The Configuration Space Method For Kinematic Design of Mechanisms

The Conﬁguration Space Method for
Kinematic Design of Mechanisms
Elisha Sacks and Leo Joskowicz

The Configuration Space Method for Kinematic Design of
Mechanisms
Mechanisms
Elisha Sacks and Leo Joskowicz
The MIT Press

Cambridge, Massachusetts
London, England
( 2010 Massachusetts Institute of Technology
All rights reserved. No part of this book may be reproduced in any form by any elec-
tronic or mechanical means (including photocopying, recording, or information
storage and retrieval) without permission in writing from the publisher.
For information about special quantity discounts, please email

[email protected]
This book was set in Stone Serif and Stone Sans on 3B2 by Asco Typesetters, Hong
Kong. Printed and bound in the United States of America.
Library of Congress Cataloging-in-Publication Data
Sacks, Elisha, 1958–

The configuration space method for kinematic design of mechanisms / Elisha Sacks
and Leo Joskowicz.
p. cm.
Includes bibliographical references and index.
ISBN 978-0-262-01389-5 (hardcover : alk. paper) 1. Machinery, Kinematics of.
2. Computer-aided design. 3. Configuration space. 4. Machine design.
I. Joskowicz, Leo, 1961– II. Title.
TJ175.S23 2010
621.8 0 11—dc22 2009030466
10 9 8 7 6 5 4 3 2 1
I dedicate this book to my wife, Jennifer, the light of my life.
Elisha
I dedicate this book to my daughters, Dana and Yael, to my parents, Esther

and Alfredo, and to all my close friends.
Leo
Contents
Preface xi
Acknowledgments xiii
1 Introduction 1
1.1 Mechanisms and Kinematic Function 3
1.2 Kinematic Design 5
1.3 Content and Organization of the Book 9
1.4 Notes 10
2 Mechanisms 11
2.1 Geometry of a Part 11
2.1.1 Planar Parts 11
2.1.2 Spatial Parts 14
2.2 Configurations and Motions of Parts 16
2.2.1 Planar Parts 16
2.2.2 Spatial Parts 20
2.3 Kinematic Pairs 21
2.3.1 Representation of Configuration Space 21
2.3.2 Lower and Higher Pairs 24
2.4 Mechanisms 25
2.5 Classification of Mechanisms 29
2.6 Notes 32
3 Contact of Features 33
3.1 Simple Planar Feature Contact 33
3.1.1 Tangency Equations 34
3.1.2 Boundary Equations 35
3.1.3 Subsumed Contacts 36
3.2 Fixed-Axis Planar Pairs 37
3.2.1 Tangency Curves 37
3.2.2 Contact Curves 43
3.3 General Planar Pairs 45
viii Contents
3.4 General Boundary Segments 50

3.5 Spatial Pairs 52
3.6 Notes 55
4 Contact of Parts in Configuration Space 57

4.1 Partition of Configuration Space 57
4.2 Fixed-Axis Pairs 60
4.3 Partition Algorithm for Fixed-Axis Pairs 62
4.3.1 Vertex and Edge Formation 64
4.3.2 Face Formation 65
4.3.3 Face Classification 66
4.5 Partition Algorithm for General Planar Pairs 69
4.6 Mechanisms 72
4.7 Partitions of Configuration Space and the Theory of Mechanisms 75
4.8 Notes 78
5 Analysis 81
5.1 Kinematic Analysis by Examination of Partition 81
5.1.1 Disk Indexing Pair 82
5.1.2 Lever Indexing Pair 83
5.1.3 Intermittent Gear Mechanism 86
5.2 Kinematic Simulation 87
5.2.1 Simulation Algorithm 88
5.2.2 Computation of Velocity for Fixed-Axis Mechanisms 90
5.2.3 Computation of Velocity for General Planar Mechanisms 93
5.3 Dynamical Simulation 95
5.4 Notes 98
6 Tolerancing 99
6.1 Specifications of Parametric Tolerance 99
6.2 Tolerance Analysis 100
6.3 Algorithms for Kinematic Pairs 103
6.4 Algorithms for Mechanisms 107
6.5 Notes 108
7 Synthesis 109
7.1 Kinematic Design Cycle 109
7.2 Parameter Synthesis 111
7.2.1 Design Change 112
7.2.2 Structure Change 113
7.2.3 Multiple Draggers 115
7.3 Parameter Optimization 117
7.4 Notes 120
Contents ix
8 Case Studies 121

8.1 Optical Filter Mechanism 121
8.2 Manual Transmission Gearshift 123
8.3 Torsional Ratcheting Actuator 125
8.4 Spatial Asynchronous Gear Pair 128
9 Conclusion 133
Research Directions 134
Appendix A Catalog of Mechanisms 137

Appendix B HIPAIR Software 175
B.1 Installation 175
B.2 Graphical User Interface 175
B.3 Mechanisms 177
B.3.1 Parts 177
B.3.2 Configuration Spaces of Fixed-Axis Pairs 178
B.3.3 Kinematic Simulation 178
B.3.4 Contact Zones of Fixed-Axis Pairs 178
B.4 Input Format 179
B.4.1 Mechanisms 179
B.4.2 Parts 179
B.4.3 Simulation and Display Data 181
Glossary 183
References 187
Index 193
Preface
This monograph presents the configuration space method for computer-

aided design of mechanisms with changing contacts of parts. Configura-
tion space is a complete and compact geometric representation of the
motions and interactions of parts in a mechanism that supports the core
design tasks of analysis, synthesis, and tolerancing. Our method is the first
general algorithmic treatment of the kinematics of higher pairs with chang-
ing contacts. It helps designers detect and correct unexpected kinematic
behaviors and design flaws, as demonstrated in automotive, micromechani-
cal, and optical case studies.
This book is intended for students, researchers, and engineers in mechani-
cal engineering, computer-aided design, computer science, and robotics.
The presentation is self-contained and is suited for a course, a seminar, or
independent study. The prerequisites are freshman mathematics and com-
puter science.
The first part describes the configuration space framework and the algo-
rithms for the kinematics of mechanisms. The second part describes the
algorithms for kinematic analysis, tolerancing, and synthesis based on con-
figuration spaces. The third part presents four case studies taken from in-
dustry in which the configuration space method supports the analysis and
design of mechanisms. A catalog of higher-pair mechanisms is given in ap-
pendix A. Appendix B describes an open-source Cþþ mechanical design
system, called HIPAIR, that implements some of the configuration space
methods described in this book, including visualization of configuration
space and kinematic simulation. HIPAIR comes with an interactive graphi-
cal user interface and many sample input files for mechanisms.
Acknowledgments
We wish to thank the many people who have contributed to this book.
Min-Ho Kyung at Ajou University, Sowon, South Korea, developed the pa-
rameter synthesis and optimization algorithms. Ku-Jin Kim at Ajou Univer-
sity developed the partition algorithm for a spatial fixed-axis configuration
space. Vijay Srinivasan from the IBM T.J. Watson Research Center, York
town Heights, participated in our early research on contact zones and
Min-Ho Kyung participated in our research on nonlinear contact zones.
Ralf Schultheiss and Uwe Hinze at Ford Motors, Cologne, Germany, intro-
duced us to automotive design, provided the manual gearshift case study,
and supported our research with a Ford grant. Jim Allen, Steve Barnes, and
many others at Sandia National Laboratories, Albuquerque, New Mexico,
introduced us to the design of MEMS mechanisms and provided the case
study of the torsional ratcheting actuator. The National Science Foundation
in Washington, D.C. and the Israel Science Foundation in Jerusalem sup-
ported this work. Ruby Shamir from Hebrew University of Jerusalem pro-
vided valuable comments on the first part of this book.
I, Elisha, wish to thank my parents for raising me to love books and in-
dependent thinking. I enjoyed and learned from discussions with many
people, especially Hal Abelson, Jim Allen, Bruce Donald, Mike Erdman, Pat
Hanrahan, Chris Hoffman, Matt Mason, Victor Milenkovic, Tomas Lozano-
Perez, Ralf Schultheiss, Herb Simon, Gerry Sussman, and Ken Yip. Leo Jos-
kowicz introduced me to mechanism design and to configuration spaces.
He hosted me for many years in Paris, New York, and Jerusalem, and has
always been a loyal friend. My wife, Jennifer, and daughters, Sarah, Talia,
and Anat, encouraged me along the way and offered marketing advice. Sad
to say, we will not be able to use ‘‘Passion in the Desert’’ as the title, nor
can a romantic oasis scene appear on the cover. I realize that equations
and diagrams are sorry substitutes. Last but not necessarily least, my cat,
xiv Acknowledgments
Fluffer, has kept me company (often asleep on the couch) for most of the
long hours that I spent writing and revising this book.
I, Leo, wish to thank my parents; my daughters, Yael and Dana, and my
many beloved friends around the world for their support and patience
in good and bad times. I thank Elisha Sacks for being a perseverant and
outstanding partner in this long journey. I also want to thank Sanjaya
Addanki, Sesh Murthy, Scott Penberthy, Jarek Rossignac, and former col-
leagues at the IBM T.J. Watson Research Center, Yorktown Heights, NY,
for their support and fruitful discussions on mechanical design and re-
search. The Institute of Computer Science, now the School of Engineering
and Computer Science at the Hebrew University of Jerusalem, Israel, pro-
vided an optimal environment for writing this book.
Mechanisms
1 Introduction
Mechanisms are assemblies of moving parts that perform useful tasks. They
are pervasive in modern life and include familiar mechanical systems, such
as door locks, transmissions, and gearboxes, and specialized ones, such as
industrial robots and microscopic mechanisms fabricated on silicon chips
(MEMS). The designing of mechanisms is an important engineering task
that motivates research in design methods and in computer-aided design
(CAD) tools.
A key aspect of this task is kinematics. This is the branch of mechanics
that studies the motion of parts independently of the forces acting on
them. It assumes that parts are rigid bodies with fixed shapes. A mecha-
nism performs a task by transforming input motions into output motions
through the contact of parts. The transformation of motions is called
the kinematic function of the mechanism. The design goal is to ensure
that the intended and actual motions match. Studying the kinematics is
an early and crucial step in designing a mechanism. It helps answer many
questions about the workings of mechanisms and is a prerequisite for
further mechanical studies involving dynamics, stress, and deformation.
The main kinematic design tasks are analysis, synthesis, and tolerancing.
Analysis derives the kinematic function of a mechanism from a specifica-
tion of its parts’ shapes and motion constraints. For example, a rotating
gear wheel is specified by its profile (how many teeth and what shape) and
the configuration (position and orientation) of its rotation axis. A pair of
gears is analyzed to determine its gear ratio. Synthesis is the inverse task
of devising a mechanism that performs a specified function. The starting
point can be a prior design or a novel design concept. Analysis and synthe-
sis disregard the imprecision of manufacturing, which causes actual parts to
vary from their intended shapes and configurations. The tolerancing design
task is to determine the kinematic effect of variation in manufacturing
2 1 Introduction
and modify the design or the manufacturing process to ensure correct

function.
The foundation for a systematic study of mechanism kinematics was laid
by the German engineer Franz Reuleaux (1829–1905). Reuleaux defined
mechanisms as collections of basic building blocks, called kinematic pairs,
and developed a classification system for mechanisms. A kinematic pair
consists of two interacting parts. There are two types of pairs: lower and
higher. Lower pairs involve a single, permanent surface contact between
two parts, such as a round pin in a matching hole. The parts have simple
relative motions, such as rotation around an axis. All other pairs are higher
pairs. Lower pairs are idealizations of higher pairs because real parts require
a small clearance to function.
The compositional model of mechanisms uses a hierarchical analysis
strategy: compute the kinematic functions of the pairs, then combine them
to obtain the kinematic function of the mechanism. The kinematic func-
tion of a pair is determined by the parts’ geometry in the neighborhood
of the points of contact. Analysis consists of formulating and solving con-
tact equations. These equations are highly nonlinear and can be solved in
closed form only in simple cases. Combining a pair’ kinematic functions
entails an analysis of the interactions among the pair’s contacts. This anal-
ysis is difficult, even for mechanisms with few parts, because there are
many potential interactions. Synthesis and tolerancing are also performed
hierarchically and pose similar challenges.
The mainstream approach to managing the complexity of kinematic de-
sign is to identify special cases that restrict the interactions of parts. Kine-
matic pairs are assumed to be lower pairs or to have fixed, closed-form
contact relations. Mechanisms are limited to assemblies of such pairs. Con-
sequently, a mechanism’s kinematic function is specified by a fixed set of
equations that can be solved with an efficient numerical algorithm. The
most common categories are linkages, cams, and gear mechanisms. Link-
ages consist of lower pairs. Cam mechanisms consist of cams in permanent
contact with followers. Gear mechanisms consist of meshed gears. These
categories exclude many important mechanisms.
Pairs with multiple contacts are more versatile than fixed-contact pairs
because they can perform multiple functions through changes in their
contacts. Higher-pair mechanisms are typically cheaper, lighter, and more
compact than lower-pair mechanisms. Examples include sewing machines,
copiers, cameras, and compact disc players.
In this book we present a general computational theory of kinematic de-
sign that covers all types of mechanisms. In this approach, the contacts of
1.1 Mechanisms and Kinematic Function 3
Figure 1.1
Kinematic design of a mechanism using configuration spaces.
parts are studied in a geometric representation called configuration space.

We have developed algorithms for analysis, synthesis, and tolerancing
within the representation of configuration space. The algorithms provide
novel computer-aided design tools that can help designers detect un-
expected behaviors, correct design flaws, and study the kinematic effects of
variations in manufacturing.
Figure 1.1 illustrates our kinematic design paradigm. Mechanical CAD
packages are used to create, modify, and visualize mechanism models. The
computation of configuration space generates a kinematic model that is
used in analysis and tolerancing, and is modified in synthesis.
The rest of this chapter is organized as follows. In section 1.1, we intro-
duce kinematic function. In section 1.2, we discuss and illustrate kinematic
design. In section 1.3, we describe the content and organization of this
book.
1.1 Mechanisms and Kinematic Function
The kinematic function of a mechanism is determined by the shapes, con-

figurations, and motions of its parts. For example, consider the indexer
mechanism in figure 1.2. It is composed of six parts: a driver, an indexer, a
pawl, a pin, a lever, and a frame. The driver is an offcenter cylinder that acts
as a cam. The indexer is a gear with 24 teeth shapes as trapezoids. The pawl
has a triangular tip shaped to follow the indexer’s teeth. The pin is a rectan-
gular block. Each of these parts rotates on a shaft mounted on a cylindrical
hole in the frame. The lever has a rounded triangular tip shaped to engage
4 1 Introduction
Figure 1.2
Indexer mechanism.
the indexer’s teeth, a cylindrical hole that fits over the cam, and a rectangu-
lar slot that fits over the pin. Its motion is constrained by the driver and the
pin. The frame holds the parts together. The indexer mechanism has six
lower pairs (indexer, pawl, driver and pin mounted on the frame, lever-
pin, and driver-lever), and two higher pairs (indexer-pawl, indexer-lever).
The rotations of the parts are as indicated by the arrows. (The following
conventions for arrows are used throughout the book. The intended part
rotation and translation directions are indicated with thick-headed arrows.
The rotation angle is always measured in the standard counterclockwise di-
rection even when the arrow points clockwise. The translation is always
measured along the standard right-handed coordinate axis even when the
arrow points left. A double thick-headed arrow indicates back-and-forth
motion.)
The kinematic function of the indexer mechanism is to advance the in-
dexer wheel by one tooth (15 ) for every turn of the driver. As the driver
turns clockwise, the lever tip traces a closed trajectory whose form is deter-
mined by the relative position of the driver and pin rotation axes, the driv-
er’s offset, and the length of the lever. This causes the indexer to rotate
counterclockwise by 15 . The pawl prevents clockwise rotation of the in-
dexer. Figure 1.3 shows four snapshots of one cycle: (a) the start configura-
tion, (b) the lever tooth engaging the wheel tooth, (c) the lever driving the
wheel, and (d) the lever tooth disengaging the wheel tooth.
In designing a mechanism, the kinematic function is usually derived first
because it sets the stage for the study of other physical phenomena. In our
example, dynamical analysis determines the driver and indexer torques
based on their mass and shape. Stress analysis determines the indexer tip
deformation based on its load and material properties. In both analyses,
the kinematic function constrains the motions and contacts of the parts.
Figure 1.3
Snapshots of the kinematic function of an indexer mechanism.
Kinematic design is an iterative process consisting of five steps (figure

1.4). Conceptual design consists of selecting a design concept that captures
the desired kinematic function. The concept determines the structure of
the mechanism: the parts, the kinematic pairs, and the intended kinematic
function. Parametric design consists of building a parametric model of the
mechanism that encodes the shapes and configurations of its parts. Param-
eter values are chosen to achieve the intended kinematic function in the
absence of manufacturing variation. The resulting design is called a nomi-
nal mechanism. Analysis, tolerancing, and synthesis are then performed as
described earlier. They are iterated until a satisfactory design is obtained.
When this is impossible, the parametric model is revised or an alternative
design concept is elaborated.
1.2 Kinematic Design
We illustrate the design cycle using the indexer mechanism. Figure 1.5
shows a parametric model for the conceptual design described here. The
parameters include the relative positions of the fixed axes, the offset of the
driver, and the radii of the indexer teeth. The parts are modeled in the xy
plane because they have a constant cross-section along the z axis. Likewise,
their motions are modeled in the xy plane. The indexer, pawl, driver, and
pin are mounted on cylindrical shafts, so each is modeled with a pin joint.
6 1 Introduction
Figure 1.4
Kinematic design cycle. Boldface indicates tasks covered in this book.
Figure 1.5
Indexer mechanism: functional parameters (in millimeters).
Figure 1.6
Kinematic function of the indexer mechanism.
Their configurations are specified with orientation parameters with respect

to fixed axes: yI and OI for the indexer, yP and OP for the pawl, yD and OD
for the driver, and yN and ON for the pin. The intended rotation directions
are indicated with arrows. The lever is not mounted on a shaft, so its con-
figuration is specified with two position parameters and an orientation
parameter. The pin fits into the lever’s rectangular slot, so the pin-lever
contact is modeled as a slider joint. The driver and lever have a permanent
contact between a circle and a matching circular hole. The pawl-indexer
and lever-indexer interactions are more complex because they involve mul-
tiple changes of contact.
Having created the parametric model, the designer faces several key ques-
tions. Will the nominal mechanism function as intended or can it fail,
owing to unexpected motions of parts? What modifications are necessary
to make the mechanism function properly despite manufacturing varia-
tion? What tolerances are required? Can they be loosened, which reduces
manufacturing cost, by modifying the nominal design? To answer these
questions, the designer analyzes the current design and synthesizes alterna-
tive designs.
The analysis derives the kinematic function of the nominal mechanism
from its parametric model. The kinematic function is quantified by plotting
the relation between the motion parameters. Figure 1.6 shows the plot for
our example. The functional relation is between the driver’s orientation an-
gle, yD , and the indexer’s orientation angle, yI . The initial values are a ¼
ð0 ; 10 Þ. Angle yD increases clockwise and yI increases counterclockwise.
The bottom curve represents the work cycle shown in figure 1.3 with
points a–d corresponding to the snapshots. The indexer is at rest on the
horizontal part of the curve and is driven on the diagonal part. At the end
of the cycle, yD equals 360 and yI has increased to 25 . The middle curve
represents the second cycle, in which yI increases from 25 to 40 , and the
top curve represents the third cycle. Since the indexer has 24 teeth, there
are 24 such segments at 15-degree intervals.
8 1 Introduction
Figure 1.7
Lever pin failures (detail): (a) tip too wide, (b) tip too long, (c) tip too short.
The kinematic function rarely has a closed form, especially for mecha-
nisms with changes in part contact. Numerical kinematic simulation is often
used instead. Kinematic simulation takes a driving motion of a part as in-
put and computes motions for the other parts according to the principle
that rigid parts cannot overlap. The snapshots in figure 1.3 were generated
from a simulation of the indexer mechanism. Rotation of the driver is the
driving motion. The lever is assigned a motion that preserves its contacts
with the driver and the pin. When the lever tip engages the indexer, the
indexer is assigned an angular velocity that prevents tip and tooth overlap.
The drawback of simulation is that it only provides information for one
input motion. Unexpected behaviors can be missed. In our example, the
simulation assumes that the only driving motion is the driver’s rotation.
An unexpected force can make the pawl rotate clockwise, disengage from
the indexer, and so fail to perform its function. Unexpected behaviors usu-
ally involve changes in the contact of parts.
Once the nominal function of the mechanism has been achieved, the
next step is to assign tolerances. The goal is to ensure that the kinematic
function is preserved for small variations in the shape and configuration
of parts. Figure 1.7 shows three examples in which failures occur that are
due to small variations of the lever tip. The nominal tip width is 1 mm
and its nominal length is 7.5 mm. Too wide a tip (1.8 mm) causes the tip
to hit the tooth top (figure 1.7a). Too long a tip (11 mm) causes blocking
(b). Too short a tip (5 mm) fails to engage the indexer tooth (c). Toler-
ance analysis derives the kinematic variation for given tolerances. Synthesis
adjusts the parameter values and tolerances to prevent these failures.
Kinematic design can be difficult and time-consuming. The designer has
to optimize the design to comply with multiple, conflicting requirements.
Many parameter values and tolerances must be adjusted. The adjustments
require extensive analysis of many design instances. The analysis involves
1.3 Content and Organization of the Book 9
many contacts of parts, with complicated contact equations. Unintended

contacts can arise from variation in parts and can lead to failure modes
that coexist with or supersede the nominal function.
A key property of mechanisms is that the interactions of parts are tightly
coupled and nonlinear. This property implies that mechanisms cannot be
decomposed into linear functional modules with narrow interfaces, where-
as other engineering disciplines, notably circuit design, rely heavily on de-
composition. The difficulty of mechanism design motivates research in
design algorithms.
1.3 Content and Organization of the Book
This book describes our configuration space method for mechanism design,
which we have developed over the past 20 years. The presentation is self-
contained and tutorial, with references to related work by ourselves and
others. It is a research monograph rather than a textbook, so the reader
should be prepared to work out some technical details.
The book is organized as follows. In chapters 2–4 we describe our config-
uration space representation of kinematics. In chapter 2, we review basic
geometry concepts, describe the configuration space representation, and
discuss kinematic pairs and classification of mechanisms. The shape of a
part in defined in terms of geometric features that form its boundary. In
chapter 3, we study contacts between two features. We formulate contact
equations for basic features, provide closed-form solutions for some cases,
and provide general numerical solutions. In chapter 4, we study the config-
uration spaces of kinematic pairs and describe efficient algorithms for com-
puting them.
In chapters 5–7 we present algorithms for kinematic analysis, toleranc-
ing, and synthesis based on configuration spaces. In chapter 5 we present
algorithms for kinematic analysis of a mechanism based on examination
of configuration space and kinematic simulation. In chapter 6 we describe
tolerance analysis based on a parametric worst-case tolerance model. In
chapter 7 we present two methods for kinematic optimization of planar
mechanisms.
In chapter 8 we illustrate the configuration space method for designing
mechanisms using four case studies taken from industry. The case studies
are an optical filter mechanism, a manual transmission gearshift, a MEMS
torsional ratcheting actuator, and an automotive asynchronous spatial re-
verse gear pair. In chapter 9 we conclude with a summary and research
directions.
10 1 Introduction
The book contains two appendices. The first is a catalog of 30 representa-

tive higher-pair mechanisms. Each mechanism is depicted by a sketch and
described with a brief narrative. Its kinematic function is described using
configuration spaces. The second appendix contains the users’ guide to an
open-source Cþþ software package, HIPAIR, which implements some of
the mechanism design methods discussed in this book using configuration
spaces.
1.4 Notes
There are numerous introductory and advanced textbooks on kinematic de-

sign of mechanisms [22, 31, 59, 78]. Other books focus on linkage [27],
cam [4, 26], and gear [53] mechanisms and their design. For an introduc-
tion and description of modern mechanical CAD, see Dimarogonas [19].
Commercial packages for mechanical computer-aided design include
Catia-CADAM, Pro/Engineer, AutoCAD, I-DEAS, and ADAMS. These pack-
ages support drafting, manipulation, and visualization, and include limited
support for kinematic analysis and tolerancing of mechanisms whose parts
interact via a fixed set of contacts.
Books on robot kinematics include those by Paul [62] and McCarthy [55].
For recent work on robotic manipulation and computational kinematics,
see Mason [54] and Angeles et al. [3]. The configuration space method for
planning robot motion is described in Latombe [47] and Choset et al. [16].
2 Mechanisms
Mechanisms consist of moving interacting parts. The geometry, motion,

and contact of parts give rise to kinematic function. This chapter intro-
duces our representations of these concepts. It describes mechanisms in
terms of their basic functional elements, which are kinematic pairs.
In section 2.1 we describe the representation of the geometry of parts. In
section 2.2, we describe the configurations (positions and orientations) and
motions of parts, and configuration spaces. In sections 2.3 and 2.4 we in-
troduce kinematic pairs and mechanisms and discuss their properties. In
section 2.5 we discuss the classification of mechanisms.
2.1 Geometry of a Part
A part in a mechanism is a rigid solid with a fixed shape. Modeling of solids

proposes two main types of shape representation: volume (also called con-
structive solid geometry) and boundary. Volume representations describe
parts in terms of set operations (union, intersection, complement) on ele-
mentary volumes (cubes, cylinders, cones, spheres). Boundary represen-
tations describe parts by their boundary elements. Boundaries are closed
surfaces that delimit a part’s interior. Volume and boundary representa-
tions are mostly interchangeable because it is possible to obtain either one
from the other. In this book we use boundary representations because they
facilitate the analysis of contacts between parts.
2.1.1 Planar Parts

Although parts are spatial, their kinematic function can often be analyzed
in the plane. A planar part is specified by a cross-section and a thickness.
For example, the lever from the indexer mechanism in figure 1.2 is a planar
part (figure 2.1). The part is obtained by extruding the cross-section perpen-
dicularly to its plane.
12 2 Mechanisms
Figure 2.1
A lever (a) and its cross-section (b).
Figure 2.2
A part with two extruded cross-sections (a) and its cross-sections (b).
A cross-section consists of closed regions called components. A compo-

nent has an outer boundary and may have inner boundaries (holes). Nei-
ther components nor boundaries can intersect, but components can nest.
The lever has one component with an outer boundary and two holes.
A general planar part consists of a stack of extruded cross-sections with a
common plane. Figure 2.2a shows a part with two extruded cross-sections
and a nested component. Figure 2.2b shows the two cross-sections. Cross-
section 1 has one rectangular component. Cross-section 2 has components
C1, C2, and C3, with C2 nested inside C1.
A boundary is a single closed curve, such as a lever’s circular hole, or
a sequence of curve segments. Closed curves include circles and ellipses.
A curve segment is the portion of a curve delimited by two endpoints.
Curve segments include line segments, circle segments, and splines. The
outer boundary of the lever consists of line segments and arc segments,
whereas the slot boundary has only line segments. The features of a planar
Figure 2.3
Planar part A coordinate system.
part are the closed curves, curve segments, and segment endpoints of its
boundary.
Closed curves and curve segments are specified with respect to a part’s
coordinate system. The coordinate system of a part A consists of a point
o A , called the origin, and of orthogonal unit vectors x A and y A , called
axes. The coordinates of point p in the coordinate system ðo A ; x A ; y A Þ are
denoted by ð pxA ; pyA Þ (figure 2.3).
Curves can be represented explicitly, implicitly, or parametrically. An ex-
plicit representation is y A ¼ f ðx A Þ. An implicit representation is f ðx A ; y A Þ ¼
0. A parametric representation is x A ðtÞ ¼ f ðtÞ and y A ðtÞ ¼ gðtÞ where t is the
curve’s parameter. For example, 2x A þ 3y A þ 5 ¼ 0 is the implicit represen-
tation of a line, and ðx A Þ 2 þ 4ðy A Þ 2 ¼ 1 is the implicit representation of an
ellipse. We use the implicit representation of curves because it is general
and convenient for deriving contact relations.
A boundary is represented by a sequence of adjacent segments. The two
endpoints of a segment are labeled tail and head, so that the part’s interior
is on the left when the segment is traversed from tail to head. Two seg-
ments are adjacent when they share an endpoint. The segment sequence
is chosen so that the head of each segment is the tail of the next segment.
This ordering implies that outer boundaries are traversed counterclockwise,
whereas inner boundaries are traversed clockwise. The outer boundary of
the lever in figure 2.4 consists of the 12 segments a1 a2 ; a2 a3 ; . . . ; a12 a1 with
ai the segment’s endpoints. The rectangular slot boundary consists of the 4
segments c1 , c2 , c3 , c4 , c1 .
In some design tasks we employ a more general, parametric representa-
tion of a part. Each boundary segment is specified by a function f ðx; y; pÞ
¼ 0 with p ¼ ð p1 ; . . . ; pn Þ a parameter vector. The value of the nominal
parameter vector defines the part’s shape. The shape can be changed by
14 2 Mechanisms
Figure 2.4
Boundary representation of a lever part. The dots represent segment endpoints and
arrows indicate segment orientation.
Figure 2.5
Parametric representation of a lever.
changing the parameter’s values. The parameters are chosen to reflect

shape changes that are relevant to the mechanism’s kinematic function.
Figure 2.5 shows a parametric model of the lever. The parameter vector is
p ¼ ðw; h; r; lÞ, where w and h are the tip’s width and height, r is the the
cam’s inner radius, and l is the slot length. Its nominal value is p ¼
ð8; 8; 12; 30Þ.
2.1.2 Spatial Parts

The boundary representation of a spatial part consists of faces, edges, and
vertices, collectively called features, that separate the part’s interior from
its exterior. Faces are surface patches, edges are line or curve segments
where faces meet, and vertices are points where edges meet.
Figure 2.6 shows the most common faces. These faces are obtained by
cutting surfaces with planes. For example, the front circle face of the cylin-
der is the intersection of the plane y A ¼ 0 with the cylinder ðx A Þ 2 þ ðz A Þ 2 ¼
1. As for planar parts, their topology is defined by the adjacency relations
among the features. Parametric spatial models of a part can be defined as
for planar parts.
Figure 2.6
Faces: (a) planar, (b) cylindrical, (c) spherical.
Figure 2.7
Spatial boundary representation: (a) a Geneva pair with part of the follower’s inner
spherical patch boundary highlighted, (b) flattened projection of part of the inner
spherical patch follower’s boundary profile. Dots represent vertices.
The spatial Geneva pair illustrates these concepts (figure 2.7). The cam
consists of a plate with a cylindrical pin and a half cylinder mounted on
it. The follower is a hollow hemisphere with four evenly spaced slots and
circular cutouts. The boundary representation consists of spherical and cy-
lindrical faces bounded by line and arc segments. The follower’s inner
boundary consists of a spherical patch with four slots and four cutouts.
The boundary of each slot consists of two parallel circle segments on the
sides and a semicircle at the bottom. The slot’s boundary and the circle seg-
ment’s cutout are connected by a circle segment. The pattern repeats four
times to form a closed boundary. Figure 2.7b shows a flattened projection
of part of this boundary. The vertices ai correspond to boundary segment
16 2 Mechanisms
endpoints: segment a1 a2 is a straight line, followed by a circle segment a2 a3

from the cutout, connected by line segment a3 a4 to the slot boundary con-
sisting of arc segments a4 a5 (which projects to a line segment), a5 a6 , and
a6 a7 . The follower’s outer boundary has the same structure, but with larger
dimensions. Both are connected by eight rectangular planar faces at the
top, as shown in figure 2.7a.
2.2 Configurations and Motions of Parts
This section describes the mathematics of the configurations and motions

of rigid parts. The configuration of a part is its position and orientation in
a world coordinate system. Its position is given by the world coordinates of
the part’s origin. The orientation is the angles between the part’s axes and
the world axes. The part’s configuration determines the set of world points
that it occupies. We begin with planar parts and then extend the discussion
to spatial parts.
2.2.1 Planar Parts

The configurations of parts are specified using two coordinate systems as
shown in figure 2.8a. The world coordinate system, ðo; x; yÞ, is fixed; for ex-
ample, it is attached to the mechanism’s frame. The part A coordinate sys-
tem, ðo A ; x A ; y A Þ, is attached to the part. The part’s position is specified by
the translation vector tA ¼ ðxA ; yA Þ. The part’s orientation is specified by the
rotation matrix
Figure 2.8
Planar part coordinate systems: (a) world and part coordinate system, (b) a configura-
tion space coordinate system.

cos yA sin yA
RA ¼
sin yA cos yA
that maps the part’s axes to the world axes. Part coordinates, p A ¼
ð pxA ; pyA Þ, are mapped to world coordinates, p ¼ ð px ; py Þ, by the Euclidean
transformation
p ¼ tA þ RA p A :
The variables xA , yA , and yA are called configuration variables. Configura-

tion variables xA and yA are called position variables and yA is an orienta-
tion variable. The domain of xA and yA is the real line, <, and the domain
of yA is the unit circle S ð180 a yA a þ180 Þ. Their Cartesian product,
< 2 S, is called configuration space, and its points, cA ¼ ðxA ; yA ; yA Þ, are
called configurations. The configuration space represents the set of all posi-
tions and orientations (figure 2.8b) of a part.
Figure 2.9 shows a planar part in configurations c1 , c2 , c3 , c4 . In configu-
ration c1 ¼ ð2; 2; 0 Þ, the part’s origin is at two horizontal and vertical units
(e.g., centimeters) from the world origin and their axes are aligned. In con-
figuration c2 ¼ ð11; 2; 0 Þ, the part’s origin is moved by nine units along the
x axis. In configuration c3 ¼ ð16; 8; 0 Þ, the part’s origin is moved by five
units further along the x axis and by six units along the y axis. In configu-
ration c4 ¼ ð16; 8; 30 Þ, the part is rotated counterclockwise by 30 . Figure
2.10 shows the four configurations in configuration space ðxA ; yA ; yA Þ.
Figure 2.9
A part in configurations c1 , c2 , c3 , c4 in the plane.
18 2 Mechanisms
Figure 2.10
Part configurations c1 , c2 , c3 , c4 in configuration space.
When a part moves, its configuration changes continuously over a time

interval. The part’s configuration is cA ðtÞ ¼ ðxA ðtÞ; yA ðtÞ; yA ðtÞÞ where xA ðtÞ,
yA ðtÞ, and yA ðtÞ are the values of xA , yA , and yA at time t. The changing con-
figuration traces a path in configuration space. When the position variables
are constant and the orientation variable changes, the motion is rotation
around a fixed axis. When the orientation variable and one position vari-
able are constant, the motion is translation parallel to the vertical or hori-
zontal axis. When the position variables are linearly related, the motion is
translation along a fixed axis defined by the linear relation. Parts with these
motions are called fixed-axis parts.
Figure 2.11 shows an example of a planar moving part. The motion starts
at configuration c1 and translates along the x axis nine units relative to c1
to reach c2 . It then translates along a diagonal to reach c3 and rotates by
30 to reach the final configuration, c4 . The part’s motion thus consists of
three stages: horizontal translation, diagonal translation, and rotation. Fig-
ure 2.12 shows the configuration variables’ time plots and figure 2.13 shows
the path in the configuration space.
In many mechanisms, parts are designed to move along fixed axes. To
simplify the kinematic analysis, we choose a part coordinate system so that
two of the three configuration variables are constant. When a part rotates,
the coordinate system’s origin is placed at the center of rotation, so that the
two position variables are constant. When a part translates, the coordinate
system’s x axis is aligned with the motion axis so that the y variable and
the orientation variable are constant. For example, suppose part A rotates
Figure 2.11
Representative motion snapshots of the planar motion of a part.
Figure 2.12
Planar motion time plots: (a) xA ðtÞ, (b) yA ðtÞ, (c) yA ðtÞ.
20 2 Mechanisms
Figure 2.13
Planar motion of a part in configuration space.
around its origin oA in figure 2.8. The part’s configuration ðxA ; yA ; yA Þ is

ð3; 2; yA Þ, which simplifies to yA .
The independent variables that specify the configuration of a part are
called the degrees of freedom of the part. The number of variables is called
the number of degrees of freedom. Fixed-axis planar parts have one degree
of freedom whereas free planar parts have three.
2.2.2 Spatial Parts

Spatial configurations of parts are analogous to planar ones (figure 2.14).
The world coordinates are ðo; x; y; zÞ and the part A coordinates are ðo A ;
x A ; y A ; z A Þ. A spatial configuration consists of six variables, versus three in
the plane. The translation vector is tA ¼ ðxA ; yA ; zA Þ. The three-by-three rota-
tion matrix, RA , is expressed in terms of three orientation variables. One
option, called Euler angles, is to express a general rotation matrix as the
product of three rotations around coordinate axes, each of which has a
one-parameter rotation matrix. In the roll-pitch-yaw convention (figure
2.14b), the inverse of RA is expressed as
2 3 2 3 2 3
cos yA sin yA 0 cos fA 0 sin fA 1 0 0
6 7 6 7 6 7
4 sin yA cos yA 0 5 4 0 1 0 5 4 0 cos cA sin cA 5
0 0 1 sin fA 0 cos fA 0 sin cA cos cA
with yA the roll angle around xA , fA the pitch angle around yA , and cA the
yaw angle around zA . The transpose of this product is RA .
The configuration space of a spatial part is six-dimensional, with position
variables xA , yA , zA and orientation variables yA , fA , cA . The position vari-
Figure 2.14
Spatial coordinates system and configuration of a part: (a) world and parts coordinate
system, (b) configuration of a part.
able domain is the real line and the orientation variable domain is ½180 ;
þ180 . As in the plane, spatial motions specify configurations of a part as
a continuous function of time. We only study spatial parts that rotate or
translate around fixed axes.
2.3 Kinematic Pairs
The basic functional element of a mechanism is the kinematic pair. A kine-

matic pair is formed by two parts in contact that move relative to each
other. We first describe the configuration space representation for planar
and spatial pairs and then discuss lower and higher pairs.
2.3.1 Representation of Configuration Space

We represent kinematic pair configurations in a configuration space. We
consider the cases of planar pairs and spatial pairs. For the planar case, let
A and B be two planar parts with configuration variables xA , yA , yA and xB ,
yB , yB as shown in figure 2.15. The configuration variables define a six-
dimensional space, called the pair configuration space, which is the Carte-
sian product of the parts’ configuration spaces. A point in this space defines
the configuration of the two parts. The parts’ configuration variables be-
come coordinates in the configuration space. For example, in figure 2.15,
part A is in configuration ðxA ; yA ; yA Þ ¼ ð3; 2; 30 Þ and part B is in configura-
tion ðxB ; yB ; yB Þ ¼ ð7; 1:75; 45 Þ, so the configuration
22 2 Mechanisms
Figure 2.15
Coordinate systems and configuration variables for a part.
ðxA ; yA ; yA ; xB ; yB ; yB Þ ¼ ð3; 2; 30 ; 7; 1:75; 45 Þ
is a contact configuration.
We consider the cases of fixed-axis pairs and general pairs. In a fixed-
axis pair, four of the six configuration variables of the pair are constant.
For example, in figure 2.15, suppose part A rotates around its origin oA
and part B translates along its xB axis. The pair configuration, ðxA ; yA ; yA ;
xB ; yB ; yB Þ is ð3; 2; yA ; xB ; 1:75; 45 Þ, so it can be simplified to ðyA ; xB Þ, which
defines a two-dimensional configuration space. There are three types of
two-dimensional configuration spaces for planar fixed-axis pairs: rotation-
rotation with configuration space coordinates yA and yB ; rotation-
translation with configuration space coordinates yA and xB , xA , and yB ; and
translation-translation with configuration space coordinates xA and xB .
In a general planar pair, each part has three degrees of freedom. Instead
of describing pair configurations in the world coordinate system, we de-
scribe the configuration of moving part A relative to fixed part B (figure
2.16). Part B is fixed at the world coordinate system’s origin with its axes
aligned with the world axes, so its configuration is ð0; 0; 0 Þ. Part A con-
figuration variables xA , yA , yA become the coordinates of A relative to B.
The pair configuration ðxA ; yA ; yA ; xB ; yB ; yB Þ is replaced by ðxA ; yA ; yA Þ, which
defines a three-dimensional configuration space.
For the spatial case, let A and B be two spatial parts with configuration
variables xA , yA , zA , yA , yA , fA , cA and xB , yB , zB , yB , yB , fB , cB . The configu-
ration variables define a 12-dimensional configuration space. In a fixed-axis
pair, each part has one degree of freedom and we choose coordinate sys-
tems in which four of the six configuration variables are constant. In a gen-
eral pair, we can employ relative coordinates.
Figure 2.16
Relative coordinates for general planar pair.
Figure 2.17
Contact types.
Contacts between parts constrain their configurations by inducing func-

tional relations among their configuration variables. For example, assume
that the table in figure 2.17 is fixed, with the world coordinate frame
chosen so that the x and y coordinates lie on the table’s surface. The sphere
A can only be placed so that the distance from its center to the table is
equal to the sphere’s radius. The sphere’s center lies on a plane parallel to
the table’s surface, so its zA position variable is constant. The variables of
the block position are similarly constrained, and two of its three orienta-
tion variables are constant. In a pair with permanent contacts of parts, the
configuration can be specified by an independent subset of the configura-
tion variables, called the degrees of freedom of the pair. For example, the
sphere-table pair has five degrees of freedom, the block-table pair has three,
and an involute gear pair has one.
The parametric representation of a kinematic pair is a direct extension of
the parametric representation of its parts. The parameter vector p defines
24 2 Mechanisms
Figure 2.18
The six kinematic lower pairs.
the parts’ shapes and motion axes, and p specifies the nominal parameter
values.
2.3.2 Lower and Higher Pairs

A contact occurs when a feature of one part touches a feature of another
part. Contacts can be permanent or changing. An example of permanent
contact is a cylindrical shaft mounted in a hub with a matching hole. An
example of changing contacts is the two transmission plates of a clutch.
Also, two parts can have multiple contacts, such as two rotating parallel
gears with several meshing teeth.
The contact region of two parts is the set of points at which two features
touch. There are three types of contact regions: surfaces, lines, and points.
Figure 2.17 illustrates the three cases: a block on a table is a surface con-
tact, a cylinder on a table is a line contact, and a sphere on a table is a point
contact.
The mechanical engineering literature classifies kinematic pairs as lower
and higher. Lower pairs involve a single, permanent surface contact be-
tween the features of two parts. Higher pairs are all other pairs, including
multiple contacts, changing contacts, line contacts, and point contacts.
Lower pairs, also called joints, are the most popular and robust. There are
six types of lower pairs: three planar and three spatial (figure 2.18). Planar
lower pairs contain parts whose surface contacts induce planar motion: rev-
olute, prismatic, and planar, also called pin, slider, and plane joints. Spatial
lower pairs contain parts whose surface contacts induce motions in space:
2.4 Mechanisms 25
cylindrical, helical, and spherical, also called cylinder, screw, and sphere
joints. Revolute and prismatic pairs have one degree of freedom: rotation
or translation along a fixed axis. Plane pairs have all three planar degrees
of freedom. Helical pairs have one degree of freedom: linearly coupled rota-
tion and translation along a fixed axis. Cylindrical pairs have independent
rotation and translation along a fixed axis. Spherical pairs have three inde-
pendent rotations around a fixed point.
Higher pairs can be further classified by their motion types: fixed-axis or
general, and planar or spatial. Planar pairs consist of two planar parts with a
common plane whose motion is parallel to that plane. Planar fixed-axis
pairs have one degree of freedom per part. General planar pairs contain
parts with planar motions and three degrees of freedom, whereas general
spatial pairs contain parts with spatial motions and up to six degrees of
freedom.
Figure 2.19 shows four examples of kinematic pairs. The examples use
thick-headed arrows to indicate intended motion directions, as explained
in section 1.1. Figure 2.19a shows a sector gear pair, which is a fixed-axis
planar pair. Each gear rotates around its center on a fixed axis. The contin-
uous rotation of the driver intermittently turns the follower half a turn in
the opposite direction. Figure 2.19b shows a general planar pair. The gear
and driver cam are fixed-axis parts, and the enclosed finger cam rotates
and translates in the plane. Figure 2.19c is a spatial fixed-axis pair in which
the worm gear and the follower gear rotate around fixed axes that are
orthogonal to each other. The spatial geometry of the worm screw and the
gear teeth makes the pair spatial rather than planar. Figure 2.19d shows a
ball-bearing ratchet mechanism. When the driver rotates counterclockwise
around the fixed axis at its center, it drives the follower clockwise by means
of four ball bearings, which form general spatial pairs with the driver and
the follower. Motion in the clockwise direction is prevented by the shape
of the cam’s inner notches.
2.4 Mechanisms
Mechanisms consist of interacting fixed and moving parts. The fixed parts,
called the frame, serve as a support for the moving parts. The moving parts
interact among themselves and with the frame via contacts. Their shapes
and interactions are designed to achieve useful kinematic functions.
As for kinematic pairs, we represent a mechanism’s configurations in
configuration space. In a planar mechanism with n moving parts, each
part contributes up to three configuration variables, so the configuration
26 2 Mechanisms
Figure 2.19
Examples of kinematic higher pairs: (a) planar fixed-axis, (b) general planar, (c) spa-
tial fixed-axis, (d) general spatial.
space is at most 3n-dimensional. In a spatial mechanism, each moving part

contributes up to six configuration variables, so the configuration space is
at most 6n-dimensional.
Kinematic functions describe the relationships among the motions of
parts. For example, the function of two parallel fixed-axis gears with mesh-
ing teeth of equal diameter is to reverse the rotation direction while keep-
ing the same rotation speed and acceleration. The kinematic functions
result from the shapes, contacts, and driving motions of the parts.
Mechanisms are most often described by their structural and functional
properties. Structural properties describe the shapes, features, relative con-
2.4 Mechanisms 27
Figure 2.20
Planar mechanisms: (a) link-gear dwell, (b) Dennis clutch.
figurations, and contacts of parts. For example, the structural description of

a gearbox indicates the number of gears, their rotation axes, number of
teeth and diameter, and the contacts between the gears. Additional struc-
tural descriptions refer to geometric properties common to all parts and
pairs. Planar mechanisms include planar parts and pairs with a common
plane, whereas spatial mechanisms include one or more spatial pairs.
Fixed-axis mechanisms consist of parts that move along fixed axes. Linkage
mechanisms consist of lower pairs only. Other structural descriptions stem
from the common name of the main part or pair of the mechanism, such
as gear, screw, piston, cam, gear-crank, and lever-ratchet.
The mechanisms in figure 2.20 illustrate these concepts. The link-gear dwell
mechanism is a planar linkage consisting of a fixed frame and five moving
parts with six pin joints (solid circles) and one slider joint. The counter-
clockwise rotation of the driver causes the follower to oscillate and the slider
to follow the closed trajectory (dotted lines). The Dennis clutch is a general
planar mechanism consisting of an arm, a ratchet, a cam, a pawl, and a frame
(not shown). The arm is attached to the frame with a pin joint. The ratchet is
mounted on a rotating shaft. The cam consists of a pin and a rim attached
to a plate, and is mounted on a rotating shaft. The pawl is attached to the
cam plate by a pin. Figure 2.7a shows a fixed-axis spatial mechanism con-
sisting of a cam and a follower mounted on a frame (not shown).
Contacts between parts are collectively referred to as the mechanism’s
contact topology. An abstract description of the contacts is a topology
graph. In this graph, each part is represented by a node and an edge be-
tween two nodes represents a contact between two parts. Solid edges rep-
resent permanent contacts, while dashed edges represent intermittent
28 2 Mechanisms
Figure 2.21
Topology graphs of the mechanisms in figure 2.20: (a) link-gear dwell and (b) Dennis
clutch.
contacts. Paths between nodes in the graph indicate interactions between

parts that are not directly in contact. A loop in the graph is a path from
a part to itself that goes through two or more parts. The structure of the
graph serves to characterize mechanisms. Topology graphs can be open
loop and closed loop, depending on whether they contain a loop. Open-
loop graphs can be trees or chains. A tree is a graph in which each node ex-
cept one, the start node, has exactly one parent and zero or more children.
A chain is a tree in which each node but one, the end node, has exactly one
child. Closed-loop graphs can have a single or multiple loops. Figure 2.21
shows the topology graphs of the mechanisms in figure 2.20.
Functional descriptions refer to the motions of parts and their relation-
ships. These motions can be simple, such as fixed-axis rotation or trans-
lation, or more complex. Motion relations describe the relationships
between a part’s degrees of freedom in terms of mathematical functions.
For example, the functional description of the gearbox indicates the gear
shifts and the ratio between the input and output rotation axes. Addi-
tional functional descriptions refer to the overall purpose of the mecha-
nism. Examples include clutch, braking, indexing, stopping, and locking
mechanisms. Often, motions of parts are interpreted as input and output
motions.
The two mechanisms in figure 2.20 illustrate these concepts. The func-
tion of the link-gear dwell mechanism is to convert a continuous rotation
of the driver to an oscillatory motion of the follower, with a dwell in-
between. The function of the Dennis clutch is to prevent the clockwise ro-
tation of the cam.
2.5 Classification of Mechanisms
It has long been recognized that the classification of mechanisms according

to their structural and functional properties poses a difficult problem. The
mechanical engineering literature often employs structural, functional, and
other descriptions, such as the inventor’s name (Dennis clutch) and the in-
vention location (Geneva cam).
The literature includes a variety of collections and encyclopedias on
mechanisms, loosely organized by structure and function. The most com-
prehensive and systematic collection is Artobolevsky’s Mechanisms in Mod-
ern Engineering Design [5]. The five-volume set contains 4,745 mechanisms
classified in a two-tier group system. The first classification is based on
structural features such as lever, cam, gear, link-gear, and slider-crank, to
name a few. The second classification is based on their function: balance,
brake, dwell, guiding, etc. The drawings have been simplified to retain
only the relevant shape details and are supplemented with short explana-
tions. Figure 2.22 shows two entries from the encyclopedia. The four-
volume collection Ingenious Mechanisms for Designers and Inventors edited
by Horton and Jones [30] includes hundreds of mechanisms loosely classi-
fied by mechanical motion, such as tripping and stop, intermittent motion,
and differential motion.
Since the focus of this book is computational, we elaborated our own
classification of mechanisms according to their geometric and kinematic
properties. We conducted a survey of about 2,500 rigid-part mechanisms
in Artobolevsky’s encyclopedia to identify the main categories and deter-
mine their relative importance. We excluded mechanisms with flexible
parts, such as belts and chains because they fall outside the scope of stan-
dard kinematics.
Tables 2.1 and 2.2 summarize the classification results for higher pairs
and for mechanisms. Of the 559 stand-alone kinematic higher pairs, 72%
are planar and 28% are spatial; 89% have one or two degrees of freedom
while 11% have three or more. Most pairs (66%) are both planar and fixed-
axis. The 1,912 mechanisms surveyed have between 3 and 25 moving
parts, and 7 on average. Linkages comprise 35% of all mechanisms, fixed-
axis 22%, and fixed-axis parts connected by linkages 9%. A total of 85%
mechanisms are planar and 18% have changing contacts (such as clutch
and ratchet mechanisms).
The mechanism compilations and our survey show that higher pairs are
pervasive. The most common are gear and cam pairs, which appear in all
types of mechanisms. Others, such as ratchet, escapement, and indexer
pairs, are common in low-torque mechanisms, such as sewing machines,
30 2 Mechanisms
Figure 2.22
Two examples of entries from Artobolevsky’s Mechanisms in Modern Engineering
Design.
Table 2.1
Survey of kinematic pairs
Degrees of
freedom Geometry
Planar
Volume Total a2 >2 planar spatial fixed-axis
Lever mechanisms 1 159 130 29 102 57 94

Lever mechanisms 2 56 43 13 46 10 37
Gear mechanisms 180 174 6 140 40 136
Cam mechanisms 164 148 16 114 50 103
Total 559 495 64 402 157 370
Percentage — 89% 11% 72% 28% 66%
Notes: The first column shows the total number of pairs. The following two columns
show the number of pairs with two or fewer degrees of freedom, or with more than
two. The fourth and fifth columns indicate the number of planar and spatial pairs.
The last column indicates the number of fixed-axis pairs.
Table 2.2
Survey of Mechanisms
Motion type
Volume Total Linkage FA Both Others VC Spatial
Lever mechanisms 1 614 341 101 18 154 123 108

Lever mechanisms 2 677 323 105 66 183 110 70
Gear mechanisms 350 1 106 36 207 72 28
Cam mechanisms 271 4 101 47 119 40 72
Total 1912 669 413 167 663 340 278
Percentage — 35% 22% 9% 35% 18% 15%
Notes: The first column shows the total number of mechanisms. The following three
columns show the number of linkage, fixed-axes (FA), and linkage and fixed-axes
(Both) mechanisms. The fifth columns shows the number of remaining mechanisms.
The sixth column shows the number of mechanisms with varying contacts (VC). The
last column shows the number of spatial mechanisms.
32 2 Mechanisms
copiers, cameras, and VCRs. Mechanisms with one degree of freedom, such
as the link-gear dwell mechanism, are the most common.
2.6 Notes
Representation of the shape of parts has been the topic of extensive re-
search in the solid modeling community for the past 30 years. An excellent
textbook on the subject is by Mantyla [53].
The motions of parts are studied in classic mechanical engineering and
robotics literature. There are textbooks by Erdmann and Sandor [22] and
Paul [62].
The configuration space representation of locations and motions of parts
is presented in Latombe [47], and more recently in Choset et al. [16].
Collections and encyclopedias describing many useful and practical
mechanisms include those by Artobolevsky [5], Horton and Jones [30], Chi-
ronis [15], and Jensen [33]. More details on the survey we conducted on
Artobolevsky’s encyclopedia can be found in Joskowicz and Sacks [40].
Tsai [79] addresses classification of mechanisms based on graph theory to
characterize the structure and function for mechanisms with permanent
contacts.
The KMODDL on-line Cornell University Reuleaux Collection of Kine-
matic Mechanisms includes many historical texts, mechanisms, and their
animations at http://kmoddl.library.cornell.edu.
Appendix B describes the representation of parts and mechanisms in
HIPAIR. It covers fixed-axis planar parts whose boundaries are line and cir-
cle segments.
3 Contact of Features
This chapter begins the study of contacts of parts with the concept of con-
tacts of features. Two features are in contact when they share one or more
points, called contact points, and do not intersect. We study the contact of
features in the configuration space of a kinematic pair. We compute the
contact configurations by formulating and solving contact equations. We
solve the equations in closed form when possible and use a numerical
solver in other cases. Using a closed-form solution is faster and more accu-
rate than using a numerical solver, but is less general.
The simplest planar features are line segments, circle segments, and their
endpoints, called vertices. These are the most common features in planar
kinematic pairs. In the first three sections, we formulate their contact equa-
tions and derive closed-form solutions. We formulate contact equations for
general planar features in section 3.4 and for spatial features in section 3.5.
Few cases have closed-form solutions, so we turn to numerical solutions.
3.1 Simple Planar Feature Contact
There are six pairs of simple planar features: circle-circle, circle-line, line-
line, vertex-vertex, circle-vertex, and line-vertex. Each pair defines a contact
type. In this section we derive contact equations for the first two types and
reduce the other four types to these two. We solve the contact equations
for fixed-axis pairs in section 3.2 and for general planar pairs in section 3.3.
Two segments are in contact when the underlying circles or lines are tan-
gent (figure 3.1a) and the point of tangency lies on both segments (figure
3.1b). The tangency configurations form the solution set of a circle-line
or circle-circle tangency equation. The contact configurations are a subset
bounded by the configurations where the point of tangency coincides with
a vertex (figure 3.1c). The boundary configurations are the solutions of
vertex-line and vertex-circle boundary equations.
34 3 Contact of Features
Figure 3.1
Tangency (a), contact (b), and boundary configuration (c) at q.
Figure 3.2
Tangency conditions: (a) circle-line, (b–c) circle-circle.
3.1.1 Tangency Equations

Figure 3.2a shows the tangency condition for a circle, a, with radius ra and
center oa , and a line, b, with point hb and unit normal nb . The distance
from the circle’s center to the line equals the circle’s radius: nb ðoa hb Þ
¼ ra where the dot denotes the inner product v w ¼ vx wx þ vy wy . We ex-
press this equation in the coordinates of parts by substituting
oa ¼ RA oaA þ tA
hb ¼ RB hbB þ tB
nb ¼ RB nbB
to obtain
RB nbB ðRA oaA þ tA tB Þ ¼ ra þ db ; ð3:1Þ

3.1 Simple Planar Feature Contact 35
Figure 3.3
Boundary conditions: (a) line, (b–c) circle.
where db ¼ nbB hbB replaces ðRB nbB Þ ðRB hbB Þ using the invariance of inner
products under rotation.
Figure 3.2b–c shows the tangency condition for a circle, a, with radius ra
and center oa , and a circle, b, with radius rb and center ob . When the circles
touch on the outside (figure 3.2b), the distance between the centers equals
the sum of the radii: kob oa k ¼ ra þ rb with k k denoting the vector norm
kvk ¼ ðvx2 þ vy2 Þ 1=2 . When the outside of a touches the inside of b (figure
3.2c), the distance between the centers equals the difference of the radii,
kob oa k ¼ rb ra , and rb > ra . We express these equations in part coordi-
nates by substituting
oa ¼ RA oaA þ tA
ob ¼ RB obB þ tB
to obtain
kRB obB þ tB RA oaA tA k ¼ rb G ra : ð3:2Þ
3.1.2 Boundary Equations

Figure 3.3a shows the boundary conditions for a vertex, v, and a line, b.
The vertex is on the line and the outward normals, na and nb , are opposite.
The equations are nb ðv hb Þ ¼ 0 and nb ¼ na . Substituting part coordi-
nates yields
ðRB nbB Þ ðtA þ RA v A tB Þ ¼ db ð3:3Þ
RB nbB ¼ RA naA ; ð3:4Þ
with db ¼ nbB hbB .

Figure 3.4
Subsumed contacts: (a) line-line, (b) identical circles, (c) vertex-vertex, (d) segment-
vertex.
Figure 3.3b shows the boundary conditions for a vertex, v, and a circle, b.
The vertex is on the line and the outward normals are opposite. The equa-
tions are ob G rb nb ¼ v, using the circle property that v ob is colinear with
nb , and nb ¼ na . Substituting the second equation into the first and intro-
ducing part coordinates yields
tA þ RA v A G rb RA naA ¼ tB þ RB obB
and the linearity of rotation yields
tA þ RA e ¼ tB þ RB obB ; ð3:5Þ
with e ¼ v A G rb naA .
3.1.3 Subsumed Contacts

We need not compute line-line contacts because they are also line-vertex
contacts. Either both vertices of the shorter segment are in contact with
the longer segment or one vertex of each segment is in contact with the
other segment (figure 3.4a). For the same reason, we need not compute
contacts for identical circles (figure 3.4b) because they are also circle-vertex
contacts. The analysis in later sections implicitly ignores this case. We need
not compute vertex-vertex contacts because they are also segment-vertex
contacts. When vertex v of a is in contact with vertex w of b, v is in contact
with b and a is in contact with w (figure 3.4c).
We rewrite a segment-vertex contact as a contact between the segment
and an artificial circle with radius zero. A contact between vertex v and seg-
ment b occurs when v lies on b and the b inward normal, m, is between the
outward normals, s and e, of the two incident segments (figure 3.4d). This
is mathematically equivalent to contact between b and a circle segment
with radius zero, center and vertices v, and vertex normals s and e.
Figure 3.5
Fixed-axis pair: (a–d) contact configurations a–d, (e) contact curve (thick) and tan-
gency curve (thin).
3.2 Fixed-Axis Planar Pairs
A fixed-axes pair has a single tangency equation whose variables are the
two degrees of freedom, while the other part coordinates are constant. The
solution set is a curve in the pair’s configuration space, called a tangency
curve. The solution set of the boundary equations is a subcurve of the tan-
gency curve, called a contact curve. Figure 3.5a–d shows contact configura-
tions a–d of a rotating circle segment, a, and a translating line segment, b.
Figure 3.5e shows the tangency and contact curves. The contact point coin-
cides with an a vertex in configuration a, is interior to both segments in
configurations b and c, and coincides with a b vertex in configuration d.
3.2.1 Tangency Curves

We solve a tangency equation by writing one variable as a closed-form ex-
pression of the other variable on a closed interval. Although the example
given here yields one interval, other equations yield several intervals. We
saw in section 2.3 that there are three types of fixed-axis pairs: rotation-
rotation, rotation-x translation, and x translation-y translation. These types
combine with the two tangency equations to generate eight cases: rotating
circle-translating line, rotating circle-rotating circle, and so on. In this sec-
tion we solve the five cases involving rotation. We solve the three transla-
tion cases at the end of section 3.3 since their main use is in general planar
pairs.
Rotating Circle-Translating Line The degrees of freedom are yA , xB and the

configuration variables xA , yA , yB , yB are constant. Since yB is constant,
n ¼ RB nbB is constant. Also tA ¼ ðxA ; yA Þ is constant. Substituting n in equa-
tion 3.1, expanding the parentheses, and rearranging yields
n tB ¼ n ðRA oaA þ tA Þ ra db :
Expanding the left side and rearranging yields
nx xB ¼ n ðRA oaA þ tA Þ ra db ny yB :
We expand tA and employ the identity

n ðRA vÞ ¼ ðv nÞ sin yA þ ðv nÞ cos yA
to obtain
nx xB ¼ ðoaA nÞ sin yA þ ðoaA nÞ cos yA ra db ny yB þ nx xA þ ny yA :
Dividing by nx gives the closed-form solution for xB in terms of yA and the

four constant configuration variables: xB ¼ k1 sin yA þ k2 cos yA þ k3 .
Figure 3.6 shows a typical pair. The constant configuration variables are
xA ¼ 0, yA ¼ 0, yB ¼ 0, and yB ¼ 0. The curve parameters are oaA ¼ ð2; 0Þ,
ra ¼ 1, hbB ¼ ð0; 0Þ, and nbB ¼ ð1; 0Þ. The curve, xB ¼ 2 cos yA 1, is a
sinusoid.
A degeneracy occurs when the line b is parallel to the xB axis, so nx ¼ 0.
The tangency curve is two vertical lines where the right side of the tan-
gency equation equals zero. Figure 3.7 shows a degenerate tangency curve.
The only change from the previous example is that b is horizontal with
hbB ¼ ð0; 2Þ and nbB ¼ ð0; 1Þ. The right-side equation is 2 cos yA ¼ 1 with
zeros at 150 and 30 .
Translating Circle-Rotating Line The degrees of freedom are xA , yB and the

configuration variables yA , yA , xB , yB are constant. Since yA is constant,
Figure 3.6
Rotating circle-translating line: (a) configuration c ¼ ð0 ; 1Þ, (b) configuration d ¼
ð90 ; 1Þ, (c) tangency curve.
Figure 3.7
Degenerate case: (a) configuration c ¼ ð150 ; 0Þ, (b) configuration d ¼ ð30 ; 0Þ, (c)
tangency curve.
o ¼ RA oaA is constant. Also tB ¼ ðxB ; yB Þ is constant. Substituting n ¼ RB nbB

and o in equation 3.1, expanding the parentheses, and rearranging yields
n tA ¼ n ðtB oÞ þ ra þ db :
Expanding the left side and rearranging yields

nx xA ¼ n ðtB oÞ þ ra þ db ny yA :
Expanding n yields the closed-form solution for xA in terms of yB .
½nbB ðtB oÞ sin yB þ ½nbB ðtB oÞ cos yB þ ra þ db ny yA

xA ¼ :
nxB cos yB nyB sin yB
yA ¼ 0, yA ¼ 0, xB ¼ 0, and yB ¼ 0. The curve parameters are oaA ¼ ð2; 0Þ,
Figure 3.8
Translating circle-rotating line: (a) configuration c ¼ ð0; 0Þ, (b) configuration d ¼
ð90 ; 0Þ, (c) tangency curve.
ra ¼ 1, hbB ¼ ð2; 0Þ, and nbB ¼ ð1; 0Þ. The solution is xA ¼ ð3 2 cos yB Þ=
cos yB . The tangency curve has vertical asymptotes at yB ¼ G90 where the
denominator equals zero. There is no tangency for any xA value because a
translates parallel to b (figure 3.8b).
Rotating Circle-Translating Circle The degrees of freedom are yA , xB and

the configuration variables xA , yA , yB , yB are constant. Hence, e ¼ RB obB tA
is constant. Define f ¼ e RA oaA . We employ the identity kvk 2 ¼ v 2 where
v 2 denotes v v. Substituting f in equation 3.2 and squaring both sides of
the equation yields
ðtB þ fÞ 2 ¼ ðrb G ra Þ 2 :
Expanding the inner product yields
tB2 þ 2f tB þ f 2 ðrb G ra Þ 2 ¼ 0:
Expanding the three inner products yields
tB2 ¼ xB2 þ yB2
A A A A
f tB ¼ xB ðex þ oay sin yA oax cos yA Þ þ yB ðey oax sin yA oay cos yA Þ
f 2 ¼ e 2 þ ðoaA Þ 2 2ðoaA eÞ sin yA 2ðoaA eÞ cos yA :
Substituting the expansions into the tangency equation yields
xB2 þ 2ðex þ oay

A A
sin yA oax A
cos yA ÞxB 2ðoax yB þ oaA eÞ sin yA
A
2ðoay yB þ oaA eÞ cos yA þ e 2 þ ðoaA Þ 2 þ 2ey yB ðrb G raÞ 2 ¼ 0:
Figure 3.9
Rotating circle-translating circle: (a) configuration c ¼ ð0 ; 1:73Þ, (b) tangency
curves.
The tangency equation is a quadratic in xB whose coefficients are linear

in sin yA and cos yA . The domain of the tangency curves is the yA intervals
where the discriminant is positive. The discriminant is quadratic in sin yA
and cos yA ; hence it has at most four real zeroes that define at most two do-
main intervals, each of which yields two tangency curves.
xA ¼ 0, yA ¼ 0, yB ¼ 0, and yB ¼ 0. The curve parameters are oaA ¼ ð0; 2Þ,
ra ¼ 1, obB ¼ ð0; 1Þ, and rb ¼ 1. The tangency equation is xB2 þ ð4 sin yA ÞxB
4 cos yA þ 1 ¼ 0. The discriminant, 16 sin 2 yA þ 16 cos yA 4, is zero at
yA ¼ G120 , so the domain of the two tangency curves is ½120 ; 120 .
Rotating Circle-Rotating Line The degrees of freedom are yA , yB , with xA ,

yA , xB , yB constant. Substituting e ¼ RA oaA þ tA tB and k3 ¼ ra þ db in equa-
tion 3.1 and expanding RB yields
ðnbB eÞ sin yB þ ðnbB eÞ cos yB ¼ k3 :
Expanding e and RA yields
k1 sin yB þ k2 cos yB ¼ k3 ; ð3:6Þ
with
k1 ¼ ðnbB oaA Þ sin yA þ ðnbB oaA Þ cos yA þ nbB ðtA tB Þ
k2 ¼ ðnbB oaA Þ cos yA ðnbB oaA Þ sin yA þ nbB ðtA tB Þ:
Let k4 ¼ ðk12 þ k22 Þ 1=2 and define a by sin a ¼ k2 =k4 and cos a ¼ k1 =k4 . Divid-
ing the tangency equation by k4 yields
Figure 3.10
Rotating circle-rotating line: (a) configuration c ¼ ð0 ; 75:52 Þ, (b) tangency curves.
cos a sin yB þ sin a cos yB ¼ k3 =k4
or sinðyB þ aÞ ¼ k3 =k4 . The domain of the tangency curve is the yA intervals

where jk3 =k4 j a 1 or k32 a k12 þ k22 . The two solutions are arcsinðk3 =k4 Þ a
and 180 arcsinðk3 =k4 Þ a.
xA ¼ 0, yA ¼ 0, xB ¼ 2 and yB ¼ 0. The curve parameters are oaA ¼ ð2; 0Þ,
ra ¼ 1, hbB ¼ ð0; 0Þ, and nbB ¼ ð1; 0Þ. The tangency equation is
2 sin yA sin yB þ 2ðcos yA þ 1Þ cos yB ¼ 1:
The domain inequality, 8ð1 þ cos yA Þ b 1, holds on the interval

½151 ; 151 .
Rotating Circle-Rotating Circle The degrees of freedom are yA , yB , with xA ,

yA , xB , yB constant. Substituting t ¼ tB tA , p ¼ RA oaA , and q ¼ RB obB in
equation 3.2 yields
ðh þ t pÞ 2 ¼ ðrb G ra Þ 2 :
Expansion yields
ðobB Þ 2 þ t 2 þ ðoaA Þ 2 þ 2q t 2p t 2q p ¼ ðrb G ra Þ 2
and rearrangement yields

q ðp tÞ þ p t ¼ k4 ;
with
k4 ¼ 0:5½ðobB Þ 2 þ t 2 þ ðoaA Þ 2 ðrb G ra Þ 2 :

Figure 3.11
Rotating circle-rotating circle: (a) configuration c ¼ ð97:4 ; 0 Þ, (b) tangency curves.
Expanding q yields
½obB ðp tÞ sin yB þ ½obB ðp tÞ cos yB þ p t ¼ k4
then expanding p yields equation 3.6 with
k1 ¼ ðoaA obB Þ sin yA ðoaA obB Þ cos yA obB t
k2 ¼ ðoaA obB Þ sin yA þ ðoaA obB Þ cos yA obB t
k3 ¼ k4 ðoaA tÞ sin yA ðoaA tÞ cos yA :
xA ¼ 0, yA ¼ 0, xB ¼ 2, and yB ¼ 0. The curve parameters are oaA ¼ ð2; 0Þ,
ra ¼ 1, obB ¼ ð0; 1Þ, and rb ¼ 1. The tangency equation is
2ð1 þ cos yA Þ sin yB þ 2 sin yA cos yB ¼ 2:5 þ 4 cos yA :
The domain inequality, 16 cos 2 yA þ 12 cos yA b 1:75, holds on the inter-

vals ½151 ; 83 and ½83 ; 151 .
A degeneracy occurs when obB ¼ ð0; 0Þ, since k1 ¼ 0, k2 ¼ 0 for all yA
values. The tangency curve is the solution of k3 ¼ 0, which is two ver-
tical lines. When oaA ¼ ð0; 0Þ as well, there are no tangency curves.
3.2.2 Contact Curves

We compute contact curves by solving the boundary equations and identi-
fying the solutions that are boundary configurations. The solutions are the
configurations where a vertex of one segment lies on a circle or line of the
other segment. A solution is a boundary configuration when the vertex lies
Figure 3.12
Contact curve (thick), tangency curve (thin), and boundary equation solutions (dots
plus feature snapshots).
on the segment. For a line segment, the vertex coordinates are between
those of the segment’s vertices. For a circle segment, the ray from the cir-
cle’s center to the vertex is between the rays from the circle’s center to
its tail and head. We split the tangency curve at the boundary configura-
tions. The contact curves are the subcurves whose midpoints are contact
configurations.
Figure 3.12 illustrates this for on a rotating circle segment, a, and a trans-
lating circle segment, b. There are six boundary equation solutions, labeled
with dots, and two boundary configurations: s where an a vertex is on b
and e where a b vertex is on a. The se subcurve is the contact curve.
Solving Boundary Equations There is one set of boundary equations for a

vertex-line contact and another for a vertex-circle contact. The three types
of fixed-axis pairs yield three cases for each equation: rotating vertex-
translating segment, translating vertex-rotating segment, and rotating
vertex-rotating segment.
The vertex-line equations are 3.3 and 3.4. When A translates and B
rotates, na is constant. Equation 3.4 implies that na RB nbB ¼ 0, which has
the form k1 sin yB þ k2 cos yB ¼ 0 with k1 and k2 constant; hence it is
solved like equation 3.6. Once yB is known, equation 3.3 is linear in the
second unknown, xA . When A rotates and B translates, nb is constant, equa-
tion 3.4 yields yA , and equation 3.3 yields xB . When both parts rotate, sub-
stituting the second equation into the first and using the invariance of
inner products under rotation yields
ðRA naA Þ ðtA tB Þ ¼ db naA v A :
This equation has the form k1 sin yA þ k2 cos yA ¼ k3 with k1 , k2 , k3 con-

stant, which is solved as earlier. Given yA , we compute na and obtain yB
from the second equation as before.
The vertex-circle equation is equation 3.5. When A and B rotate, the left
side of the equation defines a circle parameterized by yA with center tA and
radius kek, while the right side defines a circle parameterized by yB . Inter-
secting these circles (a standard computation) yields the solutions to the
boundary equations. When A rotates and B translates, the equations are
xA þ ex cos yA ey sin yA ¼ xB þ ox
yA þ ex sin yA þ ey cos yA ¼ yB þ oy ;
with o ¼ RB obB constant. The second equation is solved for yA , then the first
is solved for xB . Likewise when A translates and B rotates.
Figure 3.12 shows typical rotating arc-translating arc contact curves. The
constant configuration variables are xA ¼ 0, yA ¼ 0, yB ¼ 0, and yB ¼ 0.
The curve parameters are oaA ¼ ð0; 2Þ, ra ¼ 1, taA ¼ ð1; 2Þ, haA ¼ ð1; 2Þ, obB ¼
ð0; 1Þ, rb ¼ 1, tbB ¼ ð0:71; 0:29Þ, and hbB ¼ ð0; 2Þ. The equations for v ¼ ha are
2 cos yA 2 sin yA ¼ xB
2 sin yA þ 2 cos yA ¼ 1:
The first solution is the boundary configuration s ¼ ð24:3 ; 2:64Þ and the
second is c ¼ ð114 ; 2:64Þ where ha is only on the b circle.
3.3 General Planar Pairs
We analyze general planar pairs in the ðxA ; yA ; yA Þ configuration space of

part A relative to part B (figure 2.16). The segment contact equations define
contact surfaces in the configuration space. The cross-section at yA ¼ y0 of
the a=b contact surface consists of the contact configurations where a trans-
lates at orientation y0 . It is a contact curve in the ðxA ; yA Þ configuration
space of a rotated version of a that translates relative to b. The curve is a cir-
cle segment when both a and b are circle segments and is a line segment
otherwise. We exploit this special structure to obtain simple closed-form
expressions for contact surfaces.
We derive closed-form expressions for the three types of contact: circle
segment-line segment, line segment-circle segment, and circle segment-
circle segment. A contact surface is represented as a contact curve
Figure 3.13
General segment contacts: (a) moving circle-fixed line, (b) moving line-fixed circle,
(c) moving circle-fixed circle.
parameterized by yA and defined over an interval of yA values. The deriva-

tion of the parameterized contact curve follows the strategy of the previous
section: compute the tangency curve, compute the boundary points, and
identify the portion of the tangency curve that is the contact curve. The
main difference is that the rotated a segment contains yA as a parameter.
Moving Circle Segment-Fixed Line Segment Figure 3.13a shows the con-
tact conditions for a circle segment, a, with radius ra and center oa , and a
line segment, b, with point hb and unit normal nb . Tangency occurs when
the distance from oa to the line equals ra . The tangency equation is
nb ðRA oaA þ tA Þ ¼ ra þ db ;
using equation 3.1 with tB ¼ ð0; 0Þ and yB ¼ 0. The tangency curve,
nbx xA þ nby yA þ nb RA oaA ra db ¼ 0; ð3:7Þ
is a line parallel to b whose distance from the origin varies with yA .

Contact occurs when the point of tangency lies on a: the b inward nor-
mal angle, n ¼ arctanðnby ; nbx Þ, is in the interval of a outward normal
angles, ½aa ; b a ¼ ½aaA þ yA ; b aA þ yA , so yA A ½n b aA ; n aaA . The boundary
configurations are where the point of tangency coincides with the vertices
of b. The equations are oa ra nb ¼ gb and oa ra nb ¼ hb , as in section
3.2.2. The boundary configurations are s ¼ gb þ q and e ¼ hb þ q with q ¼
ra nb RA oaA . Point q lies on a circle of radius koaA k centered at ra nb . As y
varies, s and e rotate on translated copies of this circle. The contact curve,
se, is b translated by q.
Figure 3.14
Moving circle segment-fixed line segment: (a) contact configurations, (b) contact
curves, (c) contact surface.
Figure 3.14a shows a typical case for boundary configurations. Segment

a has center oaA ¼ ð1; 0Þ, radius ra ¼ 0:5, start angle aaA ¼ 180 , and end
angle b aA ¼ 0 . Segment b has vertices gb ¼ ð0; 1Þ and hb ¼ ð4; 1Þ, outward
normal nb ¼ ð0; 1Þ, and inward normal angle n ¼ 90 . The tangency curve
is yA þ sin yA 1:5 ¼ 0. Contact occurs for yA A ½90 ; 90 with boundary
configurations s ¼ ð0; 1Þ þ RA ð1; 0:5Þ and e ¼ ð0; 1Þ þ RA ð1; 0:5Þ. The
displayed configurations are s ¼ ð1; 1:5; 0 Þ, e ¼ ð3; 1:5; 0 Þ, c ¼ ð1:5; 2:5;
90 Þ, and d ¼ ð2:5; 0:5; 90 Þ. Figure 3.14b shows the parametric contact
curve at selected yA values and figure 3.14c shows the contact surface.
Moving Line Segment-Fixed Circle Segment Figure 3.13b shows the con-
tact conditions for a line segment, a, with point ha and unit normal na , and
a circle segment, b, with radius rb and center ob . Tangency occurs when the
distance from ob to the line equals rb . The tangency equation is
na ðob ha Þ ¼ rb ;
which yields
A A A A
ðnax cos ya nay sin ya ÞðxA obx Þ þ ðnax sin ya þ nay cos ya ÞðyA oby Þ
þ rb da ¼ 0;
with da ¼ naA haA . The tangency curve is a line that forms an angle of yA
with a and whose distance from the origin is jrb da j. Its inward normal an-
A A
gle is n þ yA with n ¼ arctanðnay ; nax Þ the inward normal angle of a.
Contact occurs when the point of tangency lies on b: n þ yA is in the
interval of b outward normal angles, ½ab ; b b , so yA A ½ab n; b b n. The
boundary configurations are where the point of tangency coincides with
Figure 3.15
Moving line segment-fixed circle segment: (a) contact configurations, (b) contact
the vertices of a. The equations are ob rb na ¼ ga and ob rb na ¼ ha . The

boundary configurations,
s ¼ ob þ RA ðgaA rb naA Þ
e ¼ ob þ RA ðhaA rb naA Þ;
rotate on circles centered at ob . The contact curve, se, is a rotated by yA

around rb naA then translated by ob rb naA .
Figure 3.15a shows a typical case. Segment a has vertices gaA ¼ ð1; 0Þ
and haA ¼ ð1; 0Þ, outward normal naA ¼ ð0; 1Þ, inward normal angle n ¼
90 , and da ¼ 0. Segment b has a center ob ¼ ð0; 1Þ, radius rb ¼ 0:5, start
angle ab ¼ 180 , and end angle b b ¼ 0 . The tangency curve is xA sin yA þ
ð1 yA Þ cos yA ¼ 0:5. Contact occurs for yA A ½90 ; 90 with boundary
configurations s ¼ ð0; 1:5Þ ðcos yA ; sin yA Þ and e ¼ ð4; 1:5Þ ðcos yA ;
sin yA Þ. The displayed configurations are s ¼ ð1; 1:5; 0 Þ, e ¼ ð3; 1:5; 0 Þ,
c ¼ ð1:5; 2:5; 90 Þ, and d ¼ ð2:5; 0:5; 90 Þ. Figure 3.15b–c shows the para-
metric contact curve and the contact surface.
Moving Circle Segment-Fixed Circle Segment Figure 3.13c shows the con-
tact conditions for a circle segment, a, with radius ra and center oa , and a
circle segment, b, with radius rb and center ob . The distance between the
centers equals the sum or difference of the radii when the circles touch on
the outside or on the inside. The tangency equation is
ktA þ RA oaA ob k ¼ r with r ¼ jrb G ra j; ð3:8Þ
using equation 3.1 with tB ¼ ð0; 0Þ and yB ¼ 0. The tangency curve is a cir-
cle of radius r centered at o ¼ ob RA oaA . As y increases, it rotates on a circle
Figure 3.16
Moving circle segment-fixed circle segment: (a) contact configurations, (b) contact
of radius koaA k centered at ob . To derive the contact curve, we parameterize

the tangency curve as qðgÞ ¼ o þ rðcos g; sin gÞ.
We derive the contact curve for two circle segments whose combined an-
gular extent, b aA aaA þ b b ab , is at most 360 . Otherwise, we split the
segments in half and analyze the four subsegment contacts. In contact con-
figuration tA ¼ qðgÞ,
oa ob ¼ qðgÞ þ RA oaA ob
¼ ob RA oaA þ rðcos g; sin gÞ þ RA oaA ob
¼ rðcos g; sin gÞ:

When the circles touch on the outside, the outward normal angles of a and
b at the point of tangency are g þ 180 and g. Contact occurs when each an-
gle is in the angular extent of its segment: g þ 180 A ½aaA þ yA ; baA þ yA and
g A ½ab ; b b . The contact curve appears as a point at yA ¼ ab b aA þ 180
where the end angle of the a interval equals the start angle of the b interval.
It grows in angular extent, then shrinks to a point and vanishes at
yA ¼ b b aaA þ 180 where the start angle of the a interval equals the end
angle of the b interval. When one circle touches the other on the inside,
the outward normal angles of a and b at the point of tangency are g and
g þ 180 , but the patch equations are the same.
Figure 3.16a shows a typical case. Segment a has a center oaA ¼ ð1; 0Þ, ra-
dius ra ¼ 0:5, start angle aaA ¼ 135 , and end angle b aA ¼ 45 . Segment
b has a center ob ¼ ð1; 1Þ, radius rb ¼ 1, start angle ab ¼ 0 , and end angle
b b ¼ 180 . The tangency curve is a circle of radius 1.5 centered at o ¼
ð1 cos yA ; 1 sin yA Þ. Contact occurs for yA A ½135 ; 135 . The displayed
configurations are s ¼ ð1; 0:5; 0 Þ, e ¼ ð1; 0:5; 0 Þ, c ¼ ð0:5; 1:1; 90 Þ,

and d ¼ ð0:5; 1:1; 90 Þ. Figure 3.16b–c shows the parametric contact curve
and the contact surface.
Fixed-Axis Translation We derive the contact curves of an x translation-y

translation pair from the yA ¼ 0 cross-section of a general pair. The config-
uration ðxA ; yA ; 0Þ of the general pair maps to the configuration ðxA ; yB Þ with
yB ¼ yA of the fixed-axis pair. The fixed-axis contact curve is thus ob-
tained by reflecting the yA ¼ 0 cross-section around the xA axis.
3.4 General Boundary Segments
This section studies contacts between general boundary segments. Figure

3.17 shows two examples. The example in figure 3.17b shows that general
segments can have simultaneous contacts, whereas simple features cannot.
We derive contact equations for general segments in the same way as for
simple features. These equations define the contact curves and surfaces of
the general segments. Unlike simple features, the equations lack closed-
form solutions, so numerical solutions are required. We discuss numerical
solution strategies.
The contact conditions are that the two segments share a boundary point
and have opposite outward normals there. Let a be a segment on A with
equation fA ðx A ; y A Þ ¼ 0 and vertices ga and ha . Let b be a segment on B
with equation fB ðx B ; y B Þ ¼ 0 and vertices gb and hb . The tangency equa-
tions are
fA ð pxA ; pyA Þ ¼ 0
fB ðqxB ; qyB Þ ¼ 0
Figure 3.17
General segment contact: (a) quadratics, (b) splines.
3.4 General Boundary Segments 51
t A þ R A p A ¼ tB þ R B q B
ðRA ‘fA Þ ðRB ‘fB Þ ¼ 0;
with ‘f ¼ ðqf =qx; qf =qyÞ. The first two equations state that p and q are on a
and b, the third equation equates these points, and the fourth equation
makes the outward normals colinear. Contact occurs when p and q lie on
a and b:
pxA A ½gax
A A
; hax ; pyA A ½gay
A A
; hay ; qxB A ½gbx
B B
; hbx ; qyB A ½gby
B B
; hby :
The inequalities assume that a and b are monotonic with respect to x and y.
If not, they are split into monotonic subsegments.
As an example, we derive the equations for the two quadratics in figure
3.17a. Segment a has equation x A þ ðy A Þ 2 ¼ 0, vertices ga and ha , and ver-
tex coordinates gaA ¼ ð1; 1Þ and haA ¼ ð1; 1Þ. Segment b has equation
x B ðy B Þ 2 ¼ 0, vertices gb and hb , and vertex coordinates gbB ¼ ð1; 1Þ and
hbB ¼ ð1; 1Þ. The tangency equations are
pxA þ ð pyA Þ 2 ¼ 0
qxB ðqyB Þ 2 ¼ 0
RA ½1; 2pyA t RB ½1; 2qyB t ¼ 0:
Each segment is split at ð0; 0Þ into two monotonic subsegments. The con-
tact inequalities for the two upper subsegments are
pxA A ½1; 0; pyA A ½0; 1; qxB A ½0; 1; qyB A ½0; 1
and likewise for the other three pairs of subsegments.

For a fixed-axis pair, we have five scalar equations in six unknowns: two
degrees of freedom and the four components of the two contact points. The
solution is a curve in a five-dimensional space whose projection into the
configuration space is the contact curve. In our example, suppose a rotates
and b translates, so the degrees of freedom are yA , xB , and the constant con-
figuration variables are xA ; yA ; yB ; yB ¼ 0. The equations are
pxA þ ð pyA Þ 2 ¼ 0
qxB ðqyB Þ 2 ¼ 0
pxA cos yA pyA sin y ¼ xB þ qxB
pxA sin yA þ pyA cos y ¼ qyB
ð4pyA qyB 1Þ sin yA 2ð pyB þ qyB Þ cos yA ¼ 0
and the unknowns are yA , xB , pxA , pyA , qxB , qyB . The solution shown in figure
3.17a is where all six unknowns equal zero. For a general planar pair, we
have three degrees of freedom, six equations in seven unknowns, and the
solution is a surface in a six-dimensional space whose projection is the con-
tact surface.
We need to approximate the contact curve or surface to a specified accu-
racy. The first step is to solve the tangency equations. If we pick values for
all but one degree of freedom, we can solve for the last one by Newtonian
iteration. The iteration requires a starting point, which can be hard to de-
rive, and can fail near multiple roots. The approximate contact curve or
surface consists of the numerical solution points that satisfy the contact
inequalities.
3.5 Spatial Pairs
Spatial contact is more general than planar contact. There are three types of
boundary features: faces, edges, and vertices. Each combination of two fea-
ture types yields a contact type. It suffices to analyze face-face, face-edge,
and edge-edge contacts. The proof that the other cases are covered by this
analysis is similar to that for the planar case (section 3.13). The analysis fol-
lows the planar pattern: derive tangency equations, solve for the tangency
configurations, and use boundary configurations to delimit the contact set.
We derive the equations for face-face, face-edge, and edge-edge contacts.
The equations are much harder to solve than in the plane because the fea-
ture dimension grows from two to three, the number of degrees of freedom
for a part grows from three to six, and the boundaries change from points
to curves. For fixed-axis pairs, we have developed closed-form solutions for
planar, spherical, and cylindrical faces with linear and circular edges.
Face-Face The tangency equations are the same as for general planar fea-
tures, except that the points, vectors, and rotations are three-dimensional.
The additional condition for face contact is that the contact points lie on
the faces. The structure of the contact inequalities reflects that of the face
boundaries.
3.5 Spatial Pairs 53
Figure 3.18
Spatial contacts: (a) cylindrical face-spherical face contact, (b) cylindrical face-linear
edge contact, (c) circular edge-linear edge contact.
Figure 3.18a shows a contact between a cylindrical face on A with equa-

tion xA2 þ yA2 ¼ 1 and a spherical face on B with equation xB2 þ yB2 þ zB2 ¼ 4.
The configurations of A and B are ð0; 0; 0; 0; 0; 0Þ and ð3; 0; 0; 0; 0; 0Þ. The
tangency equations are
ð pxA Þ 2 þ ð pyA Þ 2 ¼ 1
ðqxB Þ 2 þ ðqyB Þ 2 þ ðqzB Þ 2 ¼ 4
RA ½2pxA ; 2pyA ; 0 t RB ½2qxB ; 2qyB ; 2qzB t ¼ 0:
The cylindrical face is bounded by the half-spaces zA b 5 and zA a 5,

which yield the contact inequalities 5 a pzA a 5. The spherical face is
bounded by the half-space xB b 0, which yields the contact inequality
qxB b 0.
Face-Edge Let a be a face on A with equation fA ðx A ; y A ; z A Þ ¼ 0 and let b

be an edge on B with equations fB ðx B ; y B ; z B Þ ¼ 0 and gB ðx B ; y B ; z B Þ ¼ 0.
Contact occurs when there exist points p on a and q on b that coincide
with an a normal that is perpendicular to the b tangent. The tangency
equations are
fA ð pxA ; pyA ; pzA Þ ¼ 0
fB ðqxB ; qyB ; qzB Þ ¼ 0
gB ðqxB ; qyB ; qzB Þ ¼ 0

tA þ RA p A ¼ tB þ RB q B
RA ‘fA RB ð‘fB ‘gB Þ ¼ 0:
The structure of the inequalities in face contact is as before. The edge

inequalities state that q lies between the two vertices of the edge.
Figure 3.18b shows a contact between the above cylindrical face and a
linear edge on B (the thick line) with equations y B ¼ 0 and z B ¼ 0, vertices
ð2; 0; 0Þ and ð2; 0; 0Þ, and configuration ð0; 1; 0; 0; 0; 0Þ. The contact equa-
tions are
ð pxA Þ 2 þ ð pyA Þ 2 ¼ 1
qyB ¼ 0
qzB ¼ 0
RA ½0; 1; 0 t RB ½1; 0; 0 t ;
since the edge’s tangent direction is ð0; 1; 0Þ ð0; 0; 1Þ ¼ ð1; 0; 0Þ. The edge’s
inequalities are 2 a qxB a 2.
Edge-Edge Let a be an edge on A with equations fA ðx A ; y A ; z A Þ ¼ 0 and

gA ðx A ; y A ; z A Þ ¼ 0, and let b be an edge on B with equations fB ðx B ; y B ; z B Þ ¼
0 and gB ðx B ; y B ; z B Þ ¼ 0. Contact occurs when there exist points p on a and
q on b that coincide. The tangency equations are
fA ð pxA ; pyA ; pzA Þ ¼ 0
gA ð pxA ; pyA ; pzA Þ ¼ 0
fB ðqxB ; qyB ; qzB Þ ¼ 0
gB ðqxB ; qyB ; qzB Þ ¼ 0
and the edge inequalities are as before. Figure 3.18c shows a contact be-
tween the above linear edge and a circular edge on A with equations
ðx A Þ 2 þ ðy A Þ 2 ¼ 1 and z A ¼ 0, vertices ð0; 1; 0Þ and ð0; 1; 0Þ, and configura-
tion ð2; 0; 0; 0; 0; 0Þ.
3.6 Notes 55
Curved Features Curved features are not amenable to closed-form solu-

tions. Numerical solutions appear feasible for fixed-axis pairs because the
number of unknowns is moderate. General pairs have many more degrees
of freedom and have not been studied.
3.6 Notes
Contacts between features have been extensively studied in the mechanical

engineering and robotics literature. Mechanical engineering research ad-
dresses the kinematic analysis of pairs of features in permanent contact. It
formulates the general contact constraints, but does not provide a general
solution method. Closed-form solutions are available only for gear invo-
lutes, which induce linear constraints, and for a few specialized cam pro-
files. They do not cover imperfect involutes or nonstandard gear and cam
profiles. Numerical methods are broadly applicable, but are not robust.
They can fail to converge, can converge to spurious solutions, and cannot
handle singular contact constraints, such as those induced by a cylindrical
shaft in a cylindrical hole. Angeles and Lopez-Cajun [4] and Gonzales-
Palacios and Angeles [26] describe numerical solutions for the analysis and
synthesis of planar and spatial cams that drive standard followers. Others
describe numerical solutions for pairs of convex, planar curves [14, 65]. Lit-
vin [51] describes the feature contacts of gear teeth profiles.
The mechanical engineering literature also addresses infinitesimal kine-
matics, in which instantaneous displacements and contacts of features are
studied [31]. Expressions for instantaneous axes and centers of motion, to-
gether with contact invariants, are derived for planar and spatial contacts.
The drawback of these methods is that they are local to a linear neighbor-
hood of a known contact configuration.
Robotics studies the contact of parts in configuration space for planning
robot motions [47, 52]. The research provides contact equations for poly-
gonal [11, 52] and polyhedral [20] features. The research does not provide
practical algorithms for curved features, which are the norm in higher
pairs, because they are relatively unimportant for planning motion. Colli-
sion detection [50] provides local contact information.
We have studied feature contacts for mechanisms within a configuration
space for nearly 20 years [36]. We have studied feature contacts for planar
fixed-axis pairs [71], general planar pairs [68], and spatial fixed-axis pairs
[42]. Other work includes that by Faltings [23, 24].
Appendix B describes the feature contacts covered by HIPAIR. It covers
fixed-axis contacts of simple planar features: line and circle segments.
4 Contact of Parts in Configuration Space
In chapter 3 we began to study contact of parts by characterizing contact of

features in configuration space. In this chapter we complete the study of
contact of parts. A part contact is a feature contact where the parts’ interi-
ors are disjoint. Whereas a feature contact is local to a feature pair, a part
contact involves every pair since no features can intersect. We must ana-
lyze an entire set of feature contacts to characterize the contacts of parts.
In section 4.1, we model contacts of parts by using configuration space
partitions. We specialize partitions to fixed-axis pairs in section 4.2 and
present a partition algorithm in section 4.3. We do the same for general
planar pairs in sections 4.4 and 4.5. In section 4.6 we generalize from pairs
to mechanisms. In section 4.7 we relate our approach to the classic theory
of mechanisms.
4.1 Partition of Configuration Space
We characterize a kinematic pair by partitioning its configuration space

into free, contact, and blocked spaces. The parts are disjoint in free space,
in contact in contact space, and overlapping in blocked space. Figure 4.1
illustrates these concepts. Circle A has a center oa ¼ ð0; 0Þ, radius ra ¼ 1,
and translates with yA ¼ 0 . Circle B has a center ob ¼ ð0; 0Þ, radius rb ¼ 2,
and is fixed. The configuration space is ðxA ; yA Þ. The contact space is the cir-
cular contact curve with a center oa þ ob ¼ ð0; 0Þ and radius ra þ rb ¼ 3; free
space is its exterior and blocked space is its interior.
Free space consists of regions. Figure 4.2 shows a pair with two regions.
Circle A is as before and circle B is replaced by a box with a circular cavity.
Free space consists of an outer region where A is outside B and a circular re-
gion where A is inside the B cavity. Part A can move freely in its initial
region but cannot reach the other region. It is placed in the initial region
by moving perpendicularly to the motion plane.
58 4 Contact of Parts in Configuration Space
Figure 4.1
Circle-circle pair: (a) free configuration a ¼ ð3; 3Þ, (b) contact configuration b ¼
ð2:6; 1:5Þ, (c) blocked configuration c ¼ ð1:5; 1:5Þ, (d) configuration space partition.
Figure 4.2
Circle-box pair: (a) double contact configuration a, (b) configuration space partition.
Contact space contains subspaces where contact occurs between a pair of

features. Multiple contacts occur at configurations where the subspaces in-
tersect. In figure 4.2b, the subspaces are curves where the sole A feature
touches the b features. The curves for line segments a–d and vertices v–w
are labeled. Each curve is part of a feature contact curve, the rest of which
is in blocked space. A double contact occurs at configuration a where the a
and b curves intersect.
Configuration space partitions have a special topological structure. When
two parts are free at a configuration, there is a positive minimum distance
between pairs of boundary points. If the configuration changes slightly, the
boundaries move less than the minimum distance and the parts stay free.
Since every free configuration has a neighborhood of free configurations,
free space is an open set. Likewise, blocked space is open because blocked
parts overlap by a minimum amount and slight changes in configuration
4.1 Partition of Configuration Space 59
Figure 4.3
Translation pair: (a) configuration a ¼ ð0; 0Þ, (b) b ¼ ð2:65; 1:35Þ, (c) c ¼ ð1:29; 0Þ, (d)
d ¼ ð0:65; 0:65Þ, (e) e ¼ ð1; 1:5Þ, (f) f ¼ ð2; 1:5Þ, (g) g ¼ ð3; 1:5Þ, (h) h ¼ ð3:5; 0Þ,
(i) configuration space partition.
move the boundaries less than this amount. Contact space is a closed set
because it is the complement of the union of these two open sets. At a con-
tact configuration, outward motion along the contact normal frees the
parts and inward motion blocks them.
The configuration space partition of a pair provides a global view of its
kinematic function that reveals the pairs of features that come in contact,
the configurations where the point of contact shifts between feature pairs,
and the working modes. We illustrate this view with a simple planar pair
(figure 4.3). Circle A translates horizontally with a degree of freedom xA ,
and triangle B translates vertically with a degree of freedom yB . Figure
4.3a–h show configurations a–h, while figure 4.3i shows the configuration
space partition with these configurations labeled. Configuration a lies in
the single free region. The contact space geometry encodes the parts’
feature contacts. Between configurations b and d, the circle touches the

diagonal edge of the triangle. This corresponds to line segment bd in
the partition. The contact equation is yB ¼ xA 1:29 because the contact
curve is parallel to the triangle’s edge. It indicates a linear relation between
the translations of A and B.
Between configurations d and e, the upper left vertex of the triangle fol-
lows the circle’s boundary, so the corresponding contact curve, de, is a cir-
cular arc segment. Between configurations e and g the circle can translate
and remain in contact with the horizontal edge of the triangle without
pushing the triangle, so the contact curve is a horizontal line segment, eg.
In this interval, the circle blocks the upward translation of the triangle,
which is indicated by the functional relation yB ¼ 1:5. Around configura-
tion h, the triangle can translate without pushing the circle and blocks
its leftward translation, so the contact curve is a vertical line segment,
xA ¼ 3:5. The contact transition configurations b, d, e, and g indicate the
changes in the kinematic function resulting from changes in contact.
4.2 Fixed-Axis Pairs
A fixed-axis pair has a two-dimensional configuration space. The configu-

ration space is a plane for a translation pair, a cylinder for a rotation-
translation pair, and a torus for a rotation-rotation pair. Free space and
blocked space are two-dimensional, while the contact space consists of
curves.
Figure 4.4 shows a rotating cam that drives a translating follower. Recall
the convention that the thick-headed arrow indicates that the cam rotates
Figure 4.4
Cam pair: (a) configuration a ¼ ð0 ; 1:05Þ, (b) configuration b ¼ ð180 ; 2:5Þ, (c) con-
figuration c ¼ ð180 ; 1:05Þ, (d) configuration space partition.
4.2 Fixed-Axis Pairs 61
clockwise and the two-headed arrow indicates that the follower translates
up and down. Free space is a single region since the verticals yA ¼ G180
represent one line in the cylinder’s configuration space. The contact curves
express the vertical distance between the cam’s origin and the follower’s or-
igin as a function of the cam’s angle. For example, yA ¼ 0:55 þ 0:5 ¼ 1:05 at
configuration a, using the displayed dimensions. While the configuration
follows a contact curve, the cam rotates and the follower translates with
nonlinear coupling between the degrees of freedom.
Figure 4.5 shows a rotating one-tooth ratchet with a rotating pawl.
Clockwise rotation of the ratchet is blocked when the pawl tip engages the
ratchet tooth (contact transition configuration a). Since configuration a is a
concave point, a clockwise rotation of the ratchet cannot proceed without
the pawl rotating clockwise as well. This corresponds to the blocking func-
tion of the ratchet-pawl contact. Conversely, contact transition configura-
tions b and c are convex, so clockwise and counterclockwise rotations of
both parts are possible. In this case, the pawl slides over the ratchet tooth.
The verticals yA ¼ G180 and the horizontals yB ¼ G180 each represent
one line in the torus configuration space.
When a rotating part has rotational symmetry or a translating part has
translational symmetry, the configuration space is symmetric in the part’s
degree of freedom. A part with repeated geometry produces repeated con-
tact curves. For example, a ratchet with four identical teeth yields a config-
uration space partition with four shifted copies of the one-tooth contact
curves (figure 4.6).
Figure 4.7 shows the configuration space partition of the spatial Geneva
pair. The free space is a single channel composed of four slanted and
four horizontal segments. The channel width determines the mechanical
play, which governs the coupling between the cam and the follower. The
Figure 4.5
Ratchet pair: (a) configuration a ¼ ð111 ; 30 Þ, (b) configuration b ¼ ð103 ; 25 Þ, (c)
configuration c ¼ ð21 ; 89 Þ, (d) configuration space partition.
Figure 4.6
Four-tooth ratchet pair at a ¼ ð111 ; 30 Þ, (a) configuration, (b) configuration space
partition.
Figure 4.7
(a) A spatial Geneva pair and (b) its configuration space partition.
partition in figure 4.7b shows that the kinematic function is bidirectional

and that the cam blocks the follower’s rotation in the horizontal segments.
The ability to model planar and spatial pairs that have completely different
shapes illustrates the generality of configuration space partitions.
4.3 Partition Algorithm for Fixed-Axis Pairs
We construct the partition of a fixed-axis pair from its contact curves. The
feature contact curves are augmented with boundary lines that enforce the
Figure 4.8
Partition structure of a ratchet pair.
Figure 4.9
Partition algorithm for fixed-axis pairs.
topology of the configuration space. These are yA ¼ G180 when yA is a de-

gree of freedom and yB ¼ G180 when yB is a degree of freedom. The planar
region outside the boundary lines is not part of the configuration space.
The partition is a planar subdivision of the configuration space.
We split the curves at their intersection points. The resulting curve seg-
ments are called edges and their endpoints are called vertices. The comple-
ment of the vertices and edges consists of open regions, called faces, that
are wholly free or wholly blocked. Each face is bounded by one or more
edge loops. We orient the loops so that the face interior lies to the left
when they are traversed in order.
Figure 4.8 shows the partition of a one-tooth ratchet pair. Contact curves
are solid and boundary lines are dashed. Vertices appear as solid dots for
endpoints and gray for intersection points. The edges are bc, cd, dv, vh,
and so on. Face 1 is bounded by the edge loop bcdvhnm and face 4 is
bounded by efv. Faces 1–2 are free and faces 3–5 are blocked.
Figure 4.9 summarizes the partition algorithm. Step 1 forms the contact
curves as described in section 3.2. Steps 2–4 form the vertices, edges, and
faces, and classify each face as free or blocked. The next three sections de-
scribe these steps.
4.3.1 Vertex and Edge Formation

The first task is to compute the intersection points of n contact curves. One
strategy is to test all nðn 1Þ=2 pairs of curves, but this is inefficient when
only a few intersect. We expect a constant number of intersections per
curve because the curves are short and are distributed evenly in the plane.
We employ a standard computational geometry algorithm, called the
sweep-line algorithm, whose running time is Oðn log nÞ in this case. The
second task is to form the vertices and edges.
Sweep-Line Algorithm This algorithm sweeps a vertical line along the x

axis and tracks the y order of the curves that it crosses. The order is stored
in a sweep list. The list is updated when the sweep line reaches a curve’s
endpoints and intersection points. A curve is inserted at its left endpoint
and is removed at its right endpoint. Two curves are swapped in the list at
their intersection point. When two curves become adjacent owing to an
update, they are tested for intersection.
We illustrate the algorithm on the ratchet pair (figure 4.8). Curve ag is
inserted in the empty sweep list. Curve bc is inserted above ag and they
are tested for intersection. Curve bc is removed from the sweep list, cd is
inserted above ag, and they are tested. Curve cd is removed, df is inserted
above ag, and they are tested. Curve ef is inserted between ag and df, and
both pairs are tested. Curve eh is inserted between ef and df, both pairs
are tested, and intersection point v is found. Curves eh and df swap at v,
so the sweep list becomes (ag, ef, df, eh). Curves bc and eh are never tested
because they are disjoint in x. Curves eh and ag are never tested because
they are separated by other curves at every x.
The algorithm is implemented as follows. The events are stored in a pri-
ority queue ordered by x. The queue is initialized with an insert and a re-
move event for each curve. The first event is dequeued and handled until
the queue is empty. Handling updates the sweep list, checks for intersec-
tions between newly adjacent curves, and enqueues swap events when
intersections occur. Vertical curves, curves that overlap over an interval, or
three curves that intersect at a single point are treated as special cases that
we do not describe here.
The sweep list is represented as a balanced binary tree. A curve is inserted
in the standard way: compare its y with the the root curve y at the current
x, recurse on the appropriate subtree, and insert the curve when a leaf is
reached. Finding the curves before and after a curve, removing a curve,
and swapping two curves are standard. Each operation performs at most
log n comparisons. There are two events per curve and one per intersection,
Figure 4.10
Approximation of a contact curve.
which yields order n events, written as OðnÞ, when the number of intersec-
tions per curve is constant. Thus, the total running time is Oðn log nÞ.
We compute the intersection points of two curves by intersecting the
underlying curves and returning the solutions that lie on both curves. The
underlying curves must be intersected numerically because a closed-form
solution does not exist, even for line and circle segments. We approximate
the contact curves to a specified accuracy using line segments and then in-
tersect the approximate curves.
The approximation algorithm, illustrated in figure 4.10, is as follows. Fit a
line through the contact curve’s endpoints, s and e. Sample the curve at
mx ¼ 0:5ðsx þ ex Þ to obtain the point m. If the distance, d, from m to the
line is less than the specified accuracy, the curve approximation is ðs; eÞ.
Otherwise, split the curve at m, recursively approximate the two pieces,
and concatenate the results. In our example, the initial d is too large, so
the curve is split. The sample spacing shrinks from s to e as the curvature
increases.
Edges The edge between two vertices, p and q, is represented by two ori-
ented edges called twins: pq from p to q and qp from q to p. Edge pq is
called outgoing at p and is called incoming at q. Each vertex, v, is assigned
a list of outgoing edges in clockwise order. The list consists of the twins of
the incoming edges in sweep-list order followed by the outgoing edges in
reverse sweep-list order. In the ratchet example (figure 4.8), the incoming
edges of v are ev then dv, and the outgoing edges are vh then vf, so the
edge list is (ve, vd, vh, vf).
4.3.2 Face Formation

The faces are formed by a traversal of the vertex-edge graph. The vertices
are traversed in x order; the outgoing edges of each vertex are traversed in
clockwise order, and a loop is formed for each untraversed edge. The loop is
traversed from its first edge until the first edge is reached again. At each
Figure 4.11
(a) Simple pair and (b) its configuration space partition.
step, the next edge is the successor of the twin of the current edge in the
clockwise order of its head.
In figure 4.8, vertex p is traversed first and yields the loops pabmnhio
and poij ga. Edge ij follows oi in the second loop because the twin of
oi is io and ij follows io in clockwise order. Vertex a yields one loop, ag
jihvfevdcb, since its other outgoing edges were traversed in p loops. Vertex
b yields bcdvhnm. Vertices m, c, d yield no loops. Vertex e yields efv. The
final loop encloses face 5.
The faces are formed along with the loops. The traversal begins with a
single face that is unbounded; hence it has no outer loop. When a vertex
is first traversed, its first loop is an inner loop of the enclosing face. Every
other loop is the outer loop of a new face. In our example, face 4 has an
outer loop efv, face 3 has outer loop ag jihvfevdcb, and the unbounded
face has inner loop pabmnhio. Figure 4.11a shows the simple pair from
section 4.1 and figure 4.11b shows the four faces in its partition. Face 3
has one outer loop and one inner loop.
4.3.3 Face Classification

It remains to classify each face as free or blocked. We select a configuration
in the face, transform the parts to this configuration, and test if they over-
lap. Two parts overlap if their boundaries intersect or if one is inside the
other. The boundaries are tested for intersection with the sweep algorithm.
If they are disjoint, we check if one A boundary point is inside B and vice
versa. We pick a point, form its upward vertical ray, and count the intersec-
tions of the ray with the boundary of the other part. The point is outside
the part when the count is even. Figure 4.12 illustrates the test. Classifica-
tion takes Oðk log kÞ time for the sweep and OðkÞ time for the ray test, with
k the number of boundary curves.
We greatly reduce the classification time with a fast test during face for-
mation that detects most blocked faces. The contact curve for two features
Figure 4.12
Examples of point-in-polygon test: (a) m ¼ 0, (b) m ¼ 1, (c) m ¼ 2, (d) m ¼ 3 where m
is the intersection count of the vertical dotted line.
is a boundary between configurations where the features are disjoint and

where they overlap. We orient the curve so the disjoint configurations are
on the left when it is traversed from tail to head. When the curve is split
into edges, one twin in each pair, called the positive twin, has the same ori-
entation as the curve, and the other twin, called the negative twin, has the
opposite orientation. A face is blocked when its boundary contains a nega-
tive edge since the configurations to the left of this edge are blocked and
are inside the face.
4.4 General Planar Pairs
General planar pairs have spatial configuration spaces. Free space and
blocked space have dimension three. Contact space has dimension two
and is a subset of the feature contact surfaces. The yA ¼ y0 cross-section of
the A=B partition is the planar partition where A translates at orientation y0
relative to B. Cross-sections are useful for visualizing and computing spatial
partitions.
We illustrate spatial partitions on a fastener pair composed of a pin and a
box (figure 4.13). The pin is vertical at yA ¼ 0 (figure 4.13a) and horizontal
at yA ¼ 90 (c). The yA ¼ 0 cross-section (figure 4.13b) has two free regions,
which indicates that the pin cannot translate out of the box. The yA ¼ 90
cross-section (figure 4.13d) has one free region, which indicates that the
pin can translate out of the box. As yA decreases from 90 , the clearance be-
tween the pin and the opening decreases until the pin touches the top and
the bottom of the opening and the cross-section free region splits into two
regions.
Figure 4.13e and f show bottom and top views, respectively, of the yA A
½0 ; 90 part of the spatial partition. The y ¼ 0 cross-section appears in
the bottom view and the y ¼ 90 cross-section appears in the top view.
Figure 4.13
Fastener pair: (a–b) yA ¼ 0 cross-section, (c–d) yA ¼ 90 cross-section, (e–f) bottom
and top views of spatial configuration space partition.
Although some cross-sections have two free regions, the spatial partition
has one free region. The pin can move between any two free configurations
by rotating until it is horizontal, translating to the second position, and
rotating to the second orientation. The pin can be moved out of the box
for yA A ½46 ; 134 .
The next example is a constant-breadth cam with a square follower (fig-
ure 4.14). When the cam rotates, the follower rotates in step and translates
around a circle. In configuration space, the cam moves relative to the fol-
lower. Figure 4.14b shows the free region where the cam is inside the
follower: a curved channel with a square cross-section. Figure 4.14c shows
selected cross-sections projected onto the ðxA ; yA Þ plane. The center of the
square indicates the cam’s position when the follower has zero clearance.
For the actual clearance, the cam can translate around the nominal posi-
tion within the square. The clearance is the same regardless of the orien-
tation. A second free region in which the cam is outside the follower lies
beyond the blocked region that encloses the channel.
Figure 4.14
Cam pair (a), spatial configuration space partition (b), and projected cross-sections (c).
Figure 4.15
Partition structure for a cam pair.
4.5 Partition Algorithm for General Planar Pairs
Spatial partitions are a generalization of planar partitions. The contact sur-

faces split a configuration space into connected components, called cells,
each of which is wholly free or wholly blocked. The surface patches that
bound the cells are called faces; their boundary curves are called edges; and
the edge endpoints are called vertices. Figure 4.15 shows the yA A ½0 ; 3
part of the cam pair partition. The channel is a cell, f and g are two of its
four faces, e is one of its four edges, and there are no vertices.
The spatial partition algorithm (figure 4.16) has the same form as the pla-
nar algorithm, form contact surfaces for all pairs of part features, form the
cells, faces, edges, and vertices, and classify the cells. The first step is
explained in section 4.3. The third step is explained in section 4.3.3. The
only change is that the sample configuration is for A relative to B, rather
than for A and B relative to the mechanism’s frame. It remains to describe
the second step.
We compute the spatial partition by computing the yA ¼ 0 planar parti-
tion and tracking its evolution as yA increases to 360 . As yA increases, the
Figure 4.16
Partition algorithm for general planar pairs.
Figure 4.17
Fastener pair configurations and partitions before (a–b), at (c–d), and after (e–f) a
contact curve tangency.
faces, edges, and vertices of the planar partition trace the cells, faces, and
edges of the spatial partition. In figure 4.15, the planar face bounded by
abcd traces the channel; planar edges a and b trace spatial faces f and g;
and planar vertex v traces spatial edge e. The planar partition is a continu-
ous function of yA when its graph is fixed. The contact curves move, which
makes the edges move, which makes the vertices move, which makes the
faces deform. The graph changes when the edge set or the face set changes.
The edge set changes when yA equals the start or end angle of a contact
surface, so a curve appears or disappears. Figures 3.14, 3.15, and 3.16 show
the three types of contact curves. The edge set also changes when two
curves are tangent. Figure 4.17 shows how a tangency changes the free
Figure 4.18
Fastener pair configurations and partitions before (a–b), at (c–d), and after (e–f) a
face collapses.
space of the fastener pair from one region to two regions. The face set
changes when a face collapses to a point, which occurs when three curves
intersect. Figure 4.18 shows that a fastener pair face collapses when the
upper pin touches the upper slanted box segment (contact a), the lower
pin touches the lower slanted segment (contact b), and the upper pin also
touches the upper horizontal segment (contact c). Contacts a–c are labeled
in the triple contact configuration (figure 4.18c) and in the three partitions
(figure 4.18b,d,f).
The contact curve’s start and end angles are described in section 3.3. Two
curves are tangent when the underlying curves are tangent and the point of
tangency lies on both curves. Tangency occurs when the parametric expres-
sions from section 3.3 satisfy the contact equations from section 3.1.1.
There is one case for each pair of contact types. For example, consider the
contact line of moving circle a and fixed line b, and the contact circle of
moving circle e and fixed circle f . A line with normal n and distance from
the origin k is tangent to a circle with center o and radius r when n o ¼
r þ k. Equations 3.7 and 3.8 yield n ¼ nb , o ¼ of RA oeA , r ¼ jre G rf j, and
k ¼ ra þ db nb RA oaA . The resulting equation has a closed-form solution.
The same is true of the other cases.
Three curves intersect when the underlying curves intersect and the in-
tersection point lies on all three curves. The intersection configurations are
the solutions of the three parametric contact curves. There is one case for
each triple of contact types. Numerical solutions are required.
The planar partition is updated at angles where its graph changes. When
a curve appears, it is added to the partition, which entails splitting the
edges that it intersects and updating the vertices and faces. When a curve
disappears, it is removed from the partition, which entails merging edges
that it formerly split and faces that it formerly separated. When disjoint
curves become tangent and then intersect, the edges that contain the point
of tangency are split and the face that they bound splits into two faces
(figure 4.17b,d,f). When the opposite occurs, the edges that formerly inter-
sected are merged and two faces merge. When a face collapse occurs, the
old face is replaced with a new one.
The spatial partition is constructed from the evolving planar partition.
Planar faces, edges, and vertices generate spatial cells, faces, and edges. A
spatial vertex is formed when two spatial edges from the same surface meet
at its start or end angle, and when two edges from different surfaces meet
at a face collapse. Although the construction is straightforward, there are
many details and special cases.
4.6 Mechanisms
We have seen that the partition of the configuration space of a kinematic

pair models its parts’ contacts. We model the contacts among the parts
of a mechanism in the same way. The pairwise contacts are modeled by
constructing a configuration space partition for each kinematic pair in the
mechanism. Contacts among three or more parts are modeled by compos-
ing the pair partitions to form a partition of the mechanism’s configuration
space. The configuration space partition of the mechanism consists of free
regions where all the pairs are free, blocked regions where some pairs are
blocked, and contact regions where some pairs are in contact and the rest
are free. The contact region geometry encodes the coupling among the
degrees of freedom of the touching parts.
We illustrate configuration space partitions for a mechanism by using a
pawl indexing mechanism composed of a driver D, indexer I, and pawl P
(figure 4.19). As the driver rotates clockwise, its finger engages an indexer
pin, rotates the indexer counterclockwise by one-tenth of a turn, and disen-
gages. When the driver is disengaged (figure 4.19a), its rim aligns with the
left pawl arm, which prevents clockwise rotation, and the right pawl arm
prevents indexer rotation by engaging two pins. When the driver engages
the indexer, its notch aligns with the left pawl arm, which allows the bot-
4.6 Mechanisms 73
Figure 4.19
Indexing mechanism: (a) disengaged, (b) engaged, (c) disengaging.
Figure 4.20
Indexing mechanism pair configuration space partitions: (a) driver-indexer (detail),
(b) driver-pawl, (c) indexer-pawl.
tom indexer pin to rotate the right pawl arm out of the way (figure 4.19b).
When it disengages, the left pawl arm returns to its rim and the right arm
re-locks the indexer (figure 4.19c).
Figure 4.20 shows details of the configuration space partitions of the
indexing mechanism’s pairs. The driver-indexer free space consists of ten
diagonal channels where the finger engages the ten pins. The negative
slope indicates that the parts rotate in opposite directions. The rest of the
free space is the regions to the left and right of the channels where the
driver is disengaged. The driver-pawl free space is bounded from below by
the rim-left arm contact curve, which is horizontal, and by the notch-left
arm contact curves, which form a v-shape (figure 4.20b). The horizontal
curve blocks downward vertical motion, clockwise pawl rotation, whereas
the slanted curves permit it. The indexer-pawl free space (figure 4.20c) is
bounded from above by the pin-right arm contact curves, which form ten
inverted v-shapes. When yI increases or decreases, yP decreases along one
Figure 4.21
Illustration of the configuration space partition of the indexer mechanism as a com-
position of its pair configuration space partitions.
side of the v-shape. In other words, indexer rotation in either direction

causes clockwise pawl rotation.
The kinematic function of a mechanism involves simultaneous contact
of parts. For example, the driver is in contact with the indexer and the
pawl in configuration a, whereas the indexer and the pawl are not in con-
tact (figure 4.19a and configuration a in figure 4.20). As the driver rotates
clockwise, the indexer and the pawl also come into contact (figure 4.19b
and configuration b in figure 4.20). Partition of the mechanism’s configura-
tion space encodes this contact information.
Figure 4.21 illustrates partition of the indexer’s configuration space. We
focus on a small region because the full partition is difficult to visualize.
Each configuration in the partition of the pair’s configuration space is
extended to a line parallel to the third configuration space parameter in
partition of the mechanism’s configuration space. The intersection of the
three lines corresponds to the mechanism’s configuration space. For exam-
ple, configuration a ¼ ð14 ; 0 Þ in the driver-pawl pair configuration space
is replaced by the line ð14 ; 0 ; yI Þ; configuration ð14 ; 0 Þ in the driver-
indexer pair configuration space is replaced by the line ð14 ; yP ; 0 Þ; and
configuration ð0 ; 0 Þ in the indexer-pawl pair configuration space is

replaced by the line ðyD ; 0 ; 0 Þ. They intersect at configuration a ¼
ð14 ; 0 ; 0 Þ of the mechanism.
The computation of configuration space partitions for mechanisms is an
active research topic in robotics and in computational geometry. One ap-
proach employs set operations. Free space is the intersection of the free
spaces of all the pairs because a configuration is free when every pair is
free. Blocked space is the union of pair blocked spaces because a configura-
tion is blocked when any pair is blocked. Contact space is the complement
of the union of free space and blocked space. In set operations, the pair
configuration a is replaced by the set of mechanism configurations ða; bÞ
where b ranges over the degrees of freedom of the other parts.
The computational complexity of computing a configuration space parti-
tion has been proved to grow exponentially with the number of a mecha-
nism degrees of freedom. Even a single spatial pair, which has six degrees of
freedom, appears unmanageable. Although computing the entire partition
is impractical, partial computation is a valuable design tool. The simplest
case is to construct a mechanism contact region from a set of pair contact
regions. For kinematic analysis, the partitions of the pair configuration
space provide insights into the individual workings of the pairs and their
role in the mechanism’s kinematic function.
4.7 Partitions of Configuration Space and the Theory of Mechanisms
The configuration space model of contact of parts is an extension of the

classic models in the theory of mechanisms. The theory focuses on linkage,
cam, and gear mechanisms. In each case, the mechanism is modeled under
restrictive assumptions about the parts’ contacts. The classic models are
equivalent to the contact spaces in the configuration space models of the
mechanisms.
Cams A cam is assumed to be in permanent contact with its follower, with

the follower’s position a function of the cam’s orientation. In our cam ex-
ample (figure 4.4), xB is the function of yA given by the lower boundary of
the upper free space region. Intermittent contact is not modeled. We treat a
cam and its follower as a higher pair, so permanent contact and intermit-
tent contact are modeled.
Gears Gears are assumed to be in permanent contact. Normally the con-

tact is via involute teeth and each gear has one degree of freedom, so the
Figure 4.22
Gear pair with detail (a–b) and partition with detail (c–d).
Figure 4.23
Half-gear pair and its partition.
mechanism has one degree of freedom. Imperfect teeth, clearance, under-

cutting, hub play, and other changes in contact are not modeled. We treat
gears as higher-pair mechanisms. Gears with hub play have three degrees of
freedom, as do floating gears. Other gears are modeled as fixed-axis parts.
Figure 4.22 shows a fixed-axis gear pair in which the involutes are ap-
proximated with circle segments. As gear A rotates counterclockwise, the
contact point shifts from feature pair a to feature pair b at configuration c
(figure 4.22b,d). The a and b contact curves are nonlinear and intersect
transversely at c, whereas true involute contact curves are parallel line seg-
ments that meet at c. The free space represents clearance between the gears.
When A is stationary, B can rotate by the vertical extent of the free space.
True involutes have zero clearance, so the free space is empty.
The standard gear model does not cover sector gears, such as the half-
gear pair in figure 4.23. Meshing and unmeshing involve contact changes
between noninvolute tooth features. Partitions model uncoupled motion,
Figure 4.24
Crank-and-slider configurations: (a) a ¼ ð30 ; 17 ; 2:9Þ, (b) b ¼ ð135 ; 25 ; 1:0Þ, (c)
c ¼ ð150 ; 17 ; 0:9Þ, (d) d ¼ ð45 ; 25 ; 2:7Þ.
coupled motion, and contact changes. In our example, the free space con-
sists of 16 diagonal channels where the gears mesh, surrounded by a region
where they disengage. The 16 channels represent the 16 teeth of B that can
mesh with the lead tooth of A.
Linkages A linkage is assumed to consist of links connected by joints. The

configuration variables satisfy equality equations whose parameters are the
link lengths and the joint configurations. Most linkages have one degree of
freedom, so all the variables are determined once a single variable is known.
Joint limits, joint play, and link interference are not modeled. We treat a
linkage as an assembly of lower and higher pairs. Lower pairs model the
joints for which the classic assumptions are acceptable. Higher pairs model
the other part contacts.
We illustrate linkage partitions on a three-link crank-and-slider mecha-
nism (figure 4.24). Link A is attached to the frame by a revolute joint at
o ¼ ð0; 0Þ and link C is attached to the frame’s x axis by a prismatic joint.
Link B is attached to A and C by revolute joints at p and q. The link’s coor-
dinate frames are placed at o, p, and q with the x axes parallel to the links.
The degrees of freedom are yA , xB , yB , yB , xC with xA ; yA ; yC ; yC ¼ 0. The con-
tact equations are
lA cos yA ¼ xB
lA sin yA ¼ yB
Figure 4.25
Kinematic function of a crank and slider.
xB þ lB cos yB ¼ xC
yB þ lB sin yB ¼ 0;
with lA and lB the link’s lengths. These four equations define a curve in the
five degrees of freedom. We can eliminate xB and yB from the last three
equations using the first two equations. The reduced system defines a curve
in ðyA ; yB ; xC Þ whose projection onto ðyA ; xC Þ is the linkage input-output
function (figure 4.25).
4.8 Notes
The contact of parts is closely related to the contact of features and has
been extensively studied in the mechanical engineering and robotics litera-
ture. Mechanical engineering research addresses part contacts for cam, gear,
and linkage mechanisms. In cam and gear mechanisms, the contact se-
quence between parts is predetermined [4, 26, 51], and feature contacts are
determined as described in chapter 3. Linkage mechanisms are described as
rigid links connected by standard joints with no play.
Robotics studies the contact of parts within the configuration space
method for planning robot motion [16, 47, 52]. The research provides algo-
rithms for partitioning configuration space for planar polygons [11, 52]
and partial algorithms for polyhedral parts [20]. The research does not ad-
dress curved parts. Bajaj and Kim [6, 7] describe algorithms for computing a
configuration space partition for parts with algebraic curves and surface
contours.
For a detailed description of the sweep-line algorithm and related compu-
tational geometry tasks, see Berg et al. [18].
4.8 Notes 79
Reuleaux’s seminal work [67] analyzes mechanisms as collections of low-

er and higher pairs whose kinematics is derived by composing a pair’s kine-
matics. The configuration space partition of mechanisms introduced in this
chapter formalizes this compositional approach.
We have studied the contact of parts in mechanisms within configura-
tion space for nearly 20 years [36]. We have studied part contacts for planar
fixed-axis pairs [71], general planar pairs [68], and spatial fixed-axis pairs
[42]. Other work includes that by Falting [23, 24].
Appendix B describes the configuration space partitions constructed by
HIPAIR. The program handles fixed-axis planar mechanisms.
5 Analysis
The next four chapters explore kinematic design using configuration spaces.
This chapter addresses analysis. The purpose of analysis is to determine
how a mechanism functions. Kinematic analysis studies the functional
consequences of the requirement that the configuration of every mecha-
nism must lie in free space or in contact space. Partitions of the configura-
tion space provide a geometric view of these spaces that facilitates the
analysis. The feature contact spaces reveal the structure of the individual
contacts, while their intersection spaces reveal the coupling among con-
tacts. The benefit is greatest when seemingly unrelated features interact.
In section 5.1 we discuss kinematic analysis through an examination of
configuration space partition. The strategy is to relate the geometric proper-
ties of partitions to functional properties and then to compare mechanisms
by comparing their geometric properties. We illustrate this strategy with
several examples. The remainder of the chapter is devoted to simulation.
Simulation predicts the function of a mechanism by solving differential
equations that express Newton’s laws of motion. In kinematic simulation,
the initial values and velocities of the mechanism’s degrees of freedom
are given, whereas in dynamical simulation the initial state and the exter-
nal forces are given. Simulation is less general than examination in that
many inputs may be required to capture every aspect of the mechanism’s
function. We present a kinematic simulation algorithm in section 5.2 and
extend it to dynamical simulation in section 5.3. The use of configuration
space partitions enables us to predict contact changes, whereas prior simu-
lators are limited to mechanisms with permanent contacts or with known
contact changes.
5.1 Kinematic Analysis by Examination of Partition
We saw in chapter 4 that an examination of a mechanism’s configuration

space partitions provides qualitative and quantitative information about its
82 5 Analysis
Figure 5.1
Disk indexing pair: (a) drive configuration a, (b) dwell configuration b, (c) drive con-
figuration c in next period.
kinematic function. Kinematic function is typically specified with mixed

structural, behavioral, functional, and historical descriptions, such as
switching, balancing, engaging, locking, and braking. While a systematic
configuration space partition analysis is desirable, it would require a formal
specification of kinematic function and a classification of mechanisms into
a functional taxonomy, neither of which is available in the mechanical
engineering literature. Instead we demonstrate the efficacy of an examina-
tion of partition with numerous examples throughout the book, including
a catalog of mechanisms (appendix A).
In this section we examine two fixed-axis indexing pairs with similar ki-
nematic functions. Although the part geometries are different, the configu-
ration space partitions are similar. Changing the nominal values of the
design parameters leads to changes in kinematic function, including failure
modes, that are apparent in the configuration space partitions. We then ex-
amine an intermittent gear mechanism with two general planar pairs.
5.1.1 Disk Indexing Pair

The first indexing pair is shown in figure 5.1. The driver is a disk with a slot
cutout. The indexer has ten circular pins evenly spaced on a disk and is
spring-loaded clockwise. Continual counterclockwise rotation of the driver
causes intermittent clockwise rotation of the indexer. In drive periods, the
driver slot engages an indexer pin and advances the indexer one-tenth of a
turn. In dwell periods, the outer arc of the driver engages two adjacent in-
dexer pins and prevents rotation. When a dwell period ends, the indexer
spring rotates the next indexer pin into the driver slot and the next drive
period begins.
Figure 5.2
(a) Configuration space partition and (b) detail.
Figure 5.2a shows the configuration space partition with the three config-
urations from figure 5.1. There is one free region with ten slanted segments
that represent drive periods and ten horizontal segments that represent
dwell periods. The segment width corresponds to the play between the in-
dexer and the driver. The play is largest at transitions between drive and
dwell periods. The partition reveals a failure mode. If the spring is too
weak, the driver slot will pass the indexer pin before it can engage, so the
indexer will not advance (the path in figure 5.2b from configurations b to
c instead of from b to a).
We present two examples of how changes in the partition reflect changes
in kinematic function. Figure 5.3 shows the effect of decreasing the dis-
tance between the centers of rotation by 1% from 33 mm to 32.7 mm.
The free space splits into ten regions where the driver blocks on the ten in-
dexer pins. The driver can rotate one pin, but cannot disengage from it and
rotate to the next pin (figure 5.3c). Figure 5.4 shows the effect of decreasing
the width of the driver slot mouth by 15% from 6 mm to 5.1 mm. The free
space again contains ten regions, which show a second type of blocking
(figure 5.4b). The driver can engage and disengage the indexer pins, but
cannot rotate the indexer by one-tenth of a turn (figure 5.4c).
5.1.2 Lever Indexing Pair

The second indexing pair is shown in figure 5.5. The driver consists of a
driving lever and a locking arc mounted on a disk (not shown). The indexer
has eight pins evenly spaced on a disk. Continual counterclockwise rota-
tion of the driver causes intermittent clockwise rotation of the indexer. In
84 5 Analysis
Figure 5.3
(a) Modified disk indexing pair with centers of rotation too close, (b) partition, (c)
detail.
Figure 5.4
(a) Modified disk indexing pair with slot too narrow, (b) partition, (c) detail.
drive periods, the driver lever engages an indexer pin and advances the in-
dexer one-eighth of a turn. In dwell periods, the locking arc engages two
adjacent indexer pins and prevents rotation.
Figure 5.6 shows the configuration space partition with the six configura-
tions from figure 5.5. The slanted and horizontal segments in figure 5.6a
represent the drive and dwell periods. The intended kinematic function
occurs along the path abcde in figure 5.6b. The partition reveals two failure
modes. The indexer can rotate clockwise from b to f and block. It can also
rotate counterclockwise from d to a and repeat the previous cycle.
We can remove both failures by modifying the driver (figure 5.7). The
first failure occurs when the leading side of the locking arc touches an
Figure 5.5
Lever indexing pair: (a) drive configuration a, (b) transition to dwell at configuration
b, (c) dwell configuration c, (d) end of dwell at configuration d, (e) start of next drive
period at configuration e, (f) blocking configuration f.
Figure 5.6
(a) Configuration space partition, and (b) detail.
indexer pin and forms a contact curve with a negative slope. As the driver
angle increases, the indexer angle decreases until the configuration reaches
a blocking configuration (like the ratchet pair in figure 4.5). We assign the
contact curve a positive slope, which eliminates the blocking configura-
tion, by beveling the locking arc side. The second failure occurs because
the locking arc needs to engage two indexer pins for dwell. It must dis-
engage one pin before the dwell ends, so that it can disengage the second
86 5 Analysis
Figure 5.7
(a) Revised lever indexing pair, (b) configuration space partition, and (c) detail.
pin at the start of the next drive period. We prevent indexer rotation by
increasing the outer radius of the locking arc so it locks another pin.
The configuration space partition of the modified pair resembles that of
the spatial Geneva pair (figure 4.7). Both partitions exhibit slanted driving
segments linked by horizontal dwell segments. The number of segments
equals the drive ratio of the pair. The Geneva partition has smooth seg-
ment boundaries, which indicates smooth motion of parts, whereas the
second partition has irregular boundaries. The partition of the first indexer
pair resembles these partitions since all three pairs have the same kinematic
function, but it is also spanned by horizontal segments, owing to its failure
mode (figure 5.1).
5.1.3 Intermittent Gear Mechanism

General planar pairs have more complex kinematic functions than fixed-
axes pairs because of their extra degree of freedom. Direct examination of
the three-dimensional partition of the configuration space is practical in
relatively simple pairs, such as the constant-breadth cam pair in section
4.4. Another strategy is to examine partition cross-sections, which repre-
sent translation of a part at a fixed orientation.
We illustrate cross-section examination using the pawl indexer gear
mechanism in figure 5.8. The mechanism consists of a constant-breadth
cam, a follower with two pawls, and a gear with inner teeth. The cam and
the gear are mounted on a fixed frame and rotate around their centers with
degrees of freedom yC and yG ; the follower is free with degrees of freedom
ðxF ; yF ; yF Þ. Rotating the cam causes the follower to rotate in step while
reciprocating along its length. The right follower pawl engages a gear tooth
at cam angle yC ¼ 0 (figure 5.8a); the follower rotates the gear 45 ; the
Figure 5.8
Intermittent gear mechanism: (a) right follower pawl engaged, (b) follower dis-
engaged, (c) left pawl engaged.
right pawl disengages at yC ¼ 82 (figure 5.8b); the follower rotates inde-
pendently while the gear dwells; then the left pawl engages the gear at
yC ¼ 90 and the cycle repeats (figure 5.8c).
Figure 5.9 shows the cross-sections of the gear-follower partition that cor-
respond to the snapshots. The follower’s configuration is ð0; 0; 0Þ and the
gear degrees of freedom are ðxG ; yG Þ. In figure 5.9a, the free space consists
of three regions. In the middle region, which contains the snapshot config-
uration, the right follower pawl engages a gear tooth. The gear motion is
about 10 mm, whereas its outer radius is 175 mm. In the region above or
below, the pawl engages the tooth below or above. In figure 5.9b, the three
regions have merged into a single region, which shows that the follower
can now cross over the two gear teeth. In figure 5.9c, the free space once
again consists of three regions and the middle region contains the snapshot
configuration in which the left pawl engages a gear tooth.
5.2 Kinematic Simulation
Simulation is a method of computing the state of a time-varying system.

The state at time t is xðtÞ, it initial value is xðt0 Þ, and its time derivative is
x_ ðtÞ. The system satisfies a system of ordinary differential equations, x_ ¼
fðx; tÞ. For example, x_ ¼ 2tx 2 with xð0Þ ¼ 1 yields xðtÞ ¼ 1=ð1 þ t 2 Þ. Al-
though few systems have closed-form solutions, there are many efficient
numerical solvers that generate sequences of approximate solution points
xðt 0 Þ; xðt1 Þ; . . . ; xðtn Þ. Each point is derived by a formula that contains ear-
lier points and f values. The simplest formula, Euler’s method, is xðt þ hÞ
¼ xðtÞ þ hfðxðtÞ; tÞ with h a small constant. Better formulas employ more
points and f values.
88 5 Analysis
Figure 5.9
Gear-follower partition cross-sections: (a) yG ¼ 0 , (b) yG ¼ 5 , (c) yG ¼ 9 .
The state of a mechanism is its configuration. Kinematic simulation as-

sumes that the derivative is a function of a driving motion: a velocity for
one or more parts over a time interval. The parts’ velocities are determined
by propagating the driving motion through the parts’ contacts. Parts in
contact are assigned velocities that make the relative velocity at the contact
point tangent to the parts’ boundaries. Equivalently, the velocity vector is
tangent to the contact space.
Figure 5.10 shows a kinematic simulation of the figure 4.3 pair. Circle A
translates horizontally with a degree of freedom xA and triangle B translates
vertically with a degree of freedom yB . The state is x ¼ ðxA ; yB Þ with x_ ¼
ðx_ A ; y_ B Þ, the initial state is xð0Þ ¼ a, and the driving motion is x_ A ¼ 1 for
t A ½0; 4. In free configurations a, e, y_ B is zero. In contact configurations
b–d, y_ B is chosen to make x_ tangent to the contact space normal, n, which
gives the equation n x_ ¼ nx x_ A þ ny y_ B ¼ 0. Substituting n ¼ ð1; 1Þ and
x_ A ¼ 1 yields y_ B ¼ 1.
5.2.1 Simulation Algorithm

Standard simulation algorithms require continuous simulation formulas.
However, kinematic equations are discontinuous at changes in contact.
The discontinuities are handled by splitting the simulation into intervals
of fixed contact and simulating each interval separately. In the example of
Figure 5.10
(a–e) Kinematic simulation of a translation pair, (f) time plots, (g) motion path in
configurations a–e configuration space.
Figure 5.11
Kinematic simulation algorithm.
figure 4.3, the fixed contact intervals are from a to b, from b to d, and from
d to e.
Figure 5.11 summarizes the kinematic simulation algorithm. The input is
a mechanism, an initial configuration xð0Þ, a driving motion d, a stop time
te , a time step h, and an accuracy . The output is a sequence of configura-
tions xðt 0 Þ; . . . ; xðte Þ. Configuration xðtiþ1 Þ is computed by integrating the
equations x_ ¼ fðx; tÞ from ti to tiþ1 with accuracy . When the contacts of
the parts are fixed on ½ti ; ti þ h, tiþ1 equals ti þ h; otherwise, tiþ1 is the con-
tact change time.
90 5 Analysis
Figure 5.12
Collision: (a) free configuration a, (b) overlap configuration b, (c) contact configura-
tion c, (d) configuration space partition.
Figure 5.13
Contact breaks: (a) contact configuration a, (b) free velocity configuration b, (c) free
configuration c, (d) configuration space partition.
The inputs to the integrator are the pairs’ partitions, xðti Þ, d, h, , and the
velocity computation module described later. The integrator first computes
xðti þ hÞ under the assumption that the contacts are fixed on ½ti ; ti þ h. It
uses a standard ordinary differential equation solver that invokes the veloc-
ity module multiple times to ensure accuracy. It then tests for contact
changes. Two parts make contact when their configuration enters blocked
space. A contact breaks when the velocity points into free space. Figures
5.12 and 5.13 illustrate this using the driver-indexer pair of the indexing
mechanism from section 4.6. When a change is detected, the integrator
finds the change configuration by a bisection search on ½ti ; ti þ h.
5.2.2 Computation of Velocity for Fixed-Axis Mechanisms

The input to the velocity computation algorithm for fixed-axis mecha-
nisms is a configuration and a driving velocity. The algorithm assumes one
driving part, one contact per pair, and no cycles in the topology graph of
the mechanism. When these assumptions are violated, the simulation is
overconstrained, so the mechanism blocks.
We first discuss a pair with x ¼ ðxA ; xB Þ and a driving velocity x_ A . Define
v ¼ ðx_ A ; 0Þ. When x is free, x_ B ¼ 0 and x_ ¼ v. Suppose contact occurs with
contact normal n. If n v b 0, v points into free space or is tangential to
Figure 5.14
Blocked simulation: (a–b) configurations a–b, (c) motion path.
Figure 5.15
Computation of a mechanism’s velocity.
contact space, A does not drive B and x_ ¼ v. The translation pair exhibits
this behavior in configuration d in figure 5.10d, g. If n v < 0, v causes
overlap of parts and a tangential x_ is obtained by solving n x_ ¼ nx x_ A þ
ny x_ B ¼ 0 for x_ B . There is no solution when ny ¼ 0, so we set x_ ¼ ð0; 0Þ. The
simulation blocks because the driving velocity v is parallel to the contact
normal n (figure 5.14).
The computation algorithm for a mechanism’s velocity is as follows. The
topology graph for the input configuration is traversed starting at the driv-
ing part. When part B is traversed from A, the pair algorithm is invoked
with A as the driving part. If B is assigned a velocity, its children are tra-
versed. In figure 5.15, part A is assigned a driving velocity x_ A , B is traversed
from A and is assigned x_ B ¼ x_ A , D is traversed from B and is assigned
x_ D ¼ x_ B , C is traversed from A without being assigned a velocity, and so E
is not traversed.
Figure 5.16 shows a kinematic simulation of the indexing mechanism.
The driving velocity is y_D ¼ 1 hertz. In figure 5.16a, the driver assigns
y_I to the indexer, which assigns y_P to the pawl. The indexer time plot (fig-
ure 5.16d) shows that y_I is positive. The driver-indexer partition (figure 5.16f)
shows that this is because y_D and the contact curve slope are negative. The
92 5 Analysis
Figure 5.16
Kinematic simulation of an indexing mechanism: (a–c) configurations a–c, (d–e)
time plots, (f–h) motion paths in configuration space.
pawl time plot (figure 5.16e) shows that y_P is negative and the indexer-pawl
partition (figure 5.16h) explains why. In figure 5.16b, the driver assigns
velocities to the indexer and to the pawl. In figure 5.16c, the driver assigns a
velocity to the pawl and the indexer is stationary.
5.2.3 Computation of Velocity for General Planar Mechanisms

The algorithm for general planar mechanisms allows multiple driving parts
and general topology graphs for a mechanism. The driving velocity is
expressed as d ¼ ðx_ 1 ; y_ 1 ; y_1 ; . . . ; x_ m ; y_ m ; y_m Þ with m the number of moving
parts. The driving parts contribute non-zero elements to d and the other
elements are zero.
A contact between parts i and j is analyzed in the i-j configuration space
with coordinates ðxi ; yi ; yi ; xj ; yj ; yj Þ. The driving velocity of the pair is dij ¼
ðx_ i ; y_ i ; y_i ; x_ j ; y_ j ; y_j Þ and the contact normal is nij . If nij dij > 0, the contact
breaks immediately and is ignored. Otherwise, overlap of parts is prevented
by the velocity along the contact normal. The velocity is expressed as kn
with n the injection of nij into the mechanism’s coordinates. This means
that n is a 3m-vector whose only non-zero elements are the first three ele-
ments of nij in positions 3i, 3i þ 1, 3i þ 2 and the last three elements in
positions 3j, 3j þ 1, 3j þ 2.
When part j is the frame, the j elements are omitted from dij and nij . A
lower pair is modeled with two contacts, as explained in chapter 3: two
point and circle contacts for a revolute joint and two point and line con-
tacts for a prismatic joint. These contacts always generate velocities because
they cannot break.
Pc
The mechanism’s velocity is x_ ¼ d þ i¼1 ki ni with c the number of con-
tacts. We compute ki from the c equations x_ ni ¼ 0, which state that x_ is
tangent to the contact normals. The resulting symmetric linear system
2 32 3 2 3
n1 n1 n1 nc k1 n1 d
6 .. .. .. 766 . 7
7 6 6 .. 7
7
4 . . . 54 .. 5 ¼ 4 . 5
nc n1 nc nc kc nd d
is solved with a linear equation solver. When the equations are singular,
the simulation is overconstrained, so we assign the parts zero velocity.
We illustrate the algorithm using the constant-breadth cam pair from
section 4.4 (figure 5.17). The cam is part 1 and the follower is part 2. The
driving velocity is y_1 ¼ 1 hertz, so d ¼ ð0; 0; 1; 0; 0; 0Þ. The revolute joint
that connects the cam to the frame is modeled with contacts between
cam points a 1 ¼ ð1; 0Þ and b 1 ¼ ð0; 1Þ and the frame circle, s, with a center
94 5 Analysis
Figure 5.17
Kinematic simulation of a cam pair: (a–d) configurations a–d, (e–g) time plots.
ob ¼ ð0; 0Þ and radius rb ¼ 1. The contact equations are kak ¼ 1 and

kbk ¼ 1, which yields
x12 þ y12 þ 2x1 cos y1 þ 2y1 sin y1 ¼ 0
x12 þ y12 2x1 sin y1 þ 2y1 cos y1 ¼ 0:
The contact normals (after division by 2) are

n1 ¼ ðx1 þ cos y1 ; y1 þ sin y1 ; y1 cos y1 x1 sin y1 ; 0; 0; 0Þ
n2 ¼ ðx1 sin y1 ; y1 þ cos y1 ; y1 sin y1 x1 cos y1 ; 0; 0; 0Þ:
Figure 5.17a shows the configuration ð0; 0; 0; 0; 0; 0Þ where the cam-

follower pair is free, so the mechanism’s velocity is x_ ¼ d þ k1 n1 þ k2 n2 .
The linear equations are k1 ¼ 0 and k2 ¼ 0, since n1 ¼ ð1; 0; 0; 0; 0; 0Þ and
n2 ¼ ð0; 1; 0; 0; 0; 0Þ, so x_ ¼ d. Figure 5.17b shows the configuration ð0; 0;
29:81; 0:18; 0:04; 23:94Þ where the pair is in contact, so the velocity is x_ ¼
d þ k1 n1 þ k2 n2 þ k3 n3 . The contact is between cam point p 1 ¼ ð4:3; 2:5Þ
and the follower line with normal n 2 ¼ ð1; 0Þ and distance d from the
origin. The contact equation is n f ¼ d with f ¼ p t2 , which yields
n3 ¼ ðcos y2 ; sin y2 ; k; cos y2 ; sin y2 ; fx sin y2 fy cos y2 Þ
k ¼ ð px cos y2 þ py sin y2 Þ sin y1 þ ð py cos y2 px sin y2 Þ cos y1 :

The contact normals are n1 ¼ ð0:87; 0:5; 0; 0; 0; 0Þ, n2 ¼ ð0:5; 0:87; 0; 0;

0; 0Þ, and n3 ¼ ð0:91; 0:41; 2:04; 0:91; 0:41; 2:01Þ. The linear equations
are
2 32 3 2 3
1 0 0:99 k1 0
6 76 7 6 7
4 0 1 0:1 54 k2 5 ¼ 4 0 5
0:99 0:1 10:2 k3 2:04
and x_ ¼ ð0; 0; 0:55; 0:2; 0:09; 0:45Þ. The driver rotates, the follower rotates
and translates, and the sum of the angular velocities equals the driving an-
gular velocity. Unlike the fixed-axis algorithm, the driving velocity is modi-
fied in the same way as the other velocities of parts.
5.3 Dynamical Simulation
Kinematic simulation presupposes that the driving velocities are known.

This is a reasonable assumption when every degree of freedom in the mech-
anism is under accurate control. For example, the driver’s orientation in
our driver-indexer pairs can be controlled with a torque controller on a tor-
sional motor. More generally, each part of a mechanism moves according
to the Newtonian equations
mx€ ¼ fx þ cx
my€ ¼ fy þ cy
I y€ ¼ t þ cy
with m the mass of the part, x€ and y€ the components of its linear accelera-
tion, fx and fy the external force acting on the part, cx and cy the contact
force, I the moment of inertia, y€ the angular acceleration, t the external
torque, and cy the contact torque.
The rigid part model contradicts the Newtonian equations when parts
collide. The parts must change velocity instantaneously to prevent overlap,
but the velocity is continuous and only acceleration can change instanta-
neously. The contradiction is resolved by employing a separate collision
model. When the parts are smooth, the postcollision velocities are deriva-
ble from the conservation of momentum. Continuum mechanics provides
a more general model in which parts deform during collision, then return
to their original shapes.
Numerical solution of the extended Newtonian equations is called
dynamical simulation. It provides designers of mechanisms with informa-
tion that goes beyond the realm of kinematics. The most robust designs
96 5 Analysis
Figure 5.18
Dynamical simulation of escapement: (a–d) configurations a–d.
have a single kinematic function, meaning a single motion path in con-

figuration space, so dynamical simulation differs from kinematic simula-
tion solely in the rate at which this path is traversed. When the function
depends on dynamics, dynamical simulation identifies the actual func-
tion of a specific design.
Example We illustrate dynamical simulation for a fixed-axis escapement

composed of a wheel, an anchor, and a ratchet (figure 5.18). The kinematic
function is to rotate the ratchet clockwise by two teeth per second. A spiral
spring oscillates the wheel twice per second. The wheel pin engages the an-
chor fork and oscillates the anchor. A weight applies a constant clockwise
torque to the ratchet. The cycle begins with the wheel pin at its leftmost
point (figure 5.18a). The left anchor pallet engages a ratchet tooth and
blocks its rotation. As the wheel rotates counterclockwise, the anchor
rotates clockwise. When the wheel pin is at its lowest point, the ratchet dis-
engages the left anchor pallet (figure 5.18b), rotates clockwise by one tooth,
and engages the right anchor pallet (figure 5.18c). As the wheel rotates
clockwise, the right anchor pallet disengages (figure 5.18d), the left pallet
engages, and the cycle repeats.
Dynamical simulation verifies the kinematic function for one design in-
stance. The moments of inertia are 1 N cm 2 (newton for square centi-
Figure 5.19
Anchor-ratchet motion paths: (a) correct, (b) failure.
meter) for the balance, 1 N cm 2 for the anchor, and 5 N cm 2 for the
escapement. Friction is neglected. A spring coefficient of p 2 produces a nat-
ural spring period of two cycles per second. Simulating the mechanism
with a range of driving torques reveals that it works with 20 N cm, but
fails with 30 N cm. Figure 5.19 shows the two anchor-ratchet motion
paths. The bad path (figure 5.19b) starts correctly but fails to clear the hor-
izontal contact curve at configuration d, reverses direction, and blocks in
configuration c.
Algorithm The main steps in dynamical simulation are computation of

contact force, impact handling, and integration. Contacts between smooth
parts generate forces along the contact normals that are computed by solv-
ing linear equations, just as velocities are computed in kinematic simula-
tion. Contacts between rough parts generate tangential, frictional forces
that are harder to compute. Impact handling updates the velocities of col-
liding parts according to the conservation of momentum, continuum me-
chanics, or variants thereof. The integration algorithm is the same as in
kinematic simulation except that it uses the Newtonian equations and
invokes the impact handler on collisions.
Dynamical simulation applies to spatial parts. Each part has three rota-
tional and three translational equations. Collisions are detected by testing
the parts for overlap since partitions of configuration space are impractical.
Efficient detection of collisions is widely studied in mechanical engineering
and robotics.
98 5 Analysis
Simulation with Configuration Space Partitions The key advantage of us-

ing configuration space partition for simulation is that it provides a model
for contact of parts. This model is an effective alternative to precomputed
pairwise contacts and collision detection. Partition of the configuration
space encodes the contact conditions and relations that must be incorpo-
rated in the equations for the motion of parts. It thus allows the efficient
and robust simulation of mechanisms with complex contact geometry and
complex changing sequences of contact.
5.4 Notes
Analysis of mechanisms plays a central role in mechanical engineering. The

main analysis tools are kinematic and dynamical simulation, which have
been the subject of extensive research in the past two decades. A variety of
commercial and academic software packages are currently available, includ-
ing ADAMS, SIMPACK, DADS, and SD/FAST, to name a few. The simulators
handle multibody systems composed of parts connected by permanent
joints. For textbooks on simulating mechanisms, see de Jalon and Bayo
[32], Haug [28], Schiehlen [77], and Nikravesh [59].
We have studied analysis in the configuration space paradigm. We
show how configuration space partition regions and their adjacencies cor-
respond to kinematic function for fixed-axis mechanisms [36]. Based on
this representation, we developed a symbolic language to describe kine-
matic function [39], that can be used to compare and classify mechanisms
[37, 38]. We show how to compute and interpret the configuration space
partition for fixed-axes mechanisms [40]. See also related work on qualita-
tive kinematics by Faltings [23, 24].
Appendix B describes the HIPAIR kinematic simulation format and visu-
alization control.
6 Tolerancing
This chapter addresses tolerancing, which is the task of transforming a

nominal design into one that can be manufactured. Every manufacturing
process is subject to variation in the shape and configuration of parts that
causes kinematic variation. The manufacturing cost generally increases as
the variation decreases. The designer needs to select a manufacturing pro-
cess and to ensure that it produces mechanisms that function as intended
despite kinematic variation.
Variation in mechanisms is represented with tolerances. We use paramet-
ric tolerances because they are general and conform with tolerancing prac-
tice. The variation in manufacturing is modeled with error intervals around
the nominal parameter values of the mechanism’s parametric model. The
model is analyzed to estimate the kinematic variation. When the variation
is excessive, the mechanism is redesigned, as discussed in the next chapter,
or a more accurate manufacturing process is selected.
We present algorithms that derive the worst-case kinematic variation of
mechanisms with parametric tolerances. We describe the tolerance specifi-
cations in section 6.1, kinematic variation in section 6.2, and algorithms
for kinematic pairs and mechanisms in sections 6.3 and 6.4.
6.1 Specifications of Parametric Tolerance
We saw in section 2.1 that a parametric model specifies parts in terms of

parameters p ¼ ð p1 ; . . . ; pn Þ. A boundary segment is specified with an im-
plicit curve, f ðx; y; pÞ ¼ 0, that depends on p. The model contains parame-
ters for all the part boundaries and motion axes of the mechanism. We
specify nominal parameter values, p ¼ ð p1 ; . . . ; pn Þ, and lower and upper
parameter bounds, l ¼ ðl1 ; . . . ; ln Þ and u ¼ ðu1 ; . . . ; un Þ. The tolerance inter-
vals, pi li a pi a pi þ ui , model the variation of the manufacturing
process.
100 6 Tolerancing
Figure 6.1
(a) Lever mechanism and (b) kinematic variation.
Figure 6.1a shows a parametric model of a lever mechanism composed of

an eccentric rotating cam, a translating follower, and a rotating lever. The
parameters are p ¼ ðr; h; m; w; vÞ with r and h the cam’s radius and eccen-
tricity, m the follower’s axis height, and w and v the axis positions of the
horizontal and vertical levers. The tolerance specifications are p ¼ ð1; 0:25;
2; 2; 3Þ and li ; ui ¼ 0:01.
The cross-product of the tolerance intervals, a hyperbox in the parameter
space, is the tolerance space. Each point in the tolerance space is an in-
stance of the nominal design. We assume that the shapes of the parts have
the nominal topology in every instance; incident segments have equal end-
points and nonincident segments are disjoint.
6.2 Tolerance Analysis
The task of worst-case tolerance analysis is to compute the maximal varia-

tion from the nominal kinematic function over the tolerance space. We il-
lustrate this using the lever mechanism, which has fixed contacts, then
define kinematic variation for general mechanisms.
The driving motion of the lever mechanism is rotation of the cam. The
cam raises and lowers the follower, which raises and lowers the lever. The
kinematic function is the relation between the cam angle, yC , and the lever
angle, yL . Figure 6.1b shows the nominal function (thick curve) and its vari-
ation (shaded area). The variation is maximal when yC ¼ 0 and yC ¼ 180
and is minimal when yC ¼ 90 and yC ¼ 270 . The boundary curves of the
variation region represent the extremal values of yL at each yC value.
6.2 Tolerance Analysis 101
Figure 6.2
Circle-box pair: (a–c) nominal shapes and two valid instances in configuration a ¼
ð4; 5Þ, (d–f) nominal and instance partitions of configuration space.
The kinematic function of a mechanism is its partition of configuration

space. Each instance in a tolerance space has its own partition, owing to
its the shapes and motion axes of its parts. The variation in the partition
represents the kinematic variation of the mechanism. The instance contact
spaces form a band around the nominal contact space, called the contact
zone, that comprises the configurations where some instance is in contact.
The nominal free space minus the contact zone comprises the configura-
tions that are free for every instance; likewise for contact space.
We illustrate contact zones with a pair (figure 6.2a) composed of a circle
A of nominal radius rA ¼ 1 and a box B with a square outer profile of nom-
inal length lB ¼ 8, a circular hole at its center of nominal radius rB ¼ 2, and
a vertical slot of width wB ¼ 2:5. The clearance between circle A and the B
slot is 0.5. The parameter bounds are 0.1 for rA , rB and 0:2 for lB , wB . Figure
6.2b shows an instance with rA ¼ 1:1, lB ¼ 8:2, rB ¼ 2:1, wB ¼ 2:5, and clear-
ance 0.3. Figure 6.2c shows an instance with rA ¼ 0:9, lB ¼ 7:8, rB ¼ 1:9,
wB ¼ 2:7, and clearance 0.9.
102 6 Tolerancing
Figure 6.3
Circle-box pair contact zone (a) with detail (b).
Figure 6.4
Circle-box pair with larger variation: (a) blocked instance, (b) configuration space
partition, (c) contact zone.
The configuration space of the pair is ðxA ; yA Þ because A translates with

yA ¼ 0 and B is fixed. Figure 6.2d–f shows the nominal partition and the
two variations. Figure 6.3 shows the contact zone: the dark gray band
around the nominal contact space. Configuration a ¼ ð4; 5Þ is free for every
instance, whereas b ¼ ð7; 9:1Þ is free for the nominal shape (figure 6.2a) and
is blocked for the first valid instance (figure 6.2b).
A difference between the nominal partition topology and that of an
instance indicates a qualitative change in kinematic function. In our exam-
ple, the topology changes when the inner and outer contact zone bounda-
ries intersect (figure 6.4c). This occurs when the bounds are 0:15 for rA , rB
and 0:3 for lB , wB . There are instances in which A does not fit in the vertical
slot, as illustrated in figure 6.4a–b with rA ¼ 1:1, lB ¼ 8:2, rB ¼ 2:1, wB ¼ 2:2.

The narrow vertical channel disappears, indicating that the circle cannot
exit the box.
In our example, the outer and inner contact zone boundaries correspond
to the maximum and minimum material condition instances. This is not
the case in general. Each boundary point is generated by an instance, but
there need not exist one instance that generates the boundary. The contact
zone is thus a conservative estimate of worst-case variation. In particular, a
contact zone boundary intersection does not guarantee that any instance
exhibits a qualitative change in kinematic function.
Disk Indexer Pair We illustrate a change in contact zone topology using

the disk indexer pair from section 5.1.1 (figure 6.5). Figure 6.5a–c shows
that a narrow contact zone has the topology of the nominal partition: the
diagonal segments represent drive periods and the horizontal segments rep-
resent dwell periods. Figure 6.5d–f shows that a wider contact zone has a
different topology in which the diagonal channels are replaced by lower
and upper horns. The topology changes indicate that instances in the toler-
ance space might block, and the displayed instance shows that blocking
can occur. Although the horizontal channels of the failure instance are
blocked, the diagonal channels are open, so this aspect of the contact
zone is conservative with regard to this instance.
Intermittent Gear Mechanism We illustrate general planar pair contact

zones using the gear-follower pair of the intermittent gear mechanism
from section 5.1.3 (figure 6.6). The parameters that specify the gear teeth
and the follower’s pawls are assigned tolerances of G0:01 mm. The contact
zone has the nominal topology at gear orientations where a pawl drives a
gear tooth, as shown in figure 6.6a and c. The topology is different when
neither pawl is engaged (figure 6.6b): the nominal free region splits into
two regions, which indicates that the follower blocks when the pawls en-
gage opposing teeth.
6.3 Algorithms for Kinematic Pairs
Tolerance analysis of a kinematic pair consists of computing the contact

zone. We employ boundary representations. A fixed-axis pair has a two-
dimensional zone whose boundary consists of curve segments. A general
planar pair has a three-dimensional zone whose boundary consists of sur-
face patches. The boundaries are not expressible in closed form, except for
very simple shapes of parts, so we compute approximate contact zones.
104 6 Tolerancing
Figure 6.5
Disk indexer pair: (a) nominal design, (b) nominal partition, (c) narrow contact zone,
(d) failure instance, (e) failure partition, (f) wider contact zone.
Figure 6.6
Gear-follower contact zone cross-sections: (a) yc ¼ 0 , (b) yc ¼ 5 , (c) yc ¼ 9 .
Figure 6.7 summarizes the algorithm. Steps 1 and 2 sample the contact
space. For a fixed-axis pair, the contact curves are sampled as described
in section 4.3.1. For a general planar pair, cross-sections are formed at a
specified yA spacing and the resulting fixed-axis pairs are sampled. Step 3
computes the contact zone boundary points in the outward and inward
normal directions for each sample of contact space. Step 4 forms the con-
tact zone boundary from the boundary points. Figure 6.8 illustrates the al-
gorithm for the figure 6.3 contact zone.
Step 3 The input is a configuration, c, on a parametric contact curve or

surface, f ðc; pÞ ¼ 0. The outward normal at c is the gradient vector, fc . The
contact zone’s outer boundary point lies on the ray c þ kfc and maximizes k
106 6 Tolerancing
Figure 6.7
Algorithm for the computation of the contact zone.
Figure 6.8
Contact zone computation: (a) contact space sampling, (b) contact zone boundary
points at a and b, (c) contact zone boundaries.
under the constraints f ðc þ kfc ; p þ qÞ ¼ 0 and li a qi a ui . The inner

boundary point minimizes k. We approximate f with its linear Taylor series
f ðc þ kfc ; p þ qÞ A f ðc; pÞ þ kfc fc þ q fp ¼ kfc fc þ q fp
and solve f ¼ 0 for
1 1 X qf
k¼ q fp ¼ qi :
fc f c fc f c i qpi
This formula shows that k is maximal when qi ¼ ui for qf =qpi > 0 and
qi ¼ li otherwise. It is minimal when li and ui are interchanged.
We illustrate the computation using the circle-box pair (figure 6.8b). The
sole parameter is p ¼ ra with p ¼ 1, l ¼ 0:2, and u ¼ 0:2. Configuration a ¼
ð9; 6Þ lies on the moving circle-fixed line contact curve f ðxa ; ya ; ra Þ ¼ xa
ra 8 ¼ 0, using equation 3.7 with nb ¼ ð1; 0Þ, oaA ¼ ð0; 0Þ, and db ¼ 8. The
gradient is ðqf =qxa ; qf =qya Þ ¼ ð1; 0Þ and k ¼ qðqf =qra Þ ¼ q with 0:2 a q a
0:2. The maximum, k ¼ 0:2, occurs when q is largest and generates the con-
6.4 Algorithms for Mechanisms 107
tact zone’s outer boundary point ð9:2; 6Þ. The minimum, k ¼ 0:2, occurs
when q is smallest and generates the inner boundary point ð8:8; 6Þ. Con-
figuration b ¼ ð4; 3Þ lies on the moving circle-fixed circle contact curve
ð2 ra Þ 2 ðxa 4Þ 2 ðya 4Þ 2 ¼ 0, using equation 3.8 with oaA ¼ ð0; 0Þ,
obB ¼ ð4; 4Þ, and rb ¼ 2. The gradient is 2ð4 xa ; 4 ya Þ ¼ ð0; 2Þ and
k ¼ 0:2q. The maximum and minimum generate the outer and inner
boundary points ð4; 2:8Þ and ð4; 3:2Þ.
The error that results from the linear Taylor series is quadratic in the tol-
erance interval widths. Usually this error is negligible because the tolerance
intervals are narrow, but there are cases where the error is significant. We
can derive the true maximum and minimum via constrained nonlinear
optimization, which takes longer than linear optimization, but is still
practical.
Step 4 In a fixed-axis pair, each free region boundary yields an outer and
an inner sequence of contact zone boundary points. We link the points in
each sequence to obtain curves (figure 6.8c). These curves partition the
configuration space into faces. The faces that contain the contact space are
the contact zone. When the tolerance intervals are narrow, the contact
zone is a narrow band around the contact curves (figure 6.3). When the
intervals are wide enough, the curves intersect and the contact zone topol-
ogy is more complicated (figure 6.4). In a general planar pair, the approxi-
mate contact zone boundaries are triangular meshes that generate a spatial
partition.
6.4 Algorithms for Mechanisms
The computation algorithm for a contact zone cannot be used for mecha-
nisms because we cannot in general construct mechanism partitions. If we
could, the algorithm would extend directly. Instead we compute error
bounds at sample configurations on the nominal path for a given driving
motion, which we compute by kinematic simulation (section 5.2). A con-
figuration, c, lies on a set of contact surfaces, f j ðc þ d; p þ qÞ ¼ 0, that lin-
j j
earize to fc d þ fp q ¼ 0. We seek a maximal d subject to the linearized f j
constraints and li a qi a ui . The d metric is task specific. A common crite-
rion is that an output configuration variable be maximized with the driving
variables fixed at their nominal values.
We illustrate using a lever mechanism (figure 6.1). The configuration
space coordinates are ðyC ; yF ; yL Þ. The tolerance parameter vector is p ¼ ðr;
h; m; w; vÞ with p ¼ ð1; 0:25; 2; 2; 3Þ and with li ; ui ¼ 0:01. The cam-follower
108 6 Tolerancing
contact function is yF p1 p2 cos yC ¼ 0 and the follower-lever contact

function is p4 sin yL þ p5 p3 yF ¼ 0. We compute the variation in yL .
Eliminating yF yields p4 sin yL þ p5 p3 p1 p2 cos yC ¼ 0 and lineariz-
ing with yC fixed yields
d3 p4 cos yL ¼ q5 þ q3 þ q1 þ q2 cos yC q4 sin yL :
The d3 maximum is ð0:03 þ 0:01jcos yC j þ 0:01jsin yL jÞ=cos yL and the min-

imum is its negative. The maximum variation, d3 ¼ 2:4 , occurs at yC ¼ 0
and yC ¼ 180 where yL is extremal. The minimum variation, d3 ¼ 1:7 ,
occurs at yC ¼ 90 and yC ¼ 270 where yL ¼ 0.
6.5 Notes
Tolerancing of mechanisms is of great practical and economic importance.

Worst-case tolerancing sets the stage for statistical tolerancing in which
manufacturing variation is modeled with probability distributions and ki-
nematic variation is treated as a random variable. A manufacturing process
is accepted based on the probability that its output will function correctly
or on its expected cost.
Current tolerancing principles and standards are described by the Ameri-
can Society of Mechanical Engineers [1, 2]. Popular tolerance specifications
are parametric and geometric [66, 80]. A variety of methods for worst-
case and statistical tolerancing of mechanisms have been proposed [9]. For
a survey of commercial computer-aided tolerancing (CAT) packages, see
Prisco [63]. Tolerancing for assembly, whose task is to ensure that the man-
ufactured parts will assemble, is a closely related topic [82].
The most common kinematic tolerance analysis is tolerance chain or
stack-up analysis for static and small displacements. It consists of identify-
ing critical parameters, such as a gap, clearance, or play; building a one-
dimensional tolerance chain based on configurations and contacts of parts;
and determining the parameters’ variability range [8, 17, 81]. Kinematic
variations for large displacements are studied in Chase et al. [13].
We have developed algorithms for kinematic tolerancing in the configu-
ration space paradigm [41, 72, 73, 75]. We automate computation of kine-
matic variation with contact changes. The algorithms use the linear Taylor
series approximation described earlier. In later work [44] we developed an
exact algorithm based on constrained optimization. The linear approxi-
mation is normally accurate, but is sometimes very inaccurate at critical
configurations.
Appendix B describes HIPAIR computation of contact zones. HIPAIR han-
dles fixed-axis planar pairs with uniform offset tolerances.
7 Synthesis
This chapter studies the design task of kinematic synthesis. The goal is to
design a mechanism that realizes a kinematic function. Synthesis is the in-
verse process of analysis. It is inherently more difficult because many mech-
anisms can realize the same kinematic function. We focus on parameter
synthesis: given a parametric model of a mechanism, a kinematic function,
and its allowed variability, the goal is to find nominal parameter values and
tolerance intervals that realize the kinematic function with the allowed
variability.
In section 7.1 we review the kinematic design cycle before focusing on
parameter synthesis. In section 7.2 we present a parameter synthesis algo-
rithm that finds parameter values that realize a kinematic function. In
section 7.3, we describe a parameter optimization algorithm that adjusts
parameter values to ensure correct kinematic function for a given tolerance
space. We discuss mechanisms composed of fixed-axis higher pairs.
7.1 Kinematic Design Cycle
We saw in chapter 1 that kinematic design is an iterative process with con-

ceptual design, parametric design, analysis, synthesis, and tolerancing steps.
Conceptual design consists of selecting a design concept that captures the
desired kinematic function. Parametric design consists of building a para-
metric model of the mechanism that encodes the shapes and configura-
tions of parts. Analysis derives the kinematic function of a mechanism
from a specification of its parts’ shapes and motion constraints. Synthesis
is the inverse task of devising a mechanism that performs a specified func-
tion. Tolerancing derives the kinematic effect of manufacturing variation.
We illustrate this cycle with a driver-follower pair from an optical filter
mechanism, which is discussed further in section 8.1. The design require-
ments are to cover and uncover a lens with a filter (figure 7.1a). Figure
110 7 Synthesis
Figure 7.1
Kinematic design scenario: (a) design requirements, (b) nominal design, (c) intended
kinematic function, (d) nominal partition, (e) contact zone detail, (f) failure instance,
(g) failure partition.
7.1c shows the intended kinematic function. Counterclockwise rotation of

the driver turns the follower clockwise, so the filter covers the lens, then
locks it in place. Clockwise rotation unlocks the follower, rotates it until
the filter uncovers the lens, and re-locks it.
During conceptual design, the designer chooses a Geneva pair. During
parametric design, the geometry and rotation axes for the parts are chosen
(figure 7.1b). The intended feature contacts are a pin and slot in the driving
phase and concentric convex and concave arcs in the locking phase. The
designer creates a parametric model and assigns parameter values and toler-
ance intervals. The nominal function is correct (figure 7.1d), but the con-
tact zone (figure 7.1e) indicates possible blocking, owing to overlap
between the upper and lower boundaries. Figure 7.1f–g shows a blocking
instance and its partition. The designer can change the parameter values
to remove the blocking, can tighten the tolerance intervals, or can change
the design concept.
7.2 Parameter Synthesis
The input to our parameter synthesis algorithm is a parametric model of a

mechanism with parameter intervals. The cross-product of the intervals is
the design space. Each point in the design space is a design instance. We
seek a design instance that realizes a specified kinematic function in an op-
timal manner.
A manual search of the design space is often impractical. The designer
must examine many instances to ensure that a good design has not been
overlooked. Each instance requires a time-consuming analysis. The search
is especially difficult when the kinematic function is sensitive to small per-
turbations in the parameter values, as is common with higher pairs. Unlike
tolerance spaces, design spaces are generally too large for linearization of
the contact curves to have acceptable accuracy.
Our solution is an optimization search. We encode the design goals in
objective functions that we minimize subject to the kinematic constraints.
The design goals usually cannot be expressed with a single function be-
cause they are often incommensurate or conflicting. The entire design
space cannot be searched because of its large dimension. These considera-
tions lead us to an interactive paradigm in which the designer refines an
initial design instance through a sequence of design changes.
Figure 7.2 summarizes the algorithm. The designer examines the con-
figuration space partitions for design flaws, decides how to change the
112 7 Synthesis
Figure 7.2
Algorithm for parameter synthesis.
partitions to remove the flaws, and inputs the changes. The algorithm
translates the changes into an objective function and computes a local
minimum starting from the current parameter values. The cycle repeats
until the design is acceptable.
7.2.1 Design Change

A design change is specified with a dragger in configuration space: an arrow
whose tail, t, lies on a contact curve with equation f ðc; pÞ ¼ 0 and whose
head, h, is a configuration that should lie on that curve. The tail satisfies
f ðt; pi Þ ¼ 0 with pi the initial design instance. The objective function is
jf ðh; pÞj with parameters p.
We illustrate draggers with the circle-box pair (figure 7.3). The initial de-
sign instance (figure 7.3a) and configuration space partition (figure 7.3b)
are from section 6.2. The dragger (figure 7.3c) has a tail t ¼ ð3:75; 5:75Þ
and head h ¼ ð3:9; 5:75Þ. The contact is between A and the left side of the
B slot. Its equation is f ðc; pÞ ¼ xA rA þ 0:5wB 4 ¼ 0 with c ¼ ðxA ; yA Þ and
p ¼ ðrA ; wB Þ. The initial design instance is pi ¼ ð1; 2:5Þ. The objective func-
tion, jf ðh; pÞj ¼ j3:9 rA þ 0:5wB 4j, is obtained by setting c ¼ h in the
contact equation. Its value at p ¼ pi is j3:9 1 þ 1:25 4j ¼ 0:15.
We minimize the objective function by iterative optimization starting
from p ¼ pi . The new parameter values achieve the requested design
change when the minimum is zero and partially achieve the change when
it is positive. Our example design change can be achieved by increasing the
circle radius, rA , from 1 to 1:15 (figure 7.3d) or by decreasing the slot width,
wB , from 2:5 to 2:2 (figure 7.3e). Increasing rA changes the entire contact
space because the A circle is involved in every contact, whereas decreasing
wB only changes the channel width.
Figure 7.3
Circle-box pair: (a) initial design instance, (b) initial configuration space partition, (c)
dragger, (d) rA ¼ 1:15 modification, (e) wB ¼ 2:2 modification.
7.2.2 Structure Change

Minimization of the dragger can change the free space structure. The struc-
ture is the number of free faces, the number of edge loops that bound
each face, and the sequence of contact curves along each loop. A contact
curve that enters or leaves a loop is a local change since the kinematic func-
tion is insensitive to which features generate it. A change in the free space
topology implies a large change in kinematic function, which is often
undesirable.
We illustrate these concepts with a driver-follower pair (figure 7.4). The
initial partition of the configuration space shows a correct function (figure
7.4b). In the diagonal channel, the driver pin engages the follower slot and
rotates the follower. In the horizontal channels, the driver arc aligns with
the complementary follower arc and prevents rotation of the follower. The
partition reveals excessive follower play in the driving mode. The play
appears as the distance between the top and bottom contact curves that
bound the channel (figure 7.4c). The designer requests smaller play with
an upward dragger on the bottom curve. A local change in structure occurs
114 7 Synthesis
Figure 7.4
Driver-follower pair: (a) detail of initial instance, (b) partition, (c) detail with dragger,
(d) local change, (e–f) topology change with detail.
partway through the optimization (figure 7.4d). The contact curve, c, of the
driver arc-follower arc tip crosses the pin-slot curve, a, and enters contact
space. A topology change occurs when c hits the upper channel boundary,
b, and blocks the channel (figure 7.4e–f).
The parametric synthesis algorithm prevents topology changes. The new
free space is matched against the old one. If the structures match, the new
design instance is accepted. Otherwise the algorithm searches the line seg-
ment in parameter space between the old and the new instances for the
first point, pc , where the structure changes. If the change is local, the opti-
mization resumes from pc . In a topology change, two contact curves (on
the same loop or on different loops) are disjoint on one side of pc , tangent
at pc , and intersect on the other side of pc . The algorithm adds penalty
terms to the objective function to prevent the tangency and restarts the
optimizer at the old instance.
In our example, the tangent curves are a and c (figure 7.5a). One penalty
term prevents a from crossing a barrier point below the tangency and the
other prevents c from crossing a barrier above it. The barriers appear as
a single filled circle because they are so close together (figure 7.5b). The
Figure 7.5
Topology change (a), barriers (b), and final partition (c).
revised objective is minimized and the design goal is achieved without

changes in topology (figure 7.5c). The original channel (figure 7.4c) is
drawn with dashed lines for comparison.
It remains to describe the matching algorithm. Two edges match if their
contact curves are formed by the same pair of part features. For edge loops
e1 ; . . . ; en and f1 ; . . . ; fn to match, they must have the same length and e1
must equal fi for some i. The match succeeds when e2 ¼ fiþ1 ; e3 ¼ fiþ2 ; . . . ;
en ¼ fiþn with the indices’ modulo n. Partitions A and B match when they
have the same number of free face edge loops and every A loop matches a
single B loop. We can match the partitions by matching every A loop with
every B loop. This naive algorithm, which takes Oðm 2 Þ time with m the
total number of edges, can be slow when m is large since it is used in the
inner loop of the optimizer. A constant-time algorithm is cited in the notes.
7.2.3 Multiple Draggers

A design change can be specified with multiple draggers in multiple parti-
tions. The objective function is the sum of one term, jfi ðhi ; pÞj, per dragger.
The algorithm minimizes the objective as before while checking every pair
for structure changes. We illustrate this using a mechanism with design
flaws in two pairs, which are corrected with two draggers in one partition
and one in the other.
The mechanism is the Dennis clutch from section 2.4 (figure 7.6). Ini-
tially, the pawl is disengaged from the ratchet and is locked by the arm
(figure 7.6a). The operator releases the pawl by rotating the arm counter-
clockwise (figure 7.6b). The spring rotates the pawl clockwise until its tip
engages a ratchet slot. The ratchet, which is mounted on a rotating shaft,
drives the cam via the pawl (figure 7.6c). The rotating cam pin hits the
upper finger of the arm (figure 7.6d) and rotates it clockwise. The lower
116 7 Synthesis
Figure 7.6
Dennis clutch mechanism: (a–f) configurations a–f.
finger of the arm hits the pawl (figure 7.6e), disengages it from the ratchet,
and locks it (figure 7.6f). The cam rotates to its initial angle owing to inertia
(figure 7.6a).
The designer achieves the intended function by parameter synthesis. The
first change in design fixes a cam-arm design flaw: when the operator
rotates the arm, it hits the cam arm before it can release the pawl. Partition
of the cam-arm configuration space shows the design flaw (figure 7.7a). The
initial configuration is p and the dashed line marks the angle at which the
arm releases the pawl. The intended path for the motion is the vertical
from p to the release line, but the path ends prematurely at q when the
arm hits the cam’s rim. The designer raises the contact curve with two drag-
gers (figure 7.7b). The revised partition shows a correct function (figure
7.7c). The arm rotates from p to q, where it releases the pawl. The cam
rotates from q to r, where its pin hits the upper finger of the arm. The cam
rotates the arm from r to s, where the pin passes under the finger. The
cam rotates to p to end the work cycle.
The second design change fixes a ratchet-pawl design flaw: when the
pawl spring is too weak, the pawl tip can slide over a ratchet slot without
engaging. A ratchet-pawl partition shows the design flaw (figure 7.8a). The
intended motion path is from p, where the pawl tip reaches the edge of a
Figure 7.7
Cam-arm: (a) faulty partition, (b) synthesis, (c) correct partition.
Figure 7.8
Ratchet-pawl: (a) faulty partition, (b) synthesis, (c) correct partition.
ratchet slot, to q, where it hits the slot side, to r, where it engages. The
actual path is governed by the pawl’s spring and the parts’ inertias. If the
spring is too weak, the pawl tip can pass the ratchet slot and hit the top of
the next tooth at s. The designer lowers the right contact curve with a drag-
ger (figure 7.8b). In the revised partition (figure 7.8c), the pawl tip hits the
slot’s side and engages independently of the spring’s strength.
7.3 Parameter Optimization
The second synthesis algorithm is parameter optimization (figure 7.9). The

input is a parametric model and a tolerance space. The goal is to ensure
that every design instance has a correct kinematic function. An instance is
deemed correct when its configuration space partitions match the nominal
ones, as defined in section 7.2.2, and its motion path is within a specified
distance from the nominal one. The algorithm searches the tolerance space
for incorrect instances and then revises the nominal parameter values to
exclude them.
118 7 Synthesis
Figure 7.9
Algorithm for parameter optimization.
Figure 7.10
Circle-box pair: (a) nominal, maximal instances for (b) left, (c) right, and (d) both
channel sides.
The algorithm does not optimize tolerances. We could tighten tolerances

to exclude incorrect instances that cannot be excluded by changing the
nominal parameter values. Tighter intervals are an inferior solution because
they often increase manufacturing cost. We could also loosen tolerances
to reduce cost. Optimization of tolerance requires a design quality metric,
which is a separate research topic.
Maximal Instances We examine the instances that maximize the variation

of one or more contacts. Variation in a contact represents a maximal varia-
tion in its kinematic function. Variation in two contacts represents a maxi-
mal interaction between them. Incorrect function is most likely to arise in
such instances.
We illustrate maximal variation with a circle-box pair (figure 7.10). The
parameters are p ¼ ða; bÞ with nominal values p ¼ ð2:75; 2:75Þ. The circle’s
radius is 1 and the box length is 8. A maximal instance for the circle-left
channel contact (figure 7.10b) is ð3:25; 2:75Þ. A maximal instance for the
circle-right channel contact (figure 7.10c) is ð2:75; 3:25Þ. Both instances are
correct in that the circle can pass through the channel (figure 7.10d). The
instance ð3:25; 2:25Þ is maximal for both contacts and is incorrect.
Figure 7.11
Parameter revision: (a) closest correct instances, (b) revision of one instance, and re-
vision of two instances using one (c) and two (d) parameters.
The maximal instances are computed in two steps. The first step collects
the parameter values that maximize the variation in the individual con-
tacts. These are the values that generate the contact zone boundary points
in step 3 of section 6.3. Each boundary point is determined by the values of
the parameters in its contact equation. These values define a partial design
instance. The other design parameters are called free. In our example, the
partial instances are ð3:25; f Þ and ðf ; 3:25Þ for the left and right channels,
with f denoting a free parameter.
The second step forms the instances that maximize the variation in mul-
tiple contacts. Partial instances a and b are compatible when for every i ei-
ther ai is free, bi is free, or ai ¼ bi . They can be merged into a new partial
instance whose ith element is free when ai and bi are free, equals ai when
bi is free, and equals bi otherwise. The merge assigns each partial instance
its maximal parameter values; hence it maximizes the variation in both
contacts. We merge all compatible sets of partial instances. We obtain the
maximal instances by assigning the free parameters of the merges their
nominal values. The two partial instances in our example are compatible
and their merge, ð3:25; 3:25Þ, is the joint maximum.
Revision The tolerance space is revised to exclude the incorrect maximal

instances. Each incorrect instance is surrounded by a region of similar in-
stances because its contact curves are continuous in the parameters. We
would like to exclude the entire region, but computing it is impractical. In-
stead, we exclude the closest incorrect instance on the line segment from
the nominal design to the incorrect instance, which we find by a bisection
search. Figure 7.11a shows incorrect instances c and d with closest in-
stances a and b; the incorrect regions are shaded.
120 7 Synthesis
We would like to compute a minimal revision of the nominal design that

excludes the closest incorrect instances from the tolerance space. This task
is computationally intractable, so we employ a heuristic. The minimal revi-
sion for a single instance, a, is computed as follows (figure 7.11b). Let d be
the perpendicular distance from a to a tolerance space face with inward
normal n. Incrementing the nominal parameter values by v ¼ dn excludes
a from the tolerance space. The face that minimizes d yields the optimal re-
vision (the dashed box).
We exclude multiple instances by combining their minimal revisions.
Revisions v and w can be combined when vi and wi never have opposite
signs. The ith element of the combination is maxðvi ; wi Þ when vi ; wi b 0
and is minðvi ; wi Þ otherwise. Figure 7.11c–d show combinations that revise
one and two parameters.
7.4 Notes
Kinematic synthesis has been the subject of intense study for the past four
decades. Most research has focused on linkages and other permanent con-
tact mechanisms [25, 27, 57]. For a survey of earlier work, see Erdman [21];
a more recent survey is presented in McCarthy and Joskowicz [58]. Special-
ized techniques for synthesis of cam and gear mechanisms are described in
Gonzales-Palacios and Angeles [26] and Litvin [51]. Software for synthesis
of kinematic linkage includes Erdman’s LINCAGES-2000, KINSYN, SyMech,
and the ADAMS/View packages, among others.
Kinematic synthesis has also been studied in the broader context of
mechanism design. Formal theories of mechanism design, including con-
ceptual and configuration design, address these issues [29, 43]. Parameter
synthesis as an optimization problem has gained much acceptance as an
approach for optimal design [61] and has been applied to optimization of
cam mechanisms [4].
The configuration space paradigm has been used for the synthesis of
higher-pair mechanisms [12, 34, 49, 64]. Our work addresses parameter
synthesis [45] and robust synthesis [46]. The constant-time partition-
matching algorithm appears in Kyung and Sacks [45]. Robust synthesis
couples nominal parameter and tolerance synthesis so that nominal and
tolerance changes are evaluated together [76]. The nominal design is modi-
fied to reduce its sensitivity to variations of parts. Tolerances are then allo-
cated to guarantee correct function and minimize cost.
The algorithms in this chapter extend in principle to general planar pairs,
but they have not been implemented because of the complexity of manip-
ulating three-dimensional configuration space.
8 Case Studies
This chapter presents four case studies of actual mechanisms taken from
industry. They illustrate the use of the configuration space method in the
analysis and design of mechanisms. Section 8.1 describes the kinematic
synthesis of an optical filter mechanism. Section 8.2. describes the tolerance
analysis of a selector mechanism for an automotive gearshift. Section 8.3
describes the redesign of a torsional ratcheting MEMS device. Section 8.4
describes the redesign of a spatial asynchronous reverse gear pair from an
automotive transmission.
8.1 Optical Filter Mechanism
The optical filter mechanism [76] is from Israel Aircraft Industries and was
introduced in section 7.1. The mechanism consists of a lens, a driver, and
three filters mounted on identical followers (figure 8.1). The lens is attached
to a fixed frame (not shown). The followers are stacked on a shaft and ro-
tate independently. The driver consists of three slices that rotate together
on a common shaft. Each driver slice drives the corresponding follower.
The figure shows details of the top and middle driver slices and followers.
Each slice consists of a driving pin and a locking arc. The second and third
slices are rotated by 90 and 180 relative to the first slice. In the initial
state, the filters are off the lens. When the driver shaft is rotated counter-
clockwise, the three followers are engaged in sequence. Each driver pin
engages its follower slot and rotates the follower until the filter covers the
lens. Rotating the driver clockwise resets the filters to the initial state.
The initial design scenario is discussed in section 7.1. The design task is
to devise a mechanism to engage and reset the followers in the intended
manner. The mechanism must be robust and compact because it will
be mounted on a vehicle. The design concept is a Geneva mechanism
with one driver and one follower per filter. The parametric model has 25
122 8 Case Studies
Figure 8.1
(a) Optical filter mechanism and (b–c) details of two driver-follower pairs.
Figure 8.2
(a) Driver-follower configuration space partition with details of (b) original and (c)
tightened contact zones.
functional parameters, including the centers of rotation, the pin and lock-
ing arc radii, and the slot dimensions.
We assign nominal parameter values that produce correct function using
the synthesis algorithm from section 7.2. We assign tolerances to the
parameters and assess their effects. Figure 8.2 shows a detail of the driver-
follower contact zone in the area where the driver unlocks the follower
and the pin enters the follower slot. The width of the contact zone varies
with the sensitivity of the nominal contact configuration to the tolerance
parameters. The upper and lower zones of the diagonal channel intersect,
which suggests that the tolerance space contains design instances that
block.
8.2 Manual Transmission Gearshift 123
Figure 8.3
Gearshift mechanism.
We remove the blocking with the parameter optimization algorithm

from section 7.3. The first iteration finds 146 blocking instances out of
700 candidates. After 19 iterations, the algorithm finds parameter values
with no failures. If parameter optimization were to fail, we could tighten
the tolerance intervals until the zones become disjoint, as shown in figure
8.2c.
8.2 Manual Transmission Gearshift
The gearshift selector mechanism [74] is from an an automotive automatic

transmission from Ford Werke, Germany. The design task is to ensure cor-
rect function with manufacturing variation. Figure 8.3 shows the four main
parts: a rotating cam, a translating piston, a rotating spring-loaded pin, and
a fixed valve body. The cam pin fits in a slot in the left end of the piston, so
cam rotation causes piston translation. The cam has seven slots that repre-
sent the seven gear settings: 1, 2, 3, D, N, R, and P (not shown). The auto-
mobile driver selects a setting by rotating the cam with the gearshift (not
shown) to place the spring-loaded pin in the appropriate slot. The pin
then locks the cam, which locks the piston. For each gear setting, the pis-
ton body opens and closes the appropriate conducts in the valve body.
Variations in the pin, piston, and cam shapes and configurations affect
the piston’s displacement and thus the valve opening.
We construct a tolerance model by adding variation parameters to the
functional features of the parts. For the cam, we tolerance the segments
that form the slots and the pin that engages the piston. For the piston, we
tolerance the two vertical line segments that are in contact with the cam
124 8 Case Studies
Figure 8.4
Cam-piston (a) configuration space partition and (b) detail of contact zone.
pin. For the pin, we tolerance the circle. Line segments are toleranced by
varying the coordinates of the two endpoints; circle segments are toler-
anced by varying the radius and the center coordinates. To account for
variability in the position of the rotation axes, we also tolerance the centers
of rotation of the cam and the pin. Since we chose the piston as the refer-
ence part of the assembly, there was no need to tolerance the orientation of
its translation axis. The model has 86 tolerance parameters for the cam, 8
for the piston, 5 for the pin, and 99 overall. We assign every parameter an
independent tolerance of G0:1 mm.
Figures 8.4a and 8.5a show the configuration space partitions of the cam-
piston and cam-pin pairs. The degrees of freedom are the cam orientation,
yA , the piston offset along its axis of motion, xB , and the pin orientation,
yC . The cam-piston free space is a narrow diagonal channel whose top and
bottom boundaries represent contacts between the cam pin and the left
and right vertical line segments of the piston slot. The channel width
quantifies the functional play. The kinematic relation is nearly linear. The
cam-piston contact space consists of six ‘‘valleys’’ where the pin is in the
cam slots, separated by ‘‘hills’’ where it switches slots.
We use tolerance analysis to determine the variation of the piston dis-
placement for each cam setting. Figures 8.4b and 8.5b show details of the
pair contact zones. The piston position, xB , has a worst-case variation of
0.41 mm–0.45 mm and the pin orientation, yC , has a worst-case variation
of 0:75 –1:03 .
Figure 8.5
Cam-pin (a) configuration space partition and (b) detail of contact zone.
As described in section 6.4, we examine the sensitivity to the individual

tolerance parameters at the nominal gear 3 setting yA ¼ 21 , yC ¼ 0 , xB ¼
33 mm. The worst-case variation of xB is 0.9 mm—roughly half from each
pair. The main factors in the cam-piston variation are the cam center’s hor-
izontal position (25%) and vertical position (25%), tooth base x position
(25%), and the x coordinates of the piston vertical segments (10%). The
cam-pin variation is evenly distributed among the parameters of the touch-
ing features and the parts’ centers of rotation.
We use parameter optimization to make the piston close the correct
valves despite variation in its locked positions. The first iteration finds 331
design instances that allow excessive variation out of 715 candidates. After
150 iterations, the algorithm finds parameter values with no failures.
8.3 Torsional Ratcheting Actuator
The torsional ratcheting actuator [70] is a MEMS device designed and fabri-
cated at Sandia National Laboratories Albuquerque, New Mexico. The
design task is to remove intermittent blocking failures. The mechanism
consists of a driver, a ratchet, a ring gear, and an antireverse (figure 8.6a).
The gear and the antireverse are mounted on the substrate with pin joints
and the ratchet is attached to the driver with a pin joint. The driver is
attached to the substrate by springs that allow planar rotation but prevent
translation. The driver is rotated 2:5 counterclockwise by an electrostatic
126 8 Case Studies
Figure 8.6
Torsional ratcheting actuator: (a) image courtesy of Sandia National Laboratories, (b)
detail of the CAD model.
comb drive. The ratchet engages the inner teeth of the gear and rotates it
counterclockwise. When the voltage drops, the springs restore the driver
to its start orientation, which disengages the ratchet. The antireverse pre-
vents the gear from rotating clockwise. Its outer teeth drive an external
load.
We analyze the mechanism using the configuration space method (figure
8.7). The gear-antireverse partition shows that the pair functions correctly.
The slanted contact curve that contains the displayed configuration is due
to the long side of a gear tooth and the top of the antireverse. This contact
causes the antireverse to follow the gear profile when the gear is driven.
The steep contact curve to the left of this curve is due to the short side of
a gear tooth and the side of the antireverse. This contact prevents the gear
from rotating clockwise since every direction in which yG decreases lies in
blocked space. The contact zone shows that both contact curves have the
correct slope for every design instance.
The gear-ratchet pair has a three-dimensional configuration space be-
cause the ratchet rotates around a point on the driver and the driver rotates
around a point on the frame. The contacts are invariant when the driver
and the ratchet are rotated by the same angle, so we analyze the gear-
ratchet pair in a two-dimensional space whose coordinates are the gear ori-
entation and the angle between the driver and the ratchet. Equivalently,
we fix the driver orientation at 0o , which fixes the ratchet’s center of rota-
tion. Figure 8.7e shows the configuration space partition. The near-vertical
Figure 8.7
Initial design: (a–c) details of gear-antireverse pair, partition, and contact zone, (d–f)
details of gear-ratchet pair, partition, and contact zone.
contact curve to the left of the displayed configuration is due to the short
side of a gear tooth and the ratchet tip. This contact causes the gear to ro-
tate with the driver. The contact curve to the right is due to the long side of
the gear tooth and the ratchet back. This contact allows the ratchet to ro-
tate clockwise and disengage the gear.
The contact zone (figure 8.7f) reveals a design flaw that causes intermit-
tent failures: the near-vertical contact curve can have a positive slope in
some design instances, which implies that the gear can rotate clockwise, es-
cape the ratchet, and jump to the next tooth. The failure is intermittent be-
cause friction and inertia normally prevent jumps. We fix the design flaw
with the parameter synthesis algorithm. In the revised design (figure 8.8),
the gear tooth and ratchet slopes are larger, which makes the slope of the
contact curve negative in every design instance.
128 8 Case Studies
Figure 8.8
Gear-ratchet redesign: (a) before tooth and configuration space partition (detail),
(b) draggers, (c) after.
8.4 Spatial Asynchronous Gear Pair
The spatial asynchronous reverse gear pair [75] is from an automotive

transmission from Getrag-Ford Transmissions, Germany. The design task is
to remove occasional gear blocking. The pair consists of two modified spur
gears, the idler and the reverse gear, mounted on parallel axes (figure 8.9).
The reverse gear rotates around its axis and the idler rotates around and
translates along its axis. The idler has 31 teeth and the reverse gear has 13
teeth. Their pitch diameters are 65.62 mm and 27.52 mm. Figure 8.9b
shows the geometry of the reverse gear tooth; the idler teeth have the
same geometry with different dimensions. The tooth has involute sides, A
and B, topped by spherical patches, C and D, that form a guiding chamfer.
The spherical patches meet along a circular arc, e, and meet the involutes A
and B along curves f and g. The tooth top is not functional.
The initial prototype usually functions correctly. When the automobile
driver shifts into reverse, the idler translates toward the reverse gear until a
pair of chamfers make contact (the configuration shown in figure 8.9a).
The idler translation and the chamfer contact cause the the gears to rotate
into alignment and to mesh. Figure 8.9b shows the contact point’s path on
the reverse gear tooth. The chamfer arcs e touch (point 1), then slide along
Figure 8.9
(a) Asynchronous gear pair and (b) geometry of reverse gear tooth.
130 8 Case Studies
Figure 8.10
(a) Faculty configuration space partition, and (b) cross-section at c ¼ 0 .
each other until the D patches touch (point 2). The D patch contact ends
when the involutes B mesh (point 3).
The designers observed occasional blocking in which the idler stops
translating before the gears mesh, owing to an interfering contact. Figure
8.9b also shows the path of the blocking contact point. The chamfer arcs e
touch (point 4) then slide along each other until the C patches touch
(point 5). The motion ends prematurely, owing to a contact between the
C patches of an adjacent pair of teeth (point 6).
We use the configuration space method to find the initial configurations
that lead to blocking and modify the design to eliminate blocking. The
degrees of freedom are the idler rotation y and translation z, and the reverse
gear rotation c. Hence, the free space is three-dimensional and the contact
space is two-dimensional. The contact space consists of two ‘‘hills’’ sepa-
rated by a sloping ‘‘valley’’ (figure 8.10). We have not discussed the con-
struction of three-dimensional partitions for spatial parts. The algorithm is
analogous to the one for fixed-axis pairs, with contact curves and planar
partitions replaced by contact surfaces and spatial partitions.
Figure 8.10b shows an engaging path and a blocking path in the c ¼ 0
cross-section. The driving motion is idler translation in the negative z direc-
tion. Initially, the idler is away from the reverse gear and the gear orienta-
tions y and c can take any value. The idler translates (z decreases) until
contact and then the gears rotate together. The idler engages if the path
reaches z ¼ 0 mm and blocks if the path ends in the valley. A path blocks
Figure 8.11
(a) Correct partition of configuration space (b) with cross-section.
based on its initial y and c values. Measuring the contact space valley
shows that 4% of paths block.
We eliminate blocking in two steps. A manual search of the design space
reveals a design instance with 0.5% blocking. The contact space peaks
move closer together, but the valley remains. The parameter synthesis algo-
rithm might find a better design and would certainly be faster, but it has
not been implemented for spatial parts. We eliminate blocking by remov-
ing every second idler guiding chamfer, which is a conceptual redesign.
Figure 8.11 shows the partition of the modified design with the original
dimensions of the parts. One peak has disappeared, so there is no valley
and thus no blocking.
9 Conclusion
We have presented the configuration space paradigm for the kinematic de-
sign of mechanisms. This paradigm supports robust and optimal kinematic
design of mechanisms by automating key tasks in the design cycle (figure
1.4). The main contribution of our work is support for higher-pair design.
Our research strategy is to identify important mechanism classes, to study
them in depth, and to develop efficient algorithms that support their de-
sign and analysis.
Configuration space is a complete, compact geometric representation of
the motion and interaction of parts. It provides a principled approach to
modeling the kinematics of mechanisms uniformly and systematically,
including planar and spatial mechanisms with lower and higher pairs, con-
tact changes, and topology changes. It covers nominal kinematics and ki-
nematic variation that is due to manufacturing variation.
Within the configuration space paradigm, we have developed an efficient
computational approach to the kinematics of contacts between parts. We
systematically derived explicit, closed-form algebraic contact models for
the most common planar feature contacts and outlined a general numerical
contact model. We developed computational geometry algorithms for the
construction of two- and three-dimensional configuration space partitions
to model the contacts in kinematic pairs. The algorithms construct kine-
matic models from parametric descriptions of the parts’ shapes and degrees
of freedom.
Based on the representation of configuration space, we developed effi-
cient algorithms to supports the core design tasks of analysis, tolerancing,
and synthesis. For analysis, we developed algorithms for kinematic and
dynamical simulation of mechanisms with changes in contact. For toler-
ancing, we developed algorithms for parametric worst-case tolerance analy-
sis that are more accurate, efficient, and comprehensive than Monte Carlo
134 9 Conclusion
simulation. For synthesis, we developed an interactive parameter synthesis

algorithm and a parameter optimization algorithm. The algorithms are
most useful for mechanisms with higher pairs and changes in the contact
of parts. We presented case studies of how they have helped designers de-
tect and correct unexpected behaviors and design flaws in automotive,
MEMS, and optical mechanisms.
Research Directions
We conclude with development and research directions. The most impor-

tant development goal is to integrate the configuration space method into
a commercial CAD package (figure 1.1). This will allow designers to create
and analyze mechanism models seamlessly within the configuration space
paradigm. The main technical issues are software integration, converting
CAD shape models to our format, and development of a graphical user
interface. As a small first step, the prototype source code in appendix B pro-
vides a textual input format and a simple user interface. Another develop-
ment direction is to integrate our kinematics software with multiphysics
analysis, which will extend the configuration space method to nonrigid
parts.
The main research directions are to increase the coverage of mechanism
types and to improve the support for design tasks. The coverage of mecha-
nisms can be extended by adding planar and spatial shape features of parts
by developing contact models for the new features and by handling pairs
with more degrees of freedom. For planar parts, the extension is to curve
segments, such as splines and involutes. For spatial parts, the extension is
to surface patches, such as helicoids, torii, and Bezier patches. Closed-form
solutions for the contact equations will most likely rarely be available, so
numerical solutions will be required.
Construction of a configuration space partition is implemented for fixed-
axis pairs [42, 71]. Fast, robust construction of three-dimensional partitions
is under development. General spatial pairs have six-dimensional parti-
tions, which are impractical to construct. Reduction of dimensions, for ex-
ample by slicing or projection, is a possible alternative. Construction of a
mechanism partition is impractical for all but the simplest multipair mech-
anisms. The mechanism partition can be sampled by kinematic simulation.
Other forms of local analysis should be considered, as should reduction of
dimensions.
For tolerancing, the extensions include three-dimensional contact zones
for general planar pairs and local contact zones for mechanisms. Statistical
Research Directions 135
tolerancing can also be incorporated by considering the joint distribution

of the tolerance parameters. The task is to compute the distribution of a ki-
nematic function from the joint distribution of the tolerance parameters.
In pair analysis, we compute a distribution for each contact zone region,
whereas in mechanism analysis we compute a distribution along a nominal
motion path. The kinematic function is linearized around the nominal pa-
rameter values. The linearization specifies the kinematic variation as a lin-
ear combination of given distributions. The outputs are the distributions of
the kinematic variation in the contact zones and along the motion’s path.
Statistical tolerance analysis lays the groundwork for allocating tolerances
based on cost functions. Finally, we have found that tolerance envelopes
of the shapes of parts [60] are useful in quantifying functional errors, iden-
tifying unexpected collisions of parts, and determining if a mechanism can
be assembled.
Kinematic synthesis is an open and fertile research area. Interactive mod-
ification using draggers can be extended to three-dimensional configu-
ration spaces. Other interaction paradigms can also be contemplated.
Support for configuration design and for incomplete, abstract specifications
would also be valuable.
The systematic exploration and characterization of kinematic function is
an important research topic. The goal is to develop a comprehensive taxon-
omy of this function. A promising avenue is to match kinematic function
to configuration space partition regions and develop a symbolic and alge-
braic language to describe kinematic function [48]. The kinematic descrip-
tions should characterize the qualitative and quantitative working of
mechanisms concisely and abstractly. One approach is to summarize the
function in terms of operating modes and mode transitions. Simplification
and abstraction of kinematic function [37] is required to compare and clas-
sify mechanisms [38]. Based on this classification, a taxonomy of kinematic
function could be developed and would provide a functional index for a
database of mechanisms.
Visualization and interactive manipulation of configuration space parti-
tions is another important area of research. The main challenges are to
detect qualitative patterns in three-dimensional pair partitions and to visu-
alize mechanism partitions. We can visualize the partitions of the kine-
matic pairs in a mechanism, which are projections of the mechanism
partition, but it is difficult to determine the global topology from the pro-
jections. Intelligent projection techniques are needed to resolve this issue.
The natural approach is to find subspaces that reveal the relevant interac-
tions of parts and to visualize them in a perspicuous way.
136 9 Conclusion
Specific design tasks pose their own visualization problems. Simulation

raises the need to visualize the motions of parts that generate paths in con-
figuration space. One option is to animate the parts while tracing the cor-
responding configurations in the (full or projected) configuration spaces.
This technique also helps probe the topology of the configuration space
partition without constructing it. Tolerance analysis raises the need to visu-
alize parametric families of partitions that represent the functional effects
of manufacturing variations. Contact patches generalize to narrow zones
surrounding the nominal patch surfaces.
The configuration space paradigm presented in this book may facilitate
other contact analysis tasks that arise in mechanical engineering, robotics,
computer graphics, and biomedical engineering. In mechanical engineer-
ing, contact analysis occurs in assembly [86] and [82] fixture design [10,
83] and part feeder design. In robotics, contact analysis occurs in path plan-
ning [69], analysis and planning of grasp and compliant motion, and
robotic manipulation [54]. In computer graphics, contact analysis occurs
in physics-based modeling and simulation, and in control and interaction
in virtual environments. In biomedicine, contact analysis supports kine-
matic analysis of musculoskeletal action and joint and implant modeling
[35].
Appendix A: Catalog of Mechanisms
This chapter contains a catalog of representative mechanisms and their

configuration space partitions. It illustrates the relation between a part’s
shape, its configuration space partition, and kinematic function. The cata-
log consists of 24 higher pairs and 6 mechanisms containing 16 higher
pairs. Of the 40 higher pairs, 33 are planar fixed-axis, 3 are general planar,
and 4 are spatial fixed-axis. All the planar fixed-axis pairs and three of the
mechanisms are available as HIPAIR files.
We begin each catalog entry with a description of the mechanism’s struc-
ture and kinematic function. We then describe the configuration space par-
titions and explain their relation to the kinematic function. In some cases,
we point out failure modes and establish the relation to other examples.
We use the following conventions for the figures. The left figure shows
the parts’ shapes, their typical configurations, configuration parameters,
and motion arrows. The intended rotation of the part and translation direc-
tions are indicated with thick-headed arrows. The rotation angle is always
measured in the standard counterclockwise direction, even when the arrow
points clockwise. The translation is always measured along the standard
right-hand coordinate axis, even when the arrow points left. A double
thick-headed arrow indicates back-and-forth motion. The right figure
shows the configuration space partition. For fixed-axis pairs, the horizontal
and vertical axes are the driver and follower configurations. The part’s
configuration in the left figure is shown as a thick black dot. A kinematic
simulation appears as a thick line. The simulation starts from the dot and
lasts for one driver period. Unless otherwise specified, all angles are in
½180 ; 180 .
138 Appendix A: Catalog of Mechanisms
Contents
1 Offset three-arc cam pair 139

2 Centered three-arc cam pair 140
3 Six-arc cam pair 141
4 Unidirectional three-finger cam pair 142
5 Bidirectional three-finger cam pair 143
6 Geneva pair 144
7 Inverse Geneva pair 145
8 Intermittent cam pair 146
9 Interlock pair 147
10 Star indexer pair 148
11 Disk indexer pair 149
12 Lever indexer pair 150
13 Interlock drive pair 151
14 Ratchet pair 152
15 Escapement pair 153
16 Double-rack gears pair 154
17 Sector gears pair 155
18 Dwell gears pair 156
19 Double-sector gears pair 157
20 Reciprocating indexer pair 158
21 Spatial indexer pair 159
22 Spatial Geneva pair 160
23 Spatial orthogonal gear pair 161
24 Spatial bevel gears mechanism 162
25 Counter mechanism 163
26 Pawl indexer mechanism 164
27 Intermittent gear mechanism 166
28 Lever indexer mechanism 167
29 Movie film advance mechanism 168
30 Camera shutter mechanism 169
1 Offset Three-Arc Cam Pair 139
1 Offset Three-Arc Cam Pair
The pair consists of a rotating cam with orientation yC that drives a verti-
cally translating follower with position yF . The cam contour consists of
three arc segments whose endpoints lie on a circle. The center of rotation
coincides with the common endpoint of two incident arcs. The follower
consists of a rectangular frame whose inner parallel horizontal line seg-
ments touch the cam.
A full rotation of the cam drives the follower up and down with two
dwell periods in between. The dwell periods are at configurations where
the common endpoint at the center of rotation touches the frame’s upper
and lower horizontal line segments. The contact point between the arc seg-
ment opposite the contact point and the follower lies on a circle of con-
stant radius. The pair is bidirectional.
The configuration space partition consists of a single narrow channel
whose nearly constant vertical height is the follower’s backlash (play). The
upper and lower horizontal channel segments correspond to the dwell pe-
riod. The slanted left and right channel segments correspond to the driving
periods. Its boundaries are sinusoidal curves.
2 Centered Three-Arc Cam Pair
The pair consists of a rotating cam with orientation yC that drives a verti-
cally translating follower with position yF . The cam contour consists of
three arc segments whose endpoints lie on a circle. Its center coincides
with the cam’s center of rotation. The follower consists of a rectangular
frame whose inner horizontal line segments touch the cam. The distance
between the upper and lower contact points is constant, so the cam fits
tightly inside the follower without interference.
A full rotation of the cam drives the follower up and down three times at
evenly spaced intervals. In each interval, one cam arc segment touches one
horizontal follower line segment. The transitions occur at the cam’s end-
points. The pair is bidirectional.
whose vertical width is the follower’s backlash. The slanted left and right
channel segments correspond to the driving periods, whose boundaries are
sinusoidal curves. The short horizontal segments represent the dwell peri-
ods when the cam rotates inside the follower without pushing it. The rising
and falling segments represent the periods when the cam pushes the fol-
lower right and left. The cam arc segment breaks contact with one side of
the follower at the start of each dwell period. The next arc segment makes
contact with the opposite vertex of the cam at the end of the dwell period.
The vertical channel width is the follower’s backlash. Although the pair is
structurally similar to pair 1 (the offset three-arc cam pair), the change in
the cam rotation center yields a different kinematic function.
3 Six-Arc Cam Pair 141
3 Six-Arc Cam Pair
The pair consists of a rotating cam with orientation yC that drives a hori-
zontally translating follower with position xF . The cam’s contour consists
of three convex circular arc lobes connected by three concave tangent cir-
cular arcs. The follower consists of two circular pins mounted on a rectan-
gular frame.
A full rotation of the cam drives the follower left and right three times at
evenly spaced intervals. The cam lobes alternately push the left and right
pins. Since the distance between the left and right contacts is constant,
the cam fits tightly between the two pins without interference. The pair is
bidirectional.
whose nearly uniform vertical width is the follower’s backlash. The slanted
left and right channel segments correspond to the driving periods, whose
boundaries are sinusoidal curves. The rising and falling segments of the
kinematic simulation represent the periods when the cam pushes the fol-
lower right and left. Pair 2 has a similar function, but this pair has less back-
lash and no dwell.
4 Unidirectional Three-Finger Cam Pair
The pair consists of a rotating three-finger cam with orientation yC that

drives a horizontally translating follower with position xF . The cam finger
tips are half-circles. The inner follower boundary is mirror symmetric. It
consists of an outer arc circle, an upper horizontal segment, a vertical seg-
ment, and a middle horizontal segment.
A full clockwise rotation of the cam drives the follower left and right
three times, with short dwell periods in between. The cam finger tips alter-
nately push the upper vertical segment of the follower right and the lower
vertical segment left. Counterclockwise rotation of the cam causes it to
block when one of its finger line segments hits a horizontal segment of
the follower. The pair is thus unidirectional.
The configuration space partition consists of a single channel with
shifted, mirror-symmetric upper and lower boundaries. The slanted chan-
nel boundary segments correspond to driving periods. The vertical channel
boundary segments correspond to blocking configurations in which the
cam cannot rotate. The kinematic function follows the thick line from
right to left. The horizontal segments represent the dwell periods when
the cam does not touch the follower. The rising and falling segments are
sinusoidal curves that represent the periods when the cam pushes the fol-
lower right and left. A cam finger breaks contact with one side of the fol-
lower at the start of each dwell period. The next finger makes contact with
the other side of the cam at the end of the dwell period. Its clockwise kine-
matic function is similar to that of pairs 2 and 3.
5 Bidirectional Three-Finger Cam Pair 143
5 Bidirectional Three-Finger Cam Pair
The pair consists of a rotating three-finger cam with orientation yC that

drives a horizontally translating follower with position xF . The cam finger
tips are half-circles. The inner follower boundary is mirror symmetric. It
consists of an outer arc circle, an upper horizontal segment, a vertical seg-
ment, and a middle slanted segment.
A full clockwise rotation of the cam drives the follower left and right
three times with short dwell periods in between. The cam fingers alter-
nately push the upper vertical segment of the follower right and the lower
vertical segment left. A full counterclockwise rotation of the cam drives the
follower in the same way, but the cam fingers now push the lower slanted
segment right and the upper ones left. The pair is bidirectional.
with shifted, mirror-symmetric upper and lower boundaries. The slanted
channel segments correspond to driving periods. The horizontal segments
of the kinematic function line represent the dwell periods when the cam
does not touch the follower. The rising and falling boundary segments are
sinusoidal curves that represent the periods when the cam pushes the fol-
lower right and left (or left and right). A cam finger breaks contact with
one side of the follower at the start of each dwell period. The next finger
makes contact with the other side of the cam at the end of the dwell pe-
riod. The free space channels are narrower than those of pair 4. The config-
uration space partition is similar to those of pairs 2 and 3.
6 Geneva Pair
The pair consists of a rotating driver with orientation yD and a rotating

Geneva wheel with orientation yW . The driver consists of a driving pin
and a concentric locking arc segment mounted on a disk (not shown). The
Geneva wheel consists of a cylindrical basis with four slots on which four
locking arc segments are symmetrically mounted.
As the driver rotates, the pin engages the Geneva wheel slots and rotates
it nonuniformly in the opposite direction. The locking arc segments of the
driver engage those of the wheel during the dwell periods, thereby prevent-
ing the wheel from rotating. A full rotation of the driver causes a nonuni-
form, intermittent rotation of the wheel with four dwell periods and four
drive periods. The pair is bidirectional.
whose nearly uniform vertical width is the wheel’s backlash. The slanted
sinusoidal curves.
7 Inverse Geneva Pair 145
7 Inverse Geneva Pair
The pair consists of a rotating driver with orientation yD and a rotating in-
verse Geneva wheel with orientation yW . The driver consists of a driving
pin and a concentric locking arc segment. The Geneva wheel consists of a
cylindrical base with three slots on which three locking arc segments are
symmetrically mounted.
As the driver rotates, the pin engages the inverse Geneva wheel slots and
rotates the wheel in the same direction. The locking arc segments of the
driver engage those of the wheel during the idle periods, thereby prevent-
ing the wheel from rotating. A full rotation of the driver causes a nearly
uniform, intermittent rotation of the wheel, with three dwell periods and
three drive periods. The pair is bidirectional.
sinusoidal curves. Compared with the pair 6 partition, the channels are
slanted in the opposite direction and the driving periods are longer.
8 Intermittent Cam Pair
The pair consists of a rotating driver with orientation yD and a rotating fol-
lower with orientation yF . The driver consists of a rectangular pin mounted
on a base. The follower consists of two square pins symmetrically mounted
on a disk. The centers of rotation for both parts are offset.
As the cam rotates, its rectangular pin pushes the follower square pin for
half a turn. The pins then slide out of engagement because of the eccentric-
ity of the axes of rotation. The follower is idle during the second half of the
cam rotation. The rectangular pin of the cam then engages the second
square pin of the follower for another half turn. The pair is bidirectional.
The configuration space partition consists of two slanted elongated strips
that nearly span a full turn. The strip boundaries correspond to the contact
of the driver pin with one of the square follower pins. The kinematic simu-
lation has a horizontal segment in free space where the driver rotates and
the follower is idle, and a slanted segment where the driver rotates the fol-
lower for nearly half a turn.
9 Interlock Pair 147
9 Interlock Pair
The pair consists of a rotating driver lock with orientation yD and a rotating
star-shaped follower with orientation yF . The driver lock consists of an arc
segment centered at the center of rotation and of a line segment. The fol-
lower star consists of six arc segments whose centers coincide with the cen-
ter of rotation of the lock.
The driver lock locks the follower star at six orientations. The orienta-
tion of the follower star can be changed by first turning the lock so that the
line segment faces it vertically, then rotating it. When the follower star is
spring loaded one way and the driver rotates the other way, the follower
turns one-sixth of a turn for each rotation of the driver lock. The pair is
bidirectional.
The configuration space partition consists of six identical slabs that span
360 . The narrow horizontal channels between slabs correspond to config-
urations in which the driver rotates and the follower is locked. At the end
of the channel, the follower star turns until it enters the next horizontal
channel.
10 Star Indexer Pair
The pair consists of a rotating driver with orientation yD and a rotating star-
like indexer with orientation yI . The driver consists of a disk with a recess
that allows the cusps of the indexer arc segment to pass underneath. The
indexer consists of ten arc segments whose centers coincide with the center
of rotation of the driver. The indexer is spring loaded counterclockwise.
A full counterclockwise rotation of the driver advances the indexer clock-
wise by one-tenth of a turn and locks it for the rest of the rotation. If the
indexer were not spring loaded counterclockwise, it would not engage the
driver slot and thus would not advance. If the spring is too weak, the driver
slot passes the indexer pin before it can engage, so the indexer does not ad-
vance. The pair is unidirectional.
The configuration space partition consists of ten horizontal and ten
slanted narrow channels. The horizontal channels correspond to the dwell
periods of the indexer. The slanted channels correspond to the driving peri-
ods. The channel width is the backlash between the driver and the indexer,
which is largest at transitions between drive and dwell periods. The config-
uration space partition, and thus the kinematic function, is nearly identical
to that of pair 11.
11 Disk Indexer Pair 149
11 Disk Indexer Pair
The pair consists of a rotating driver with orientation yD and a rotating

indexer with orientation yI . The driver is a disk with a slot cutout that
engages the indexer pins. The indexer has ten circular pins evenly spaced
on a disk and is spring loaded clockwise.
Continuous counterclockwise rotation of the driver causes intermittent
clockwise rotation of the indexer. In drive periods, the driver slot engages
one indexer pin and advances the indexer one-tenth of a turn. In dwell
periods, the outer arc of the driver engages two adjacent indexer pins and
prevents rotation of the indexer. When a dwell period ends, the indexer
spring rotates the next indexer pin into the driver slot and the next drive
period begins. If the indexer were not spring loaded, its configuration
would stay constant as the driver rotates. If the spring is too weak, the
driver slot passes the indexer pin before it can engage, so the indexer does
not advance.
ods. The channel width is the play between the driver and the indexer,
which is largest at transitions between drive and dwell periods. The config-
uration space partition, and thus the kinematic function, is nearly identical
to that of pair 10.
12 Lever Indexer Pair
The pair consists of a rotating driver with orientation yD and a rotating in-
dexer with orientation yI . The driver consists of a driving finger and a lock-
ing arc mounted on a disk (not shown). The indexer has eight trapezoidal
pins evenly spaced on a circular base.
Continuous counterclockwise rotation of the driver causes intermittent
clockwise rotation of the indexer. In drive periods, the driver pin engages
one indexer pin and advances the indexer one-eighth of a turn. In dwell
periods, the locking arc engages two adjacent indexer pins and prevents
the rotation of the indexer. The pair is unidirectional.
The configuration space partition consists of eight horizontal narrow
channels and eight free space regions in between. The horizontal channels
correspond to the dwell periods of the indexer. The channels’ width is the
play between the driver and the indexer. The free space regions have a nar-
row channel. The kinematic simulation consists of a slanted segment where
the driver pushes the indexer, followed by a long horizontal segment, fol-
lowed by a short slanted segment. Unlike pairs 10 and 11, spring loading
the indexer is unnecessary. The configuration space partitions are similar.
13 Interlock Drive Pair 151
13 Interlock Drive Pair
The pair consists of a rotating driver with orientation yD and a horizontally

translating rack with position xR . The driver is a disk with a single tooth
and two cutouts to allow the follower teeth to pass. The rack has five teeth.
The tooth top is a concave arc segment whose radius is the same as that of
the driver disk.
At each driver rotation, its tooth engages a rack tooth and pushes it. Then
its locking arc engages the concave top of the tooth and locks the rack.
Continuously rotating the driver makes the rack translate with dwell peri-
ods. The pair is bidirectional.
The configuration space partition consists of four horizontal and four
slanted channels. The horizontal channels correspond to the dwell periods
of the indexer. The channels’ width corresponds to the play between the
driver and the indexer. The top and bottom channels represent contacts
with the leftmost and rightmost tooth. Pair 7 has a similar partition.
14 Ratchet Pair
The pair consists of a rotating ratchet with orientation yR and a rotating

pawl with orientation yP . The ratchet consists of 16 triangular teeth. The
pawl is a quadrangle whose slanted tip matches the shape of the ratchet
teeth.
Rotating the ratchet counterclockwise causes the pawl to rise and then
fall under its own weight, following the profile of the teeth. The pawl
blocks clockwise rotation of the ratchet after at most one-sixteenth of a ro-
tation. Being unidirectional is the primary kinematic function.
The configuration space partition consists of an upper and lower bound-
ary. The lower boundary corresponds to configurations where the pawl is
above the ratchet, as shown in the figure. The upper boundary corresponds
to configurations where the pawl side is in contact with the ratchet teeth to
the right of the ratchet. The lower boundary realizes the kinematic func-
tion. It has one notch for each tooth of the ratchet. The notch slanting pre-
vents simultaneous clockwise rotation of the ratchet and counterclockwise
rotation of the pawl.
15 Escapement Pair 153
15 Escapement Pair
The pair consists of a rotating wheel with orientation yW and a rotating an-
chor with orientation yA . The wheel consists of 30 slanted triangular teeth.
The anchor consists of left and right triangular pallets that alternately en-
gage the wheel teeth. Attached to the anchor is a pendulum (not shown)
that oscillates with a constant period. The wheel rotates counterclockwise,
owing to a torque acting on its hub.
As the anchor oscillates, the pallets alternately engage teeth on the left
and right sides of the wheel. The wheel advances one tooth each oscilla-
tion. It rotates at a constant velocity that is determined by the period of
the anchor. The timing is such that the wheel advances one tooth every
second.
The figure shows a detail of the configuration space partition. The upper
and lower boundaries correspond to contacts of the right and left anchor
pallets with the teeth. The interleaved boundaries prevent the wheel from
rotating freely. In the kinematic simulation curve, the nearly horizontal
segments on the boundary show that the anchor oscillates back and forth
while the wheel follows it. The nearly vertical segments in free space show
that the wheel rotates quickly in between.
16 Double-Rack Gears Pair
The pair consists of a rotating sector gear with orientation yG and a hori-
zontally translating follower with position xF . The sector gear has teeth
that span less than half of its circumference. The follower has geared racks
at the top and the bottom of its inner perimeter. The upper and lower racks
are offset with respect to each other. The sector gear alternately engages the
upper and lower racks, pushing the follower left and right.
Continuous rotation of the sector gear causes the follower to move left
and right without dwell. The maximum displacement of the follower is
equal to the perimeter of the sector gear. The pair is bidirectional.
The configuration space partition consists of a narrow channel with three
nearly linear segments: two right slanted and one left slanted. The channel
width is the play between the driver and the indexer. The slanted segments
correspond to the right and left motions of the follower.
17 Sector Gears Pair 155
17 Sector Gears Pair
The pair consists of a rotating sector gear with orientation yA and a rotating
full gear with orientation yB . The gears have equal radii. The sector gear has
8 teeth, while the full gear has 16. When the gears are engaged and one
gear rotates in either direction, the other gear rotates in the opposite direc-
tion by the same angle. Two pairs of teeth are engaged at all times. The
transmission ratio is almost linear, with slight backlash and chatter. When
the gears are disengaged, their rotations are independent.
Continuous rotation of the sector gear causes the full gear to alternately
rotate half a turn and dwell. Continuous rotation of the full gear causes the
half gear to rotate in the opposite direction by at most half a turn. The pair
is bidirectional.
The configuration space partition consists of 16 slanted narrow channels
and a free space region (right). The slanted channels correspond to the driv-
ing periods at each of the 16 possible relative orientations in which the
gears are engaged. The channel width is the play between the two gears.
The horizontal segment of the kinematic simulation represents the dwell
period when the sector gear turns and the full gear is at rest. The slanted
line segment represents the gears rotating in opposite directions.
18 Dwell Gears Pair
The pair consists of a rotating driver sector gear with orientation yD and a
rotating follower gear with orientation yF . The driver is a concentric arc seg-
ment mounted on a gear sector with ten trapezoidal teeth. The follower is a
star-shaped disk mounted on a gear wheel. It has ten outer locking arc seg-
ments and ten teeth. The pair is a combination of pairs 9 and 17.
As the driver rotates counterclockwise, the driver sector gear engages and
turns the gear wheel clockwise. When the sector disengages, the concentric
arc segment engages one of the ten concave arcs of the star-shaped disk,
thereby locking it in a dwell position. A full rotation of the driver advances
the follower by one-tenth of a rotation. The pair is bidirectional.
ods. The channel width is the play between the driver and the follower
gears, which is largest at transitions between drive and dwell periods. The
configuration space partition is similar to those of pairs 9–12.
19 Double-Sector Gears Pair 157
19 Double-Sector Gears Pair
The pair consists of two rotating parts with orientations yA and yB . Each
part consists of concentric large and small sector gears. Part A has a large
gear sector with 10 teeth on top and a small gear sector with 17 teeth be-
low. Part B has a small gear sector with 10 teeth on top and a large gear sec-
tor with 18 teeth below. The corresponding small and larger gear sectors on
top and bottom alternatively engage. The pair has a structure similar to pair
17.
Rotating A causes B to rotate once in the opposite direction. The gear
ratio is 2:3 when the lower gears segments mesh and is 2:1 when the upper
gears segments mesh. The transmission is almost linear, with slight play
and chatter. The pair is bidirectional.
The configuration space partition consists of two types of slanted narrow
channels and two free space regions. The slanted channels correspond to
the driving periods at each relative orientation in which the gears engage.
The channel width corresponds to the play between the two gears. The dif-
ferent inclinations of the channels correspond the gear ratios of 2:1 and
2:3. The configuration space partition is the overlay of two shifted versions
of the pair 17 partition.
20 Reciprocating Indexer Pair
The pair consists of a reciprocating driver with position yD and a rotating

follower with orientation yF . The driver has two triangular teeth slanted in
opposite directions. The follower consists of an inner ring with 64 triangu-
lar teeth whose shape is complementary to that of the driver teeth. The
driver teeth alternately engage the left and right follower slots.
As the drive moves left and right, it rotate the follower clockwise by one
tooth per cycle. The pair is unidirectional.
The figure shows a detail of the configuration space partition. The
boundaries of the left and right triangular-shape free regions correspond to
the driver tooth contact with the follower teeth. The narrow passage be-
tween the two regions is the transition between the left and right contacts.
21 Spatial Indexer Pair 159
21 Spatial Indexer Pair
The pair consists of a rotating driver with orientation yD and a rotating fol-
lower with orientation yF . The driver consists of two cylindrical guides con-
nected by a slanted crossover. The follower is a gear with 21 rectangular
teeth. The rotation axes of the parts are perpendicular.
A full rotation of the driver advances the follower by one tooth. The
crossover rotates the follower, then the guides lock it for the remainder of
the driver’s rotation. The pair is bidirectional.
The figure shows a detail of the configuration space partition; the full
partition consists of 21 horizontal and slanted channels. The horizontal
channels correspond to the dwell periods of the follower. The slanted chan-
nels correspond to the driving periods. The channel width is the play be-
tween the driver and the follower.
22 Spatial Geneva Pair
The pair consists of a rotating cam with orientation yA and a rotating fol-
lower with orientation yB . The cam is a plate with a cylindrical pin and a
half cylinder mounted on it. The follower is a hollow hemisphere with
four evenly spaced slots and circular cutouts. The follower’s inner boundary
consists of a spherical patch with four slots and four cutouts. As the cam
rotates, the pin engages the follower slots and rotates it nonuniformly in
the opposite direction. The cam cylinder engages a follower cutout during
the idle periods, preventing its rotation.
As the driver rotates, the pin engages the Geneva wheel slots and rotates
it nonuniformly in the opposite direction. The locking arc segments of the
driver engage those of the wheel during the idle periods, thereby prevent-
ing the wheel from rotating. A full rotation of the driver causes a nonuni-
form, intermittent rotation of the wheel with four dwell periods and four
drive periods. The pair is bidirectional.
The configuration space partition consists of single narrow channel
sinusoidal curves. The configuration space partition is nearly identical (with
a 90 shift) to that of the planar Geneva pair 6.
23 Spatial Orthogonal Gear Pair 161
23 Spatial Orthogonal Gear Pair
The pair consists of identical toothed gears rotating around orthogonal

axes with orientations yA and yB . Each gear consists of five cylindrical teeth
mounted on a cylindrical base at evenly spaced orientations.
When one gear rotates in either direction, the other gear rotates in the
opposite direction by the same angle. Two pairs of teeth are engaged at all
times. The transmission ratio is almost linear. The pair is bidirectional.
The configuration space consists of a single narrow channel that wraps
around the vertical and horizontal axes. The channel width shows the
amount of gear backlash. The channel boundaries are nearly linear. The
rough edges reflect chatter and slight variations in the motion’s transmis-
sion ratio. The configuration space partition is similar to that of the en-
gaged (left) portion of the pair 17 partition.
24 Spatial Bevel Gears Mechanism
The mechanism consists of three rotating bevel gears: an 8-tooth driver

gear with orientation yD and two 15-tooth follower gears with orientations
yA and yB around a common rotation axis. The transmission ratio is almost
linear, with slight play and chatter.
A rotation of the driver gear alternately engages the right and left gears
and turns them by half a turn. When the driver gear engages one follower
gear, the other can rotate freely. The driver gear is bidirectional.
The two driver-follower configuration space partitions consist of 15
slanted narrow channels and a free space region. They are identical with
yD shifted by half a turn. The slanted channels correspond to the driving
periods at each of the 15 possible relative orientations in which the gears
25 Counter Mechanism 163
engage. The channel width is the play between the two gears. The free
space corresponds to independent gear rotations. The configuration space
partitions are identical to that of pair 17.
25 Counter Mechanism
The mechanism consists of a rotating pinion with orientation yD , a rotating

gear with orientation yG , a rotating pawl with orientation yP , and a fixed
anchor. The pinion consists of a disk and a single tooth. The gear consists
of 72 trapezoidal teeth. The pawl consists of a finger and a notch on its left
side. A torsional spring (not shown) rotates the pawl counterclockwise.
The anchor is a square that prevents the counterclockwise rotation of the
pawl.
The pinion rotates clockwise and advances the gear by one tooth per ro-
tation. Reverse rotation is prevented by the pawl blocking against the an-
chor. Forward rotation is limited to one tooth per cycle.
The figure shows details of the configuration space partitions of the two
pairs. Both partitions consist of slanted elongated strips forming narrow
channels. The kinematic simulation consists of horizontal segments in free
space where the pinion rotates and the gear is idle, and of a slanted seg-
ment where the pinion rotates the gear clockwise by one-seventy-second
turn, which in turn rotates the pawl clockwise.
26 Pawl Indexer Mechanism
The mechanism consists of a rotating driver with orientation yD , a rotating

indexer with orientation yI , and a rotating pawl with orientation yP . The
driver consists of a disk with a notch and a finger. The indexer consists of
a disk with ten evenly spaced circular pins. The pawl consists of left and
right tips. The left tip engages the driver notch and the right tip engages
the indexer pins.
As the driver rotates clockwise, its finger engages an indexer pin, rotates
the indexer counterclockwise by one-tenth of a turn, and disengages. When
26 Pawl Indexer Mechanism 165
the driver is disengaged (a), its rim aligns with the left pawl arm, which pre-
vents clockwise rotation, and the right pawl arm prevents indexer rotation
by engaging two pins. When the driver engages the indexer, its notch
aligns with the left pawl arm, which allows the bottom indexer pin to ro-
tate the right pawl arm out of the way (b). When it disengages, the left
pawl arm returns to its rim and the right arm re-locks the indexer (c). The
mechanism is unidirectional.
The driver-indexer free space consists of ten diagonal channels where the
finger engages the ten pins. The negative slope indicates that the parts ro-
tate in opposite directions. The rest of free space consists of the regions to
the left and right of the channels where the driver is disengaged. The driver-
pawl free space is bounded from below by the rim-left arm contact curve,
which is horizontal, and by the notch-left arm contact curves, which form
a v-shape. The horizontal curve blocks downward vertical motion, which
corresponds to clockwise pawl rotation, whereas the slanted curves permit
it. The indexer-pawl free space is bounded from above by the pin-right arm
contact curves, which form ten inverted v-shapes. When yI increases or
decreases, yP decreases along one side of the v-shape. Thus, indexer rota-
tion in either direction causes clockwise rotation of the pawl.
27 Intermittent Gear Mechanism
The mechanism consists of a constant-breadth cam, a square follower with

two pawls, and a gear with inner square teeth. The cam and the gear are
mounted on a fixed frame and rotate around their centers. The follower is
free and has thus three degrees of freedom. The mechanism has two general
planar pairs: cam-follower and gear-follower.
Rotating the cam causes the follower to rotate in step while reciprocating
along its length. The right follower pawl engages a gear tooth at cam angle
yC ¼ 0 (a), the follower rotates the gear 45 , the right pawl disengages at
yC ¼ 82 (b), the follower rotates independently while the gear dwells, then
the left pawl engages the gear at yC ¼ 90 and the cycle repeats (c).
The bottom row of figures shows details of the cross-sections of the gear-
follower configuration space partitions at the gear angles of the top fig-
28 Lever Indexer Mechanism 167
ures. The follower configuration is ð0; 0; 0Þ and the gear degrees of free-
dom are ðxG ; yG Þ. In (a), free space consists of three regions. In the middle
region, which contains the snapshot configuration, the right follower pawl
engages a gear tooth. In the region above and below, the pawl engages the
tooth below and above, respectively. In (b), the three regions have merged
into a single region, which shows that the follower can now cross over the
two gear teeth. In (c), free space once again consists of three regions and
the middle region contains the snapshot configuration in which the left
pawl engages a gear tooth.
28 Lever Indexer Mechanism
The mechanism consists of a rotating driver with orientation yD , a rotating

indexer with orientation yI , a rotating pawl with orientation yP , a rotating
pin with orientation yN , a free lever with degrees of freedom ðxL ; yL ; yL Þ, and
a frame. The driver is an offcenter cylinder that acts as a cam. The indexer
is a gear with 24 trapezoidal teeth. The pawl has a triangular tip shaped
to follow the indexer teeth. The pin is a rectangular block. The lever has a
rounded triangular tip shaped to engage the indexer teeth, a cylindrical
hole that fits over the cam, and a rectangular slot that fits over the pin.
The mechanism advances the indexer wheel by one tooth (15 ) for every
turn of the driver. As the driver turns clockwise, the lever tip traces a closed
trajectory whose form is determined by the relative position of the driver
and pin rotation axes, the driver’s offset, and the length of the lever. This
causes the indexer to rotate counterclockwise by 15 . The pawl prevents
clockwise rotation of the indexer. The mechanism is unidirectional.
The indexer-pawl partition detail shows that the vertical boundary seg-
ment is what prevents clockwise rotation of the indexer. The pair 14 parti-
tion is similar. The right-hand figure is a detail of the cross-section of the
indexer-lever partition.
29 Movie Film Advance Mechanism

30 Camera Shutter Mechanism 169
The mechanism consists of a rotating driver with orientation yD , a follower

with degrees of freedom ðxF ; yF ; yF Þ, and a vertically translating film strip
with position yS . The driver is a constant-breadth cam. The follower con-
sists of a horizontal slot, a square in which the driver cam fits, and a finger
that engages the film notches. The follower is attached to the frame by a
pin joint. The film is orthogonal to the figure’s plane. Its cross-section con-
sists of notches.
As the driver rotates counterclockwise, the follower tip engages the film,
pushes it down one frame, and retracts. The mechanism is bidirectional.
The coordinates of the driver-follower configuration space partition are
ðxA ; yA ; yA Þ. The free space is a narrow spiral channel. The coordinates of
the film-follower configuration space partition are ðxB ; yB ; yB Þ. The figure
shows a yB ¼ 0 cross-section detail. The other cross-sections are similar,
with rotated engagement curves.
30 Camera Shutter Mechanism
Figure A.1a–b shows the shutter mechanism of a 35-mm film camera. It

consists of a rotating driver with orientation yD , a rotating shutter with ori-
entation yS , a rotating shutter lock with orientation yL , a rotating advance
wheel with orientation yW , a rotating pawl with orientation yP , a rotating
film counter (not shown) with orientation yC , and a horizontally translat-
ing film with position xF . The driver consists of a cam, a slotted wheel, a
film wheel, and a counter wheel mounted on a shaft. The shutter consists
of a tip and a pin, and is spring loaded counterclockwise. The shutter lock is
spring loaded clockwise. Figures A.2a,c,e and A.3a,c,e show details of six
pairs.
The mechanism alternately loads and exposes the film frames. In the
loading mode, the user turns the wheel, which advances the film, which
rotates the driver. Loading ends when the driver engages the shutter in
the lock. In the exposure mode, the user presses the release button, which
rotates the lock, which releases the shutter. The shutter spring rotates the
shutter, which trips the curtain, which rotates away from the lens and
exposes the film.
Figure A.2b,d,f shows details of configuration space partitions for three
driver pairs. The boundary of the driver-shutter tip configuration space par-
tition shows how the driver cam raises and lowers the shutter tip (figure
A.2b). The configuration space partition for the driver-shutter lock shows
how the shutter lock tip prevents the driver wheel from rotating when it
is inside its slot (figure A.2d). The configuration space partition for the
Figure A.1
Camera shutter mechanism: (a) 3D view, (b) 2D view.
Figure A.2
Pairs and details of their configuration space partitions: (a–b) driver-shutter tip, (c–d)
driver-shutter lock, (e–f) driver-film.
Figure A.3
Pairs and details of their configuration space partitions: (a–b) driver-film counter,
(c–d) shutter lock-shutter pin, (e–f) advance wheel-pawl.
driver-film is a single narrow channel that shows the nearly linear relation
between the driver’s rotation and the film advance with almost no play.
Figure A.3b,d,f shows details of configuration space partitions for three
driver pairs. The configuration space partition for the driver counter-film
counter consists of narrow channels at each tooth orientation. Note that
the driver counter locks the film counter when not advancing it (figure
A.3b). The notch in the configuration space partition for the shutter lock-
shutter pin (figure A.2d) shows how the shutter lock locks and releases the
shutter pin. The configuration space partition for the advance wheel-pawl
(figure A.2f) is similar to the configuration space partition for pair 14.
Appendix B: HIPAIR Software
In this appendix, we describe the HIPAIR mechanism design program built

around computation, visualization, and manipulation of the configuration
space. We describe installation, the graphical user interface (GUI), and the
input format. The program is written in Ansi Cþþ using openGL and glut.
It is free for education and research. Send comments, bug reports, and re-
quest for commercial licenses to Elisha Sacks ([email protected]).
B.1 Installation
The website is http://www.cs.purdue.edu/archives/2008/eps/. The

program runs under Linux and Windows. The archive hipair.tar con-
tains the source files, which are the same for both operating systems, and
the ex subdirectory, which contains input files for most of the mechanisms
in appendix A. The file hipair.exe is a windows executable.
Here is how to install and run the Linux version.
% tar -xvf hipair.tar
% make
% hipair ex/3finger
If the argument is omitted, HIPAIR prompts for it and reads it from stan-
dard input. In Windows, create a project with the source file suffix changed
to .cpp and with glut installed.
B.2 Graphical User Interface
The program opens a parts windows and one cspace (configuration space)
window per pair (figure B.1). The windows can be resized. The parts win-
dow shows the parts in the current configuration. Each cspace window
shows its configuration space with the animation path in red, the
176 Appendix B: HIPAIR Software
Figure B.1
Parts window (a) and cspace window (b) for ex/3finger.
current configuration as a green dot, and the contact zone in green when
requested. The program normally displays the current animation step. The
user can increase or decrease the step, or can animate the mechanism. The
animation starts at the current step and ends at the last step or when inter-
rupted. The animation speed is specified as a multiple of one-thirtieth of a
second. The user can specify an arbitrary configuration with the mouse.
The next animation request resets the configuration to the current anima-
tion step. The user can set the contact zone width and can toggle the con-
tact zone display.
The program is controlled by the mouse. Press, drag, and release the left
button to zoom to a rectangle. Click the middle button in a cspace window
to set its two configuration parameters to the mouse’s position. Click the
right button to bring up this menu. The menu entries can also be invoked
by typing the letters in parentheses.
n
Start/stop animation (a).
n
Next animation step (n).
n
Previous animation step (p).
n
Set animation speed (s).
n
Draw parts in wireframe/filled (f).
n
Restore initial viewing box (r).
n
Show/hide contact zones (z).
n
Set contact zone width (w).
n
Save window in postscript (o).
n
Exit (q).
B.3 Mechanisms 177
Figure B.2
Square-circle pair (a) and its configuration space partition (b).
B.3 Mechanisms
A mechanism consists of planar parts that form higher kinematic pairs.

HIPAIR computes configuration spaces for the pairs, computes the part’s
motions for given driving motions, and computes contact zones that
bound the kinematic variation for the given part tolerances. This section
explains these concepts.
B.3.1 Parts
A part is specified in a parts coordinate system. The part consists of regions
in the xy plane that are extruded over intervals on the z axis. A region has
an outer boundary and may also have inner boundaries. A boundary is a
simple loop of line and circle segments.
For example, a cylinder of 10 mm height is specified as a circle in the xy
plane that is extruded over 0 a z a 10. The region boundaries of a slice
must be disjoint, but slices may overlap. A part has an initial position and
orientation in world xy coordinates. It has one degree of freedom: rotation
around its origin, which is fixed at its initial position, or translation along a
line that passes through its initial position.
Figure B.2a shows a simple mechanism composed of two parts. Each part
consists of a single slice with z range ½0; 1. The first part is a unit square. Its
origin is the bottom left corner. Its boundary consists of four line segments.
The part rotates with initial position ð1; 1Þ and orientation 0. The second
part is a circle of radius 1 whose origin is the circle’s center. The part trans-
lates horizontally with initial position ð3; 1Þ and orientation 0.
B.3.2 Configuration Spaces of Fixed-Axis Pairs

Interactions between pairs of parts are modeled as configuration spaces.
The configuration space is a two-dimensional manifold whose coordinates
are the parts’ degrees of freedom. Figure B.2b shows the configuration space
of the square-circle pair. It is a cylinder whose coordinates are the square
orientation y and the rectangle offset x. The configuration space is a torus
when both parts rotate and is a rectangle when both translate. Configura-
tion space partitions into free space where the parts do not touch (the
white region in the figure), blocked space where they overlap (the gray re-
gion), and contact space where they touch (the black curves). The large dot
marks the displayed configuration of ð0; 3Þ.
B.3.3 Kinematic Simulation

Kinematic simulation takes the driving motion of a part as input and com-
putes motions for the other parts that prevent overlap. In our example, the
driving motion is clockwise rotation of the square by one full turn. The cir-
cle translates clear of the square, then stands still. The motion’s path is the
thick black curve in figure B.2b.
B.3.4 Contact Zones of Fixed-Axis Pairs

Variation in the shapes of parts is modeled as a zone around the nominal
boundaries of the part where the actual boundaries are allowed to lie.
HIPAIR supports constant-width offsets. As the boundaries of the parts in a
kinematic pair range over their zones, their contact spaces vary in a zone
around the nominal space, called the contact zone of the pair. The contact
zone topology and geometry model the kinematic variation that is due to
the parts’ variation. Figure B.3 shows the square-circle contact zone for off-
sets of 0.1 on both parts. It forms a narrow band around the nominal con-
Figure B.3
Square-circle contact zone.
tact space, which shows that variation in the parts merely perturbs the
nominal kinematic function.
B.4 Input Format
The input is a mechanism, simulation data, and display data.
B.4.1 Mechanisms
The mechanism format is specified by the following BNF formulas. The first
line specifies the number of parts and pairs, then come the parts, then the
pairs. The parts are assigned indices starting from zero. A pair is specified by
two part indices and the accuracy of the configuration space computation.
MECHANISM ¼ 1 nparts npairs 0 0 PART PAIR

PART ¼ name nslices color parameters SLICEþ
SLICE ¼ nboundaries zlower zupper BOUNDARYþ
BOUNDARY ¼ ngroups SGROUPþ

SGROUP ¼ nsegments SEGMENT þ noffsets offset
SEGMENT ¼ LINE_SEGMENT or ARC_SEGMENT
LINE_SEGMENT ¼ index 0 tx ty hx hy flag

ARC_SEGMENT ¼ index 1 tx ty hx hy flag cx cy r s e
PAIR ¼ index1 index2 accuracy
B.4.2 Parts
A part is specified by a name, number of slices, color, parameters, and slices.
The name is a character string. The color is three integers in the range
½0; 255 followed by a dummy integer. The parameters are seven doubles.
Parameter 1 is the motion type: 1 for translation along the x axis, 2 for
translation along the y axis, 3 for rotation, and 4 for translation along a
slanted axis. Parameters 2–4 are the initial position and orientation of the
part. Parameters 5–6 are the lower and upper bounds of the part’s motion
parameter. For motion type 4, parameter 7 is the slope of the translation
axis. The motion parameter and its bounds are measured along this axis.
A slice is specified by the number of boundaries, the lower and upper z
values, the outer boundary, and the inner boundaries. A boundary is speci-
fied by the number of segment groups followed by the groups. A group is a
sequence of incident segments and a sequence of offsets. A segment is
specified by a dummy index, 0 or 1 for a line or circle, a tail ðtx ; ty Þ, a head
Figure B.4
Square-circle file format.
ðhx ; hy Þ, and a flag that is 1 when the part’s interior lies to the left when the
segment is traversed from tail to head. A circle segment also has a center
ðcx ; cy Þ, radius r, start angle s, and end angle e.
The segments in the group are repeated noffset times and the ith copy is
offset by the ith element of the offset sequence. The offset is along the part’s
motion axis: the segments of a rotating part are rotated and the segments
of a translating part are translated. Offsets are optional since any boundary
can be specified as one group with noffset 1 and offset 0.
Figure B.4 illustrates the file format with the square-circle mechanism
from figure B.2. The first line indicates two parts and one pair. Lines 3–11
describe the first part. Line 3: name ‘‘square,’’ one slice, and red green blue
(RGB) color ð255; 0; 0Þ. Line 4: motion type 3 (rotation), initial position
ð1:0; 1:0Þ, initial orientation 0:0 radians, and rotation range ½p; p radians.
Line 5: the slice has one boundary and has z extent ½0:0; 1:0. Line 6: the
boundary consists of one sgroup with four segments. Lines 7–10: four line
segments. Line 11: trivial shift sequence, explained next. Lines 13–20 spec-
ify the circle: name ‘‘circle,’’ one slice, RGB color ð0; 255; 0Þ, motion type 1
Figure B.5
Geneva pair (a) and its configuration space (b).
(x translation), initial position ð1:0; 4:0Þ, initial orientation 0:0, translation

range ½5; 5, the slice has one boundary with two segments, two circle seg-
ments, and the shift sequence. The last line specifies the one pair: part
index 0 (the square), part index 1 (the circle), and configuration space accu-
racy 0.001.
Figure B.5 shows a Geneva mechanism. The configuration space is a torus
because both parts rotate. The driver has three slices with z interval ½0; 1: a
locking arc, a driving pin, and a central pin. The three slices are connected
by a plate (not shown) with z interval ½1; 0. The wheel has one slice. The
outer boundary consists of a six-segment sgroup with four offsets. The in-
ner boundary is one circle. Figure B.6 shows the follower specifications.
The offsets, 0, p=2, p, 3p=2 radians, generate four rotated copies of the six
segments.
B.4.3 Simulation and Display Data

The simulation data are specified by the following BNF formulas.
MOTIONS ¼ nm MOT_DESCR CONFIG
MOT_DESCR ¼ mtime nm_parts PART_MOT_DESCR

PART_MOT_DESCR ¼ pindex velocity
CONFIG ¼ one double per part
The number of motions is followed by one descriptor per motion. A

motion descriptor is the motion’s time, the number of moving parts, then
the descriptors of the part’s motion. A part’s motion descriptor is a part
index followed by a velocity. The motion begins with the part’s motion
Figure B.6
Follower file format.
parameters specified in CONFIGURATION. In the square-circle example,

the motion sequence of rotating the square with angular velocity 1.0 for
7.0 seconds, then translating the circle with x velocity 3.0 for 5.0 seconds
has the following file descriptor. The starting configuration is y ¼ 0 and
x ¼ 3.
2
7.0 1 0 -1.0
5.0 1 1 3.0
0 3
The display data are two dummy values followed by a part bounding box.
It is
0 0 -5 5 -5 5
in our example.
Glossary
axis a fixed line in space.
boundary representation representation of a part’s shape describing the shape in

terms of its features.
conceptual design the task of selecting a design concept that captures the desired
kinematic function.
configuration position and orientation of one coordinate frame with respect to

another.
configuration space vector space that specifies all possible configurations for a part
or a mechanism.
configuration space, blocked space subspace of configuration space where parts

overlap.
configuration space, cross-section cross-section of a general planar pair configura-

tion space in which the orientation of the moving part is fixed.
configuration space, contact space subspace of configuration space where parts are
in contact.
configuration space, free space subspace of configuration space where parts are
disjoint.
configuration space partition decomposition of a configuration space into free,

blocked, and contact spaces.
configuration space, region subspace of configuration space.
configuration space, topology configuration space regions and their adjacency

relations.
configuration variable position and orientation variables.
contact parts whose interiors are disjoint and whose boundaries overlap.
184 Glossary
contact change transition from one pairwise feature contact to another.
contact configuration configuration in which parts are in contact.
contact curve zero set of contact equation in two variables.
contact equations equation in the configuration variables of two parts whose zero
set consists of contact configurations.
contact type contact between two features of a part. Contact types between simple
planar features include circle-circle, circle-line, line-line, vertex-vertex, circle-vertex,
and line-vertex contacts.
contact zone union of the contact spaces of the mechanisms in a tolerance space.
cross-section two-dimensional slice of a part.
degrees of freedom independent configuration variables that are needed to specify

the configuration of a part or mechanism.
design space hyperbox in a configuration space that is the cross-product of one de-
sign interval per parameter.
dynamical simulation computation of the motions of parts resulting from the input
and external forces acting on the parts.
feature element of a part’s boundary. Simple features include points and line and
arc segments in the plane and vertex, edges, or faces in space.
fixed-axis mechanism mechanism whose parts move along axes that are fixed in
space.
fixed-axis motion rotation or translation along an axis that is fixed in space.
fixed-axis pair kinematic pair composed of fixed-axis parts.
fixed-axis part part with one degree of freedom (rotation or translation).
general planar pair two parts whose relative motions are circumscribed to a plane.
higher pair two parts not in permanent surface contact that move relative to each
other.
HIPAIR open-source Cþþ software package that implements some of the kinematic
design methods for mechanisms described in this book.
kinematic analysis derivation of the kinematic function of a mechanism from spec-

ification of the shapes of its parts and their motion constraints.
kinematic design process of designing a mechanism to achieve a desired kinematic

function.
kinematic function function describing the transformation of input motions into

output motions by the contact of parts.
Glossary 185
kinematic pair two parts whose motions are constrained by contacts.
kinematic simulation computation of compliant motions of parts resulting from an

input driving motion.
kinematic synthesis creation or modification of parts’ shapes and motion con-

straints to achieve a desired kinematic function. The inverse of kinematic analysis.
kinematic tolerancing derivation and modification of the kinematic variation of a

mechanism that is due to the shape of its parts and variations in configuration.
kinematic variation changes in the kinematic function of a mechanism that are due
to variations in the shapes and configuration of its parts.
kinematics branch of mechanics that studies the motions of parts independently of

the forces acting on them.
linkage mechanism formed by lower pairs only.
lower pair two parts in permanent surface contact that move relative to each other.
mechanism assembly of moving and fixed parts.
mechanism classification classification of mechanisms according to their structural

and functional characteristics.
mechanism configuration position and orientation of a mechanism’s parts.
mechanism design the process of creating or modfying a mechanism to achieve a

desired function.
motion constraint motion ranges and relations induced by contact of parts.
nominal model parametric model with parameter values that achieve the intended
kinematic function in the absence of variation that is due to manufacturing.
nominal parameter values parameter values of the nominal model.
pair configuration position and orientation of two parts.
parameter real-valued variable.
parametric design search for values of parameters that achieve a design goal.
parametric model description of the shapes and configuration of a mechanism’s

parts in terms of their parameters.
parametric optimization search for values of parameters that maximize a design

quality metric.
parametric synthesis synonym of parametric design.
parametric tolerancing search for values of parameters that minimize design

variation.
186 Glossary
part rigid solid with a fixed shape.
part configuration position and orientation of a part.
planar mechanism mechanism whose parts are planar.
planar motion motions occuring in a plane (one rotation, two translations).
planar part part with a fixed cross-section with respect to a plane.
subsumed contact contact prevented by other feature contacts.
tolerance interval of allowable values for a parameter.
tolerance analysis analysis of the variation in a design due to tolerances on the

parameters of a parametric model.
tolerance, nominal value numerical value of a design parameter in the nominal

design.
tolerance space hyperbox in parameter space that is the cross-product of one toler-
ance interval per parameter.
tolerance, worst-case computation of the maximal variation from the nominal

kinematic function over the tolerance space.
tolerancing the task of assigning tolerances to the shapes and configurations of

parts.
topology graph graph representation of the contacts of parts in a mechanism. Each

part is represented by a node, and edges between nodes represent contacts.
References
[1] American Society of Mechanical Engineers. Dimensioning and Tolerancing Stan-

dard. ASME Y14.5. New York, 1994.
[2] American Society of Mechanical Engineers. Mathematical Definition of Dimension-

ing and Tolerancing Principles. ASME Y14.5M. New York, 1994.
[3] Jorge Angeles, Gunter Hommel, and Peter Kovacs. Computational Kinematics.
Springer-Verlag, New York, 1999.
[4] Jorge Angeles and Carlos Lopez-Cajun. Optimization of Cam Mechanisms. Kluwer
Academic Publishers, Dordrecht, Boston, London, 1991.
[5] Ivan Artobolevsky. Mechanisms in Modern Engineering Design, vols. 1–4. MIR Pub-
lishers, Moscow, 1979. English translation.
[6] Chandrajit L. Bajaj and M.-S. Kim. Generation of configuration space obstacles:
The case of moving algebraic curves. Algorithmica, 4(1):157–172, 1989.
[7] Chandrajit L. Bajaj and M.-S. Kim. Generation of configuration space obstacles:
The case of moving algebraic surfaces. International Journal of Robotics Research,
9(1):92–112, 1990.
[8] Eric Ballot and Pierre Bourdet. A computation method for the consequences of
geometric errors in mechanisms. In H. Elmaraghy, editor, Geometric Design Toleranc-
ing: Theories, Standards, and Applications, pages 197–207. Chapman and Hall, 1998.
[9] Pierre Bourdet and Luc Matthieu. Geometric Product Specification and Verification:
Integration of Functionality. Kluwer Academic Publishers, Dordrecht, Boston, London,
2003.
[10] Randy Brost and Kenneth Goldberg. A complete algorithm for synthesizing
modular fixtures for polygonal parts. IEEE Transactions on Robotics and Automation,
12(1):31–46, 1996.
[11] Randy C. Brost. Computing metric and topological properties of configuration-

space obstacles. In Proc. of the IEEE Conference on Robotics and Automation, pages 170–
176. IEEE Press, 1989.
188 References
[12] Michael E. Caine. The design of shape interactions using motion constraints. In
Proc. of the IEEE Int. Conference on Robotics and Automation, pages 366–371, 1994.
[13] Kenneth Chase. Basic tools for tolerance analysis of mechanical assemblies. In
H. Geng, editor, Manufacturing Engineering Handbook, chapter 7. McGraw-Hill, 2004.
[14] Sao-Chyi Chen, Gary L. Kinzel, and David J. Kuhlmann. A numerical method
for the kinematic analysis of planar higher pairs in rolling contact. Mechanism and
Machine Theory, 20(6):565–575, 1985.
[15] N. Chironis. Mechanisms, Linkages, and Mechanical Controls. McGraw-Hill, New

York, 1967.
[16] H. Choset, K. M. Lynch, S. Hutchinson, G. Kantor, L. E. Kavraki, W. Burgard,

and S. Thrun. Principles of Robot Motion: Theory, Algorithms, and Implementations. MIT
Press, Cambridge, MA, 2005.
[17] André Clemént, Alain Rivière, and Catherine Valade. The TTRS: 13 oriented
constraints for dimensioning, tolerancing, and inspection. In F. Pavese P. Ciarlini,
M. G. Cox and D. Richter, editors, Advanced Mathematical Tools in Metrology III, vol-
ume 45 of Series on Advances in Mathematics for Applied Sciences, pages 24–42. World
Scientific, 1997.
[18] M. de Berg, M. van Kreveld, M. Overmars, and O. Schwarzkopf, editors. Com-

putational Geometry: Algorithms and Applications. 2nd ed. Springer-Verlag, Berlin,
2000.
[19] A. D. Dimarogonas. Machine Design: A CAD Approach. Wiley-Interscience, New

York, 2000.
[20] Bruce R. Donald. A search algorithm for motion planning with six degrees of
freedom. Artificial Intelligence, 31(3):295–353, 1987.
[21] Arthur G. Erdman. Modern Kinematics: Developments in the Last Forty Years.
Wiley, New York, 1993.
[22] Arthur G. Erdmann and George N. Sandor. Mechanism Design: Analysis and Syn-
thesis, vols. 1–2. 3rd ed. Prentice Hall, Englewood Cliffs, NJ, 1997.
[23] Boi Faltings. Qualitative kinematics in mechanisms. Artificial Intelligence,

44(1–2):89–120, 1990.
[24] Boi Faltings. A symbolic approach to qualitative kinematics. Artificial Intelligence,

56(2–3), 1992.
[25] F. Freudenstein and E. R. Maki. The creation of mechanisms according to kine-

matic structure and function. Environment and Planning B, 6(4):379–391, 1979.
[26] Max Gonzales-Palacios and Jorge Angeles. Cam Synthesis. Kluwer Academic Pub-
lishers, Dordrecht, Boston, London, 1993.
References 189
[27] R. S. Hartenberg and J. Denavit. Kinematic Synthesis of Linkages. McGraw-Hill,

New York, 1964.
[28] Edward J. Haug. Computer-Aided Kinematics and Dynamics of Mechanical Systems.

Allyn and Bacon, Needham Heights, MA, 1989.
[29] S. P. Hoover and J. Rinderle. A synthesis strategy for mechanical devices.

Research in Engineering Design, 1(2):87–103, 1989.
[30] J. Horton and F. Jones, editors. Ingenious Mechanisms for Designers and Inventors,
vols. 1–4. Industrial Press, New York, 1951.
[31] K. H. Hunt. Kinematic Geometry of Mechanisms. Oxford Engineering Science

Series, Clarendon Press, Oxford, 1978.
[32] Javier Garcia de Jalon and Eduardo Bayo. Kinematic and Dynamic Simulation of
Multibody Systems. Springer-Verlag, New York, 1993.
[33] Preben W. Jensen. Classical and Modern Mechanisms for Engineers and Inventors.
Marcel Dekker, New York, 1991.
[34] Leo Joskowicz and S. Addanki. From kinematics to shape: An approach to inno-
vative design. In Proc. of the National Conference on Artificial Intelligence, pages 347–
352. AAAI Press, 1988.
[35] Leo Joskowicz and R. H. Taylor. Interference-free insertion of a solid body into a
cavity: An algorithm and a medical application. International Journal of Robotics Re-
search, 15(3):211–229, 1996.
[36] Leo Joskowicz. Simplification and abstraction of kinematic behaviors. In Proc. of

the 11th Int. Joint Conference on Artificial Intelligence, pages 1337–1342. Morgan Kauff-
man Publishers, 1989.
[37] Leo Joskowicz. Reasoning about the kinematics of mechanical devices. Interna-
tional Journal of Artificial Intelligence in Engineering, 4(1):22–31, 1989.
[38] Leo Joskowicz. Mechanism comparison and classification for design. Research in
Engineering Design, 1(3–4):149–167, 1990.
[39] Leo Joskowicz and Dorothy Neville. A representation language for mechanical
behavior. Artificial Intelligence in Engineering, 10(1):109–116, 1996.
[40] Leo Joskowicz and Elisha Sacks. Computational kinematics. Artificial Intelligence,
51(1–3):381–416, 1991.
[41] Leo Joskowicz, Elisha Sacks, and Vijay Srinivasan. Kinematic tolerance analysis.
Computer-Aided Design, 29(2):147–157, 1997.
[42] Ku-Jim Kim, Elisha Sacks, and Leo Joskowicz. Kinematic analysis of spatial fixed-
axis higher pairs using configuration spaces. Computer-Aided Design, 35(3):279–291,
2002.
190 References
[43] Shridar Kota and Shean-Juinn Chiou. Design representation and computational
synthesis of mechanical motions. In Design theory and Methodology. ASME Press,
1992.
[44] Min-Ho Kyung and Elisha Sacks. Nonlinear kinematic tolerance analysis of pla-
nar mechanical systems. Computer-Aided Design, 35(10):901–911, 2003.
[45] Min-Ho Kyung and Elisha Sacks. Parameter synthesis of higher kinematic pairs.
Computer-Aided Design, 35:567–575, 2003.
[46] Min-Ho Kyung and Elisha Sacks. Robust parameter synthesis for planar higher
pair mechanical systems. Computer-Aided Design, 38(5):518–530, 2006.
[47] Jean-Claude Latombe. Robot Motion Planning. Kluwer Academic Publishers, Bos-
ton, 1991.
[48] Leo Joskowicz and Dorothy Neville. A representation language for mechanical
behavior. Artificial Intelligence in Engineering, 10:109–116, 1996.
[49] C. L. Li, K. W. Chan, and S. T. Tan. A configuration space approach to the auto-
matic design of multiple-state mechanical devices. Computer-Aided Design, 31:621–
653, 1999.
[50] Ming C. Lin, Dinesh Manocha, Jon Cohen, and Stefan Gottschalk. Collision de-
tection: Algorithms and applications. In Jean-Paul Laumond and Mark Overmars,
editors, Algorithms for Robotic Motion and Manipulation, pages 129–141. A. K. Peters,
Boston, MA, 1997.
[51] Faydor L. Litvin. Gear Geometry and Applied Theory. Prentice Hall, Engleword
Cliffs, NJ, 1994.
[52] Tomás Lozano-Pérez. Spatial planning: A configuration space approach. In IEEE

Transactions on Computers, vol. C-32, pages 108–120. IEEE Press, Piscataway, NJ,
1983.
[53] Martti Mantyla. An Introduction to Solid Modeling. Computer Science Press, Col-
lege Park, MD, 1988.
[54] Mathew T. Mason. Mechanics of Robotic Manipulation. MIT Press, Cambridge,

MA, 2001.
[55] John M. McCarthy. Kinematics of Robot Manipulators. MIT Press, Cambridge, MA,
1987.
[56] John M. McCarthy. Introduction to Theoretical Kinematics. MIT Press, Cambridge,

MA, 1990.
[57] John M. McCarthy. Geometric Design of Linkages. Interdisciplinary Applied Math-

ematics Series, Springer-Verlag, New York, 2000.
References 191
[58] Michael McCarthy and Leo Joskowicz. Kinematic synthesis. In E. K. Antonsson

and J. Cagan, editors, Formal Engineering Design Synthesis, pages 321–362. Cambridge
University Press, 2001.
[59] Parviz E. Nikravesh. Planar Multibody Dynamics. CRC Press, Abingdon, England,
2007.
[60] Yaron Ostrovsky-Berman and Joskowicz. Tolerance envelopes of planar me-

chanical parts with parametric tolerances. Computer-Aided Design, 37(5):531–544,
2005.
[61] Panos Papalambros and Douglass Wilde. Principles of Optimal Design. Cambridge
University Press, 1988.
[62] Richard P. Paul. Robot Manipulators: Mathematics, Programming, and Control. MIT
Press, Cambridge, MA, 1981.
[63] Umberto Prisco and Giuseppe Giorleo. Overview of current CAT systems. Inte-
grated Computer-Aided Engineering, 9(4):373–387, 2002.
[64] Rajan Ramaswamy. ‘‘Computer tools for preliminary parametric design.’’ PhD
thesis, Massachussetts Institute of Technology, Cambridge, MA, 1993.
[65] Charles F. Reinholtz, Sanjay G. Dhande, and George N. Sandor. Kinematic anal-
ysis of planar higher pair mechanisms. Mechanism and Machine Theory, 13:619–629,
1978.
[66] Aristides A. G. Requicha. Mathematical definition of tolerance specifications.

Manufacturing Review, 6(4):269–274, 1993.
[67] Franz Reuleaux. The Kinematics of Machinery: Outline of a Theory of Machines.

Dover Publications, New York, 1963.
[68] Elisha Sacks. Practical sliced configuration spaces for curved planar pairs. Inter-
national Journal of Robotics Research, 18(1):59–63, 1999.
[69] Elisha Sacks. Path planning for planar articulated robots using configuration
spaces and compliant motion. IEEE Transactions on Robotics and Automation, 19(3),
2003.
[70] Elisha Sacks and Steven M. Barnes. Computer-aided kinematic design of a tor-
sional ratcheting actuator. In Proc. of the Fourth Int. Conference on Modeling and Simu-
lation of Microsystems. Nara Sciences and Technology Institute, Hilton Head, SC,
2001.
[71] Elisha Sacks and Leo Joskowicz. Computational kinematic analysis of higher
pairs with multiple contacts. Journal of Mechanical Design, 117(2(A)):269–277, 1995.
[72] Elisha Sacks and Leo Joskowicz. Parametric kinematic tolerance analysis of pla-
nar mechanisms. Computer-Aided Design, 29(5):333–342, 1997.
192 References
[73] Elisha Sacks and Leo Joskowicz. Parametric kinematic tolerance analysis of gen-
eral planar systems. Computer-Aided Design, 30(9):707–714, 1998.
[74] Elisha Sacks, Leo Joskowicz, Ralf Schultheiss, and Min-Ho Kyung. Towards ro-
bust kinematic synthesis of mechanical systems. In P. Bourdet and L. Mathieu, edi-
tors, Geometric Product Specification and Verification: Integration of Functionality, pages
135–144. Springer, New York, 2003.
[75] Elisha Sacks, Leo Joskowicz, Ralf Schultheiss, and Uwe Hinze. Computer-assisted
kinematic tolerance analysis of a gear selector mechanism with the configuration
space method. In 25th ASME Design Automation Conference, Las Vegas, ASME Press,
1999.
[76] Elisha Sacks, Leo Joskowicz, Ralf Schultheiss, and Uwe Hinze. Redesign of a spa-
tial gear pair using configuration spaces. In Proc. of the 2002 Association of Mechanical
Engineers Design Automation Conference, Montreal, 2002.
[77] W. Schiehlen. Multibody Systems Handbook. Springer-Verlag, Berlin, 1990.
[78] Joseph E. Shigley, Charles R. Mischke, and Richard G. Budynas. Mechanical Engi-
neering Design. 7th ed., McGraw-Hill, New York, 2004.
[79] Lung-Wen Tsai. Enumeration of Kinematic Structures According to Function. CRC

Press, Boca Raton, FL, 2001.
[80] Herbert Voelcker. A current perspective on tolerancing and metrology. Manufac-

turing Review, 6(4):258–268, 1993.
[81] Daniel Whitney, Olivier Gilbert, and Marek Jastrzebski. Representation of geo-
metric variations using matrix transforms for statistical tolerance analysis. Research
in Engineering Design, 6(4):191–210, 1994.
[82] Daniel E. Whitney. Mechanical Assemblies. Oxford University Press, Oxford, En-
gland, 2004.
[83] Kyeonah Yu and Kenneth Goldberg. A complete algorithm for fixture loading.
International Journal of Robotics Research, 17(11):1214–1224, 1998.
Index
analysis, 1–10, 82 mechanism, 72

axis, 17 partition, 57–60, 67–69, 72, 75
region, 57, 72
boundary representation, 11, 178 surface, 67
topology, 62, 102, 113
CAD, 1, 10, 98, 134 vertex, 64
cam, 2, 29, 55, 75 contact, 1–10, 21, 23, 27, 33
centered three-arc, 140 boundary equations, 35, 44
intermittent, 146 change, 2, 8, 60
offset three-arc, 139 configuration, 32
six-arc, 141 curve, 37
three-finger bidirectional, 143 edge-edge, 54
three-finger unidirectional, 142 equations, 2, 33
camera shutter, 169 face-edge, 53
conceptual design, 5, 109 face-face, 52
configuration fixed-axis, 37
mechanism, 25 general boundary segments, 50
pair, 21 general planar, 45
part, 16 line segment-line segment, 36
variables, 17, 20 moving circle segment-fixed circle
configuration space, 17, 20, 55, 121 segment, 48
blocked, 57 moving circle segment-fixed line
cell, 69 segment, 46
contact, 57 moving line segment-fixed circle
cross-section, 67 segment, 47
curve, 58 multiple, 2, 58
edge, 64 part, 2, 57
face, 65 rotating circle-rotating circle, 42
fixed-axis pair, 60, 179 rotating circle-rotating line, 41
free, 57 rotating circle-translating circle,
general pair, 69 40
194 Index
contact (cont.) Geneva

rotating circle-translating line, 38, 41 planar, 122, 144, 181
segment-vertex, 36 planar inverse, 145
space, 57 spatial, 15, 61, 160
spatial, 52
subsumed, 36 HIPAIR, 10, 55, 79, 98, 108, 175
tangency curve, 37–43
tangency equations, 37 indexer, 3, 72
translating line-translating line, 38 disk, 82, 103, 149
translating circle-rotating line, 38 lever, 83, 150, 167
type, 33, 52 pawl, 73, 86, 164
vertex-vertex, 36 reciprocating, 158
zone, 101, 134 spatial, 159
zone computation, 103, 106 star, 148
conventions, 4, 137 interlock pair, 147, 151
coordinate system, 13, 16, 18, 22
counter mechanism, 163 kinematics
cross-section, 11 analysis, 81
design, 1–3, 5, 109
degrees of freedom, 20 fixed-axis pairs, 22, 25
Dennis clutch, 27, 29, 115 function, 1–10, 14, 74–75, 81
dragger, 112 higher pairs, 2, 24
dynamical simulation, 95 lower pairs, 2, 24
pairs, 2, 21
escapement, 96, 153 simulation, 81, 87–98, 178
synthesis, 109
fastener, 67 tolerancing, 100
feature, 9 variation, 99, 121
contact, 33
planar, 12 lever mechanism, 100, 107
simple, 33 linkage, 2, 27, 29, 75
spatial, 14
manipulation, 135, 175
gear, 2, 29, 75 manual transmission gearshift,
asynchronous, 128 123
double rack, 154 material conditions, 103
dwell, 156 mechanism, 1, 11, 25, 179
intermittent, 26, 102, 166 classification, 29
sector, 26, 76, 155 configuration, 25
sector double, 157 design, 1, 5, 8
spatial bevel, 162 fixed-axis, 27
spatial orthogonal, 161 kinematics, 1
worm, 26 planar, 25
Index 195
spatial, 26 tolerance
theory, 75 analysis, 100–103
topology, 27, 90 model, 99
motion nominal, 99
constraints, 1 parametric, 99
fixed-axis, 18 revision, 119
rotation, 18 space, 100
translation, 18 specifications, 99
movie film advance, 168 worst-case, 99
tolerancing, 1, 8, 99–108, 134
optical filter, 110, 121 topology graph, 27
torsional ratcheting actuator, 125
parametric
design, 5 visualization, 135, 175
model, 5, 99 volume representation, 11
optimization, 117
representation, 13
synthesis, 109
tolerancing, 99
part
acceleration, 95
configuration, 16
force, 94
orientation, 16
planar, 12, 177
position, 16
spatial, 14
torque, 94
velocity, 88, 90, 93
ratchet, 61, 125, 152

Reuleaux, 2, 32, 79
simulation, 8, 81, 87
solid modeling, 11
state, 88
sweep-line algorithm, 64
synthesis, 1, 110, 134
design change, 111
draggers, 115
maximal instances, 118
parametric, 114
structure change, 113

The Configuration Space Method For Kinematic Design of Mechanisms

Uploaded by

Copyright:

Available Formats

The Configuration Space Method For Kinematic Design of Mechanisms

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

The Configuration Space Method For Kinematic Design of Mechanisms

Uploaded by

Copyright:

Available Formats

What is the book about?

What is the book about?

What topics are covered in the book?

What topics are covered in the book?

The Conﬁguration Space Method for

Kinematic Design of Mechanisms

Elisha Sacks and Leo Joskowicz

Elisha Sacks and Leo Joskowicz

The MIT Press

For information about special quantity discounts, please email

Library of Congress Cataloging-in-Publication Data

Sacks, Elisha, 1958–

I dedicate this book to my daughters, Dana and Yael, to my parents, Esther

3.4 General Boundary Segments 50

4 Contact of Parts in Configuration Space 57

8 Case Studies 121

Appendix A Catalog of Mechanisms 137

This monograph presents the configuration space method for computer-

and modify the design or the manufacturing process to ensure correct

parts are studied in a geometric representation called configuration space.

1.1 Mechanisms and Kinematic Function

The kinematic function of a mechanism is determined by the shapes, con-

Kinematic design is an iterative process consisting of five steps (figure

1.2 Kinematic Design

Their configurations are specified with orientation parameters with respect

many contacts of parts, with complicated contact equations. Unintended

1.3 Content and Organization of the Book

The book contains two appendices. The first is a catalog of 30 representa-

There are numerous introductory and advanced textbooks on kinematic de-

Mechanisms consist of moving interacting parts. The geometry, motion,

2.1 Geometry of a Part

A part in a mechanism is a rigid solid with a fixed shape. Modeling of solids

2.1.1 Planar Parts

A cross-section consists of closed regions called components. A compo-

changing the parameter’s values. The parameters are chosen to reflect

2.1.2 Spatial Parts

endpoints: segment a1 a2 is a straight line, followed by a circle segment a2 a3

2.2 Configurations and Motions of Parts

This section describes the mathematics of the configurations and motions

2.2.1 Planar Parts

The variables xA , yA , and yA are called configuration variables. Configura-

When a part moves, its configuration changes continuously over a time

around its origin oA in figure 2.8. The part’s configuration ðxA ; yA ; yA Þ is

2.2.2 Spatial Parts

2.3 Kinematic Pairs

The basic functional element of a mechanism is the kinematic pair. A kine-

2.3.1 Representation of Configuration Space

ðxA ; yA ; yA ; xB ; yB ; yB Þ ¼ ð3; 2; 30 ; 7; 1:75; 45 Þ

Contacts between parts constrain their configurations by inducing func-

2.3.2 Lower and Higher Pairs

space is at most 3n-dimensional. In a spatial mechanism, each moving part

figurations, and contacts of parts. For example, the structural description of

contacts. Paths between nodes in the graph indicate interactions between

2.5 Classification of Mechanisms

It has long been recognized that the classification of mechanisms according

Lever mechanisms 1 159 130 29 102 57 94

Volume Total Linkage FA Both Others VC Spatial

Lever mechanisms 1 614 341 101 18 154 123 108

3.1 Simple Planar Feature Contact

½nbB ðtB oÞ sin yB þ ½nbB ðtB oÞ cos yB þ ra þ db ny yA

k4 ¼ 0:5½ðobB Þ 2 þ t 2 þ ðoaA Þ 2 ðrb G ra Þ 2 :

½obB ðp tÞ sin yB þ ½obB ðp tÞ cos yB þ p t ¼ k4