Robotica 1 3 Fotos2 PDF

FUNDAMENTAIS
OF ROBOTICS
Analysis
and Control
FUNDAMENTAIS
OF ROBOTICS
Analysis
and Control
7
Robotic Manipulation
The term robot can convey many different meanings in the mind of the reader, de-
pending on the context. In the treatment presented here, a robot will be taken to
mean an industrial robot, also called a robotic manipulator or a robotic arm. An ex-
ample of an industrial robot is shown in Fig. 1-1. This is an articulated robotic arm
and is roughly similar to a human arm. It can be modeled as a chain of rigid links
interconnected by flexible joints. The links correspond to such features of the human
anatomy as the chest, upper arm, and forearm, while the joints correspond to the
shoulder, elbow, and wrist. At the end of a robotic arm is an end-effector, also called
a tool, gripper, or hand. The tool often has two or more fingers that open and cióse.
To further characterize industrial robots, we begin by examining the role they
play in automation in general. This is followed by a discussion of robot classifica-
tions using a number of criteria including: drive technologies, work envelope ge-
ometries, and motion control methods. A brief summary of the most common appli-
cations of robots is then presented; this is followed by an examination of robot
design specifications. Chap. 1 concludes with a discussion of the use of notation and
a summary of the notational conventions adopted in the remainder of the text.
1-1 AUTOMATION AND ROBOTS
Mass-production assembly lines were first introduced at the beginning of the twenti-
eth century (1905) by the Ford Motor Company. Over the ensuing decades, special-
ized machines have been designed and developed for high-volume production of me-
chanical and electrical parts. However, when each yearly production cycle ends and
new models of the parts are to be introduced, the specialized machines have to be
shut down and the hardware retooled for the next generation of models. Since peri-
odic modification of the production hardware is required, this type of automation is
1
7
Robotic Manipulation
The term robot can convey many different meanings in the mind of the reader, de-
pending on the context. In the treatment presented here, a robot will be taken to
mean an industrial robot, also called a robotic manipulator or a robotic arm. An ex-
ample of an industrial robot is shown in Fig. 1-1. This is an articulated robotic arm
and is roughly similar to a human arm. It can be modeled as a chain of rigid links
interconnected by flexible joints. The links correspond to such features of the human
anatomy as the chest, upper arm, and forearm, while the joints correspond to the
shoulder, elbow, and wrist. At the end of a robotic arm is an end-effector, also called
a tool, gripper, or hand. The tool often has two or more fingers that open and cióse.
To further characterize industrial robots, we begin by examining the role they
play in automation in general. This is followed by a discussion of robot classifica-
tions using a number of criteria including: drive technologies, work envelope ge-
ometries, and motion control methods. A brief summary of the most common appli-
cations of robots is then presented; this is followed by an examination of robot
design specifications. Chap. 1 concludes with a discussion of the use of notation and
a summary of the notational conventions adopted in the remainder of the text.
1-1 AUTOMATION AND ROBOTS
Mass-production assembly lines were first introduced at the beginning of the twenti-
eth century (1905) by the Ford Motor Company. Over the ensuing decades, special-
ized machines have been designed and developed for high-volume production of me-
chanical and electrical parts. However, when each yearly production cycle ends and
new models of the parts are to be introduced, the specialized machines have to be
shut down and the hardware retooled for the next generation of models. Since peri-
odic modification of the production hardware is required, this type of automation is
1
referred to as hard automation. Here the machines and processes are often very
efficient, but they have limited flexibility.
More recently, the auto industry and other industries have introduced more
flexible forms of automation in the manufacturing cycle. Programmable mechanical
manipulators are now being used to perform such tasks as spot welding, spray paint-
ing, material handiing, and component assembly. Since computer-controlled me
chanical manipulators can be easily converted through software to do a variety of
7 • ‘ % «\
Figure 1-1 An industrial robot. (Courtesy of Intelledex, Inc., Corvallis, OR.)
tasks, they are referred to as examples of soft automation. A qualitative comparison

of the cost effectiveness of manual labor, hard automation, and soft automation as a
function of the production volume (Dorf, 1983) is summarized in Fig. 1-2.
It is evident that for very low production volumes, such as those occurring in
small batch processing, manual labor is the most cost-effective. As the production
volume increases, there comes a point V\ where robots become more cost-effective
than manual labor. As the production volume increases still further, it eventually
reaches a point Vi where hard automation surpasses both manual labor and robots in
cost-effectiveness. The curves in Fig. 1-2 are representative of general qualitative
trends, with the exact data dependent upon the characteristics of the unit being pro-
duced. As robots become more sophisticated and less expensive, the range of pro-
2 Robotic Manipulation Chap. 1

referred to as hard automation. Here the machines and processes are often very
efficient, but they have limited flexibility.
More recently, the auto industry and other industries have introduced more
flexible forms of automation in the manufacturing cycle. Programmable mechanical
manipulators are now being used to perform such tasks as spot welding, spray paint-
ing, material handiing, and component assembly. Since com puter-controlled me
chanical manipulators can be easily converted through software to do a variety of
Figure 1-1 An industrial robot. (Courtesy of Intelledex, Inc., Corvallis, OR.)
tasks, they are referred to as examples of soft automation. A qualitative comparison

of the cost effectiveness of manual labor, hard automation, and soft automation as a
function of the production volume (Dorf, 1983) is summarized in Fig. 1-2.
It is evident that for very low production volumes, such as those occurring in
small batch processing, manual labor is the most cost-effective. As the production
volume increases, there comes a point v\ where robots become more cost-effective
than manual labor. As the production volume increases still further, it eventually
reaches a point vi where hard automation surpasses both manual labor and robots in
cost-effectiveness. The curves in Fig. 1-2 are representative of general qualitative
trends, with the exact data dependent upon the characteristics of the unit being pro-
duced. As robots become more sophisticated and less expensive, the range of pro-
2 Robotic M anipularon Chap. 1

\
Unit cost
Figure 1-2 Relative cost-eífectiveness

Production volume of soft automation.
duction volumes [ui, U2] over which they are cost-effective continúes to expand at
both ends of the production spectrum.
In order to more clearly distinguish soft automation from hard automation, it is
useful to introduce a specific definition of a robot. A number of definitions have been
proposed over the years. However, as robotic technology continúes to evolve, any
definition proposed may need to be refined and updated before long. For the purpose
of the material presented in this text, the following definition is used:
Definition 1-1-1: Robot. A robot is a software-controllable mechanical de-

vice that uses sensors to guide one or more end-effectors through programmed mo-
tions in a workspace in order to manipúlate physical objects.
Contrary to popular notions about robots in the Science fiction literature (see,
for instance, Asimov, 1950), today’s industrial robots are not androids built to im-
personate humans. Indeed, most are not even capable of self-locomotion. However,
many of today’s robots are anthropomorphic in the sense that they are patterned af-
ter the human arm. Consequently, industrial robots are often referred to as robotic
arms or, more generally, as robotic manipulators.
1-2 ROBOT CLASSIFICATION
In order to refine the general notion of a robotic manipulator, it is helpful to classify

manipulators according to various criteria such as drive technologies, work envelope
geometries, and motion control methods.
1-2-1 Drive Technologies
One of the most fundamental classification schemes is based upon the source of
power used to drive the joints of the robot. The two most popular drive technologies
are electric and hydrauíic. Most robotic manipulators today use electric drives in the
form of either DC servomotors or DC stepper motors. However, when high-speed
See. 1-2 Robot Classification 3

manipulation of substantial loads is required, such as in molten Steel handiing or
auto body part handiing, hydraulic-drive robots are preferred. One serious drawback
of hydraulic-drive robots lies in their lack of cleanliness, a characteristic that is im-
portant for many assembly applications.
Both electric-drive robots and hydraulic-drive robots often use pneumatic tools
or end-effectors, particularly when the only gripping action required is a simple
open-close type of operation. An important characteristic of air-activated tools is
that they exhibit built-in compliance in grasping objects, since air is a compressible
fluid. This is in contrast to sensorless rigid mechanical grippers, which can easily
damage a delicate object by squeezing too hard.
1-2-2 Work-Envelope Geometries
The end-effector, or tool, of a robotic manipulator is typically mounted on a flange

or píate secured to the wrist of the robot. The gross work envelope of a robot is
defined as the locus of points in three-dimensional space that can be reached by the
wrist. We will refer to the axes of the first three joints of a robot as the major axes.
Roughly speaking, it is the major axes that are used to determine the position of the
wrist. The axes of the remaining joints, the minor axes, are used to establish the ori-
entation of the tool. As a consequence, the geometry of the work envelope is deter-
mined by the sequence of joints used for the first three axes. Six types of robot
joints are possible (Fu et al., 1987). However, only two basic types are commonly
used in industrial robots, and they are listed in Table 1-1.
TABLE 1-1 TYPES OF ROBOT JOINTS
Type Notation Symbol Description
Revolute R Rotary motion about an axis
Prismatic P Linear motion along an axis
Revolute joints (R) exhibit rotary motion about an axis. They are the most
common type of joint. The next most common type is a prismatic joint (P), which
exhibits sliding or linear motion along an axis. The particular combination of revo
lute and prismatic joints for the three major axes determines the geometry of the
work envelope, as summarized in Table 1-2. The list in Table 1-2 is not exhaustive,
since there are eight possibilities, but it is representative of the vast majority of com-
mercially available robots. As far as analysis of the motion of the arm is concerned,
prismatic joints tend to be simpler than revolute joints. Therefore the last column in
Table 1-2, which specifies the total number of revolute joints for the three major
axes, is a rough indication of the complexity of the arm.
For the simplest robot listed in Table 1-2, the three major axes are all pris
matic; the resulting notation for this configuration is PPP. This is characteristic of a
Cartesian-coordinate robot, also called a rectangular-coordínate robot. An example

TABLE 1-2 ROBOT WORK ENVELOPES BASED ON MAJOR AXES
Robot Axis 1 Axis 2 Axis 3 Total revolute
Cartesian P P P 0
Cylindrical R P P 1
Spherical R R P 2
SCARA R R P J2bt
Articulated R R R 3
P = prismatic, R = revolute.
of a Cartesian-coordinate robot is shown in Fig. 1-3. Note that the three sliding
joints correspond to moving the wrist up and down, in and out, and back and forth.
It is evident that the work envelope or work volume that this configuraron generales
is a rectangular box. When a Cartesian-coordinate robot is mounted from above in a
rectangular frame, it is referred to as a gantry robot.
Figure 1-3 Cartesian robot.
If the first joint of a Cartesian-coordinate robot is replaced with a revolute

joint (to form the configuration RPP), this produces a cylindrical-coordínate robot.
An example of a cylindrical-coordinate robot is shown in Fig. 1-4. The revolute
joint swings the arm back and forth about a vertical base axis. The prismatic joints
then move the wrist up and down along the vertical axis and in and out along a ra
dial axis. Since there will be some minimum radial position, the work envelope gen-
erated by this joint configuration is the volume between two vertical concentric
cylinders.
Figure 1-4 Cylindrical robot.
Sec. 1-2 Robot Classification 5

If the second joint of a cyündrical-coordinate robot is replaced with a revolute
joint (so that the configuration is then RRP), this produces a spherical-coordinate
robot. An example of a spherical-coordinate robot is shown in Fig. 1-5. Here the
first revolute joint swings the arm back and forth about a vertical base axis, while
the second revolute joint pitches the arm up and down about a horizontal shoulder
axis. The prismatic joint moves the wrist radially in and out. The work envelope
generated in this case is the volume between two concentric spheres. The spheres
are typically truncated from above, below, and behind by limits on the ranges of
travel of the joints.
Figure 1-5 Spherical robot.
Like a spherical-coordinate robot, a SCARA robot (Selective Compliance As-

sembly Robot Arm) also has two revolute joints and one prismatic joint (in the
configuration RRP) to position the wrist. However, for a SCARA robot the axes of
all three joints are vertical, as shown in Fig. 1-6. The first revolute joint swings the
arm back and forth about a base axis that can also be thought of as a vertical shoul
der axis. The second revolute joint swings the forearm back and forth about a verti
cal elbow axis. Thus the two revolute joints control motion in a horizontal plañe.
The vertical component of the motion is provided by the third joint, a prismatic
joint which slides the wrist up and down. The shape of a horizontal cross section of
the work envelope of a SCARA robot can be quite complex, depending upon the
limits on the ranges of travel for the first two axes.
LnJ
1 <r
7 Figure 1-6 SCARA robot
6 Robotic Manipulation Chap ‘

h a revolute When the last rem aining prismatic joint is replaced by a revolute joint (to
[‘Coordínate 1 yield the configuration RRR), this produces an articulated-coordínate robot An
5. Here the f articulated-coordínate robot is the dual of a Cartesian robot in the sense that all
axis, while three of the m ajor axes are revolute rather than prismatic. The articulated-coordinate
tal shoulder robot is the most anthropomorphic configuration; that is, it most closely resembles
rk envelope the anatomy of the human arm. Articulated robots are also called revolute ro
rhe spheres bots. An example of an articulated-coordinate robot is shown in Fig. 1-7. Here the
te ranges of first revolute joint swings the robot back and forth about a vertical base axis. The
second joint pitches the arm up and down about a horizontal shoulder axis, and the
third joint pitches the forearm up and down about a horizontal elbow axis. These
motions create a complex work envelope, with a side-view cross section typically
being crescent-shaped.
t.
pliance A s-
lint (in the
the axes of
swings the Figure 1-7 Articulated robot.
tical shoul-
3ut a verti-
1-2-3 Motion Control Methods
ntal plañe.
I prismatic
Another fundam ental classification criterion is the method used to control the move-
section of
ment of the end-effector or tool. The two basic types of movement are listed in
; upon the
Table 1-3. The first type is point-to-point motion, where the tool moves to a se-
quence of discrete points in the workspace. The path between the points is not ex-
plicitly controlled by the user. Point-to-point motion is useful for operations which
are discrete in nature. For example, spot welding is an application for which point-
to-point motion o f the tool is all that is required.
TABLE 1-3 TYPES OF ROBOT MOTION CONTROL
Control method Applications

p-
*
Point to point Spot welding
Pick-and-place
Loading and unloading
Continuous path Spray painting
Are welding
Gluing
►
Chap. 1
Sec. 1-2 Robot Classification 7
The other type of motion is continuous-path motion, sometimes called con-
trolled-path motion. Here the end-effector must follow a prescribed path in three-
dimensional space, and the speed of motion along the path may vary. This clearly
presents a more challenging control problem. Examples of applications for robots
with continuous-path motion control include paint spraying, are welding, and the
application of glue or sealant.
1-3 APPLICATIONS
Robotic applications often involve simple, tedious, repetitive tasks such as the load-
ing and unloading of machines. They also include tasks that must be performed in
harsh or unhealthy environments such as spray painting and the handiing of toxic
materials. A summary of robot applications (Brody, 1987) is displayed in Table 1-4.
Here the percentages listed in the second column represent shares of the overall
robot market in the United States for the year 1986. The size of the market in 1986,
excluding sepárate sales of visión systems, was $516 million (Brody, 1987).
TABLE 1-4 U. S. ROBOT MARKET (1986)
Application Percent
Material handiing 24.4

Spot welding 16.5
Are welding 14.5
Spray painting and finishing 12.4
Mechanical assembly 6.2
Electronic assembly 4.8
Material removal 4.5
Inspection and testing 2.9
Water jet cutting 2.7
Other 11.1
Traditional applications of material handiing, welding, and spray painting and

finishing continué to dominate. The market share of assembly applications, both me
chanical and electrical, has grown steadily over the past decade, and there is clearly
potential for more applications in this area. However, the general assembly problem
has turned out to be quite challenging. The assembly process can be modeled as a
sequence of carefully planned collisions between the manipulator and the objeets in
its workspace. The delicate motion control that is required for assembly tasks dic-
tates the use of feedback ffom external sensors. This allows the robot to adapt its
motion in order to compénsate for part tolerances and other uncertainties in the en-
vironment. Often customized fixtures and jigs are used to secure parts and present
them to the robot at known positions and orientations (Boyes, 1985). This has the
effect of reducing uncertainty, but it can be an expensive approach.
The world population of installed robots as of the end of 1984 (Wolovich,
1987) was reported to be approximately 68,000. The country that has the largest
population of industrial robots is Japan; it is followed by the United States, West
Germany, and Sweden, as can be seen in Table 1-5. If the robot population figures
in Table 1-5 are normalized by the human population of the country, the leading
country in per capita robot population is Sweden.

TABLE 1-5 DISTRIBUTION
OF WORLD ROBOT POPULATION (1984)
Country Percent
Japan 44.1
USA. 22.1
West Germany 8.8
Sweden 7.4
France 2.9
Great Britain 2.9
Italy 2.2
Others 9.6
1-4 ROBOT SPECIFICATIONS
W hile the drive technologies, work-envelope geom etries, and motion control m eth
ods provide convenient ways to broadly classify robots, there are a number o f addi
tional characteristics that allow the user to further specify robotic m anipulators
Some o f the m ore common characteristics are listed in Table 1-6.
TABLE 1-6 ROBOT CHARACTERISTICS
Characteristic Units
Number o f axes
Load carrying capacity kg
Máximum speed, cycle time mm/sec
Reach and stroke mm
Tool orientation deg
Repeatability mm
Precisión and accuracy mm
Operating environment —
1-4-1 Number of Axes
Each robotic manipulator has a number of axes about which its links rotate or along
which its links transíate. Usually, the first three axes, or major axes, are used to es-
tablish the position of the wrist, while the remaining axes are used to establish the
orientation of the tool or gripper, as shown in Table 1-7. Since robotic manipulation
is done in three-dimensional space, a six-axis robot is a general manipulator in the
sense that it can move its tool or hand to both an arbitrary position and an arbitrary
orientation within its workspace. The mechanism for opening and closing the fingers
TABLE 1-7 AXES OF A ROBOTIC MANIPULATOR
Axes TVpe Function
1 -3 Major Position the wrist

4 -6 Minor Orient the tool
1-n Redundant Avoid obstacles
Sec. 1-4 Robot Specifications 9

or otherwise activating the tool is not regarded as an independent axis, because it
does not con tribute to either the position or the orientation of the tool. Practical in
dustrial robots typically have from four to six axes. Of course, it is possible to have
manipulators with more than six axes. The redundant axes can be useful for such
things as reaching around obstacles in the workspace or avoiding undesirable geo-
metrical configurations of the manipulator.
1-4-2 Capacity and Speed
Load-carrying capacity varies greatly between robots. For example, the Minimover
5 Microbot, an educational table-top robot, has a load-carrying capacity of 2.2 kg.
At the other end of the spectrum, the GCA-XR6 Extended Reach industrial robot has
a load-carrying capacity of 4928 kg (Roth, 1983-1984). The máximum tool-tip
speed can also vary substantially between manipulators. The Westinghouse Series
4000 robot has a tool-tip speed of 92 mm/sec, while the Adept One SCARA robot
has a tool-tip speed of 9000 mm/sec (Roth, 1983-1984). A more meaningful mea-
sure of robot speed may be the cycle time, the time required to perform a periodic
motion similar to a simple pick-and-place operation. The Adept One SCARA robot
carrying a 2.2-kg payload along a 700-mm path that consists of six straight-line seg-
ments has a cycle time of 0.9 sec. Thus the average speed over a cycle is 778
mm/sec, considerably less than the 9000 mm/sec máximum tool-tip speed.
Although the load-carrying capacities and máximum operating speeds of ro
bots vary by several orders of magnitude, it is, of course, the mix of characteristics
that is important when selecting a robot for a particular application. In some cases, a
large load-carrying capacity may not be necessary, while in other cases accuracy
may be more important than speed. Clearly, there is no point in paying for addi-
tional characteristics that are not relevant to the class of applications for which the
robot is intended.
1-4-3 Reach and Stroke
The reach and the stroke of a robotic manipulator are rough measures of the size of
the work envelope. The horizontal reach is defined as the máximum radial distance
the wrist mounting flange can be positioned from the vertical axis about which the
robot rotates. The horizontal stroke represents the total radial distance the wrist can
travel. Thus the horizontal reach minus the horizontal stroke represen ts the mini-
mum radial distance the wrist can be positioned from the base axis. Since this dis
tance is nonnegative, we have:
Stroke ^ reach (1-4-1)
For example, the horizontal reach of a cylindrical-coordinate robot is the ra-
dius of the outer cylinder of the work envelope, while the horizontal stroke is the
difference between the radii of the concentric outer and the inner cylinders, as
shown in Fig. 1-8.
The vertical reach of a robot is the máximum elevation above the work surface
that the wrist mounting flange can reach. Similarly, the vertical stroke is the total
vertical distance that the wrist can travel. Again, the vertical stroke is less than or
10 Robotic Manipulation Chap. I

Figure 1-8 Reach and stroke of a cylindrical robot.
equal to the vertical reach. For example, the vertical reach of a cylindrical robot will
be larger than the vertical stroke if the range of travel of the second axis does not
allow the wrist to reach the work surface, as shown in Fig. 1-8. One of the useful
characteristics of articulated robots lies in the fact that they often havefu ll work en-
velopes, in the sense that the stroke equals the reach. However, this feature gives
rise to a need for programming safeguards, because an articulated robot can be pro-
grammed to collide with itself or the work surface.
1-4-4 Tool Orientation
W hile the three major axes of a robot determine the shape of the work envelope, the
remaining axes determine the kinds of orientation that the tool or hand can assume.
If three independent minor axes are available, then arbitrary orientations in a three -
dimensional workspace can be obtained. A number of conventions are used in the
robotics literature to specify tool orientation (Fu et al., 1987). The tool orientation
convention that will be used here is the yaw-pitch-roll (YPR) system. Yaw, pitch,
and roll angles have long been used in the aeronautics industry to specify the ori
entation of aircraft. They can also be used to specify tool orientation, as shown in
Fig. 1-9.
Figure 1-9 Yaw, pitch, and roll of

tool.

To specify tool orientation, a mobile tool coordínate fram e M — { m \ m 2, m3}
is attached to the tool and moves with the tool, as shown in Fig. 1-9. Here m 3 is
aligned with the principal axis of the tool and points away from the wrist. Next, m 1
is parallel to the line followed by the fingertips of the tool as it opens and closes.
Finally, m 1 completes the right-handed tool coordínate frame M.
By convention, the yaw, pitch, and roll motions are performed in a specific or
der about a set of fixed axes. Initially, the mobile tool frame M starts out coincident
with a fixed wrist coordínate frame F = i / 1, / 2, / 3} attached at the end of the fore-
arm. First the yaw motion is performed by rotating the tool about wrist axis f \
Next, the pitch motion is performed by rotating the tool about wrist a x i s / 2. Finally,
the roll motion is performed by rotating the tool about wrist axis / 3. In each case
positive angles correspond to counterclockwise rotations as seen looking along the
axis back toward the origin.
The order in which the yaw, pitch, and roll motions are performed is signif-
icant because it affects the final orientation of the tool. For example, a yaw of 7t / 2
followed by a pitch of 7 t / 2 yields a different final tool orientation from the orienta
tion produced by a pitch of tt/2 followed by a yaw of 7t / 2 . Thus, implicit in the
YPR convention is the order of rotations summarized in Table 1-8.
TABLE 1-8 YAW, PITCH, AND ROLL MOTION
Operation Description Axis
1 Yaw f l
2 Pitch P
3 Roll P
An alternative way to specify the YPR motions is to instead perform the rota
tions in reverse order about the axes of the mobile tool frame M rather than the fixed
wrist ffame F. That is, first a roll motion is performed about m 3, then a pitch motion
is performed about m2, and finally a yaw motion is performed about m l. This is
equivalent to performing the rotations about the axes of the fixed wrist frame F in
the original YPR order, in the sense that the tool ends up at the same orientation.
For this reason, the YPR system is often referred to as the RPY system. A sequence
of rotations about the mobile frame M axes is often easier to visualize, particularly
when the angles are not múltiples of tt/2 .
Definition 1-4-1: Spherical W rist, A robot has a spherical wrist if and only
if the axes used to orient the tool intersect at a point.
The robot wrist shown in Fig. 1-9 is an example of a spherical wrist because
the three axes used to orient the tool, {m1, m2, m 3}, intersect at a point. More gen-
erally, if a robot possesses fewer than three axes to orient the tool, then it can still
have a spherical wrist. However, in these cases the spherical wrist will not be a gen
eral spherical wrist, because, with fewer than three axes, some orientations of the
tool cannot be achieved. Spherical wrists are useful because they simplify the mathe-
matical descriptions of robotic manipulators. In particular, the larger n-axis problem
can be partitioned into a two simpler problems, a 3-axis problem and an (n — 3)-

axis problem (Pieper, 1968). Many industrial robots are designed with spherical
wrists.
1-4-5 Repeatability, Precisión, and Accuracy
Repeatability is another important design characteristic. Repeatability is a measure

of the ability of the robot to position the tool tip in the same place repeatedly. On
account of such things as backlash in the gears and flexibility in the links, there will
always be some repeatability error, perhaps on the order of a small fraction of a mil-
limeter for a well-designed manipulator (Groover et al., 1986).
Closely related to the notion of repeatability is the concept of precisión. The
precisión of a robotic manipulator is a measure of the spatial resolution with which
the tool can be positioned within the work envelope. For example, if the tool tip is
positioned at point A in Fig. 1-10 and the next closest position that it can be moved
to is B, then the precisión along that dimensión is the distance between A and B.
Precisión Adjacent
tool
A B position
V
A
V
Accuracy Figure 1-10 Adjacent tool positions.
More generally, in three-dimensional space the robot tool tip might be posi
tioned anywhere on a three-dimensional grid of points within the workspace. Typi-
cally, this grid or lattice of points where the tool tip can be placed is not uniform.
The overall precisión of the robot is the máximum distance between neighboring
points in the grid. For example, an interior grid point of a Cartesian-coordinate
robot will have eight neighbors in the horizontal plañe plus nine neighboring grid
points in the planes above and below. One of the virtues of a Cartesian-coordinate
robot is that the precisión is uniform throughout the work envelope; the state of one
axis does not influence the precisión along another axis. However, this is not the
case for other types of robots, as can be seen in Table 1-9, where the overall preci
sión is decomposed into horizontal and vertical componen ts.
For the case of a cylindrical-coordinate robot, the horizontal precisión is
highest along the inside surface of the work envelope and lowest along the outside
surface. This is because the are length swept out by an incremental movement A<£ in
TABLE 1-9 HORIZONTAL AND VERTICAL PRECISION
Robot Horizontal Vertical

Type Precisión Precisión
Cartesian Uniform Uniform

Cylindrical Decreases radially Uniform
Spherical Decreases radially Decreases radially
SCARA Varies Uniform
Articulated Varies Varies

the base joint is proportional to the radial position r, as can be seen in Fig. 1-11.
Note how the shape of a horizontal grid element is not square and grid elements near
the outside surface of the work envelope are larger than grid elements near the inside
surface of the work envelope.
r A0
Figure 1-11 Horizontal precisión of a

cylindrical robot.
Here the radial precisión is Ar, while the angular precisión about the base ex-
pressed in terms of the are length is r A</>. In the worst case, the angular precisión is
R A<¿>, where R is the horizontal reach of the robot. Thus the overall horizontal pre
cisión in this case is approximately
A/i « [(Ar)2 + (R A<¿>)2]1/2 (1-4-2)
Here we have assumed that A</>and Ar are sufficiently small and that the angle bac
in Fig. 1-11 is approximately tt/2 . If Az represents the vertical precisión of a cylin
drical-coordinate robot, then the total robot precisión is
Ap ~ [(Ar)2 + (R A<f))2 + (Az)2]1/2 (1-4-3)
For a spherical-coordinate robot, both the horizontal precisión and the vertical
precisión are highest along the inside surface and lowest along the outside surface of
the work envelope. For SCARA robots, the horizontal precisión varies, but the ver
tical precisión is uniform. It is this feature, together with its open work envelope,
that makes the SCARA robot useful for light assembly work such as the vertical in-
sertion of electrical components into circuit boards. With articulated robots, both
the horizontal precisión and the vertical precisión vary over the work envelope.
An alternative way to specify the precisión of a robot is to specify the precisión
of the individual joints. Clearly this is easier to analyze, although it is perhaps less
useful. The precisión or spatial resolution of an individual joint is affected by the
drive system, the position sensors, and the power transmission mechanism, including
gears, sprockets, and cables. For example, suppose the reference position signal is
generated internally in a control Computer and then sent to an external analog feed-
back position control system through an n-bit digital-to-analog converter (DAC). If
the range of load shaft positions attainable goes from </>min through (¿w , then the
load shaft precisión AÓdac in this case is given by:
a i W
'ta
kx Ymin . ,, & .\
A<Pdac = ------ —------ degrees per count (1-4-4)

On many robots, incremental encoders are used to determine shaft position
rather than absolute encoders such as potentiometers. Incremental encoders typically
consist of a slotted disk with infrared emitter-detector pairs. As the shaft turns, the
infrared beams are either interrupted or not, depending upon the position of the slot
ted disk. If two or more emitter-detectors are used, then both the direction and the
magnitude of the rotary motion can be determined. This can be seen in Fig. 1-12,
where a slotted disk with a pair of slots arranged in phase quadrature is displayed.
Here a broken beam is represented by a 1, while an unbroken beam is represented
by a 0. Note that there are four possible States. Since only one bit changes between
adjacent States, the overlapping slots generate what is called a gray code. If the en-
coder output is sampled sufficiently fast, then the direction of the movement can be
inferred by comparing the oíd State with the new State. The counter which stores the
accumulated shaft position is then either incremented or decremented as appropriate.
Figure 1-12 Slotted disk with 2-bit

gray code.
More generally, if d emitter-detector pairs are used, and if there are k slots
around the circumference of the disk, then the precisión with which the shaft posi
tion can be measured is 360/{kl*) degrees per count. The encoded disk is usually
mounted on the high-speed shaft of the motor, rather than the load shaft. If the gear
reduction ratio from the high-speed shaft to the load shaft is m : 1, then the overall
incremental load shaft precisión A<frnc is
A<£mc = , —d degrees per count (1-4-5)
The accuracy of a robotic manipulator is a measure of the ability of the robot

to place the tool tip at an arbitrarily prescribed location in the work envelope.
Robotic arm accuracy is not nearly as easy to analyze as precisión. Accuracy de-
pends, for example, on such things as gear backlash and elastic deformations in the
links due to loading. However, a simple bound can be placed on the error or inaccu-
raey, in terms of the precisión, namely:
_ precisión
Error > E— ----- (1-4-6)
Sec. 1-4 Robot Specifications 491482 15

Clearly, if the precisión of the robot dictates that only a discrete grid of points
can be visited within the work envelope, then the accuracy with which an arbitrary
point can be visited is, at best, half of the grid size, as shown in Fig. 1-10. In gen
eral, the accuracy will not be that good. For example, the robot may gradually drift
out of calibration with prolonged operation. To maintain reasonable accuracy, robots
are periodically reset or zeroed to a standard hard home position using limit switch
sensors.
Although robotic manipulators are not nearly as accurate as their more expen-
sive and more specialized hard automation counterparts, they do provide increased
flexibility. It is one of the great challenges of robotics to make use of this increased
flexibility to compénsate for less than perfect accuracy. In particular, clever al-
gorithms and control strategies are needed that make use of external sensors to com
pénsate for uncertainties in the environment and the uncertainties in the exact posi
tion of the robot itself.
1-4-6 Operating Environment
The operating environments within which robots ftmction depend on the nature of
the tasks performed. Often robots are used to perform jobs in harsh, dangerous, or
unhealthy environments. Examples include the transport of radioactive materials,
spray painting, welding, and the loading and unloading of furnaces. For these appli
cations the robot must be specifically designed to opérate in extreme temperatures
and contaminated air. In the case of paint spraying, a robotic arm might be clothed
in a shroud in order to minimize the contamination of its joints by the airborne paint
particles.
At the other extreme of operating environments are clean room robots. Clean
room robots are used in the semiconductor manufacturing industry for such tasks as
the handiing of Silicon wafers and photomasks. Specialized equipment has to be
loaded and unloaded at the various clean room processing stations. Here the robot is
operating in an environment in which parameters such as temperature, humidity, and
airflow are carefully controlled. In this case it is the robot itself that must be de
signed to be very clean so as to minimize the contamination of work pieces by sub-
micron particles. Some clean room robots are evacuated internally with suction in
order to scavange particles generated by ffiction surfaces. Others use special non-
shedding materials and employ magnetic washers to hold ferromagnetic lubricant in
place.
1-4-7 An Example: Rhino XR-3
In recent years a number of educational table-top robots have appeared in the mar-
ketplace. Many educational robots employ open-loop position control using DC
stepper motors. Some table-top educational robots are also available which use
closed-loop position control with DC servomotors and incremental encoders. An ex
ample of the latter type of educational robot is the Rhino XR-3 robot, pictured in
Fig. 1-13.

Figure 1-13 An educational robot. (Courtesy of Rhino Robots, Inc., Champaign,
IL. Robotic arm developed and manufactured in the U.S.A.)
In term s o f broad classification, the Rhino XR-3 robotic manipulator is a five-

axis electric-drive articulated-coordinate robot with point-to-point motion control
(H endrickson and Sandhu, 1986). The specifications of the Rhino XR-3 are summa-
rized in Table 1-10. The load-carrying capacity, máximum tool-tip speed, and re
peatability are rough estimates based on use of this robot in an instructional labora-
tory. The reach and stroke specifications are measured with respect to the tool tip
assum ing that standard fingers are used for the tool.
A nother usefiil specification for the Rhino robot is the load shaft precisión for
each o f the five axes. Each axis is driven by a 12-V DC servomotor that has an in-
TABLE 1-10 RHINO XR-3 SPECIFICATIONS
Characteristic Valué Units
Number of axes 5 —
Load-carrying capacity 0.5 kg
Máximum tool-tip speed 650.0 mm/sec
Horizontal reach and stroke 628.6 mm
Vertical reach and stroke 889.0 mm
Tool pitch range 270.0 deg
Tool roll range 360.0 deg
Repeatability ±2.5 mm

Figure 1-13 An educational robot. (Courtesy of Rhino Robots, Inc., Champaign,
IL. Robotic arm developed and manufactured in the U.S.A .)
In terms of broad classification, the Rhino XR-3 robotic manipulator is a five-

axis electric-drive articulated-coordinate robot with point-to-point motion control
(Hendrickson and Sandhu, 1986). The specifications of the Rhino XR-3 are summa-
rized in Table 1-10. The load-carrying capacity, máximum tool-tip speed, and re
peatability are rough estimates based on use of this robot in an instructional labora-
tory. The reach and stroke specifications are measured with respect to the tool tip
assuming that standard fingers are used for the tool.
A nother useful specification for the Rhino robot is the load shaft precisión for
each o f the five axes. Each axis is driven by a 12-V DC servomotor that has an in-
TABLE 1-10 RHINO XR-3 SPECIFICATIONS
Characteristic Valué Units
Number of axes 5 ——
Load-carrying capacity 0.5 kg
Máximum tool-tip speed 650.0 mm/sec
Horizontal reach and stroke 628.6 mm
Vertical reach and stroke 889.0 mm
Tool pitch range 270.0 deg
Tool roll range 360.0 deg
Repeatability ± 2 .5 mm

cremental encoder attached to the high-speed shaft. The encoders resolve the high-
speed position to 60 degrees. Since each motor has a built-in gear head with a tums
ratio of 66.6 : 1, this results in a precisión for each load shaft of 0.901 degrees per
count. Finally, each joint has its own set of sprockets and chains which then deter
mine the joint angle precisión. The resulting precisión in degrees per count for each
joint and for the tool is summarized in Table 1-11.
TABLE 1-11 LOAD SHAFT PRECISION
OF THE RHINO XR-3 ROBOT
Motor Joint Precisión
F Base 0.2269
E Shoulder 0.1135
D Elbow 0.1135
C Tool pitch 0.1135
B Tool roll 0.3125
A Tool closure 1.2500
Units of precisión: degrees per count.
The Rhino XR-3 robot is one of several robots that are used as vehicles for il-
lustrating theoretical concepts covered in the remainder of this text. Direct and in-
verse kinematics, work-envelope calculations, trajectory planning, and control are
examined. A series of laboratory projects designed around a robotic work cell which
features a Rhino XR-3 robot and a Micron Eye camera are included in a sepárate
Laboratory Manual that accompanies this text (Schilling and White, 1990, in press).
These laboratory programming projects are designed to play an integral part in any
course using this text and are strongly recommended. If a Rhino XR-3 or other edu
cational robot is not available, a three-dimensional graphics robot simulator program
called SIMULATR that is supplied with the Laboratory Manual can be used as a
substitute. The use of the graphics simulator for development of robot control pro-
grams is highly recommended even when a physical robot is available. This way
robot control programs can be conveniently developed and debugged off-line with
the final versión of the program tested on an actual robot.
1-5 NOTATION
Robotics is a broad interdisciplinary field that encompasses a variety of traditional

topics from electrical engineering, mechanical engineering, and Computer science.
This breadth makes the field quite exciting, but at the same time it poses a challenge
when it comes to developing a clear and consistent notation. Since notation and ter-
minology play an important role in any detailed treatment of robotics, we pause at
this point to discuss and summarize the notational conventions that will be used in
the remainder of this text.
Scalars, vectors, matrices, and sets are used to simplify the presentation and
make it more concise. To distinguish vectors and matrices fforn scalars, boldface
print is traditionally employed. Although this approach is effective for printed mate
rial, the use of boldface characters is, at best, inconvenient when instructors are
18 Robotic Manipulation Chap-1

w riting on a blackboard or when students are taking notes in a class notebook. One
com m on alternative for these media is the underlining convention whereby vectors
are underlined once and matrices twice. However, repeated underlining becomes
very cumbersome when vectors and matrices are used extensively.
The general convention employed here is to instead regard all mathematical
quantities as either column vectors, matrices, or sets. Column vectors are denoted
w ith lowercase letters, both English and Greek, while matrices and sets are denoted
w ith uppercase letters. Single subscripts denote scalar components of column vec
tors, whereas double subscripts denote scalar components of matrices. In the case of
m atrices, the first subscript is the index of the row, while the second is the index of
the column. These general notational conventions are summarized in Table 1-12.
TABLE 1-12 NOTATIONAL CONVENTIONS
Entity Notation Examples
Scalars Subscripted a ,, o2, a i. «2, A n , A 12

Column vectors Lowercase a , b , c, a , /3, y , x \ x 2
Matrices, sets Uppercase a, b, c , d , r, íl, ¥ , Y
Note that scalars can be regarded as important special cases of column vectors,
that is, column vectors with only one component. Similarly, vectors are themselves
special cases of matrices, that is, matrices with only one row or only one column.
Vectors. A column vector is represented by arranging its components in an

n x 1 array and enclosing them in square brackets as follows:
X\
x = X2 (1-5-1)
*3
The superscript T is reserved to denote the transpose of a vector or, more gen-
erally, of a matrix. The transpose is obtained by interchanging the rows with the
columns. Thus the transpose of an n x 1 column vector is a 1 x n row vector, and
conversely. The following is therefore a space-saving way of writing a column vec
tor as a transposed row vector:
x = [x\, x2, x 3]r (1-5-2)
Individual superscripts are used as Índices to distinguish members of a set of
vectors. In those cases where a superscript might be confused with taking a quantity
to a power, parentheses are used to remove the ambiguity. For example (jti)2 is used
to denote the square of the first component of the vector jc, while jc2 denotes the first
com ponent of the second member of a set of vectors {jc1, jc2, . . . }.
Sets. While square brackets are reserved for enclosing components of vec
tors and matrices, curly brackets, or braces, are used to endose members of a set.
Thus a set T of vectors jc1, jc2, . . . , jc" is represented as follows:
Sec. 1-5 Notation 19

We often need to specify a set of vectors satisfying a certain property P—for
example, the set of all Solutions to a particular equation or perhaps the set of all vec
tors whose first component is zero. In these cases the set of all vectors belonging to
subset U that also satisfy property P is represented as follows:
r = { jc E U: P ( jc) } . (1-5-4)
Here E is the set membership symbol, and P(x) is a property of x. For example,
P (x) might represent one or more equality or inequality constraints involving x or its
components.
Matrices. Just as a vector can be represented as a one-dimensional array of

scalar components, as in Eq. (1-5-1), a matrix can be represented by a two-dimen-
sional array of scalar components as follows:
It is also useful to represent a matrix in terms of its columns. The convention

used here is that the corresponding lowercase letter with a superscript index is used
to denote a column of a matrix. Thus aj denotes the jth column of the matrix A, in
which case AkJ = ai, for 1 < k < m and 1 < j < n. The en tire m x n matrix A
can be written in terms of its n columns as follows:
A = [a1, a 2, . . . y a n] (1-5-6)
The matrix A can be viewed as a linear transformation or mapping from the
vector space Rn of column vectors with n real components to the vector space Rmof
column vectors with m real components. Every m x n matrix A has a nuil space
N (A) and a range space R(A) associated with it; these are defined as follows:
N(A) = {x E R": A* = 0} (1-5-7)
R{A) = {Ax E Rm: jc E Rn} (1-5-8)
The nuil space of A is the subspace of R” consisting of the set of all vectors
x E R” which are mapped into the zero vector in Rmby the transformation A. Thus
the nuil space is the inverse image of the zero vector in Rmunder the transformation
A. Similarly, the range space of A is the subspace of Rmconsisting of all vectors y
such that y = Ax for at least one x E R". Thus the range space R(A)is the image of
R" under the transformation A. A pictorial representation of the nuil and range
spaces of a matrix A is shown in Fig. 1-14.
The dimensión of N(A) is called the nullity of A, and the dimensión of R(A) is
called the rank of A. One of the fundamental results of linear algebra is that the di
mensión of the nuil space plus the dimensión of the range space equals the dimensión
of the space over which the transformation A is defined. Therefore:
nullity (A) + rank (A) = n (1-5-9)

Rn R m
Figure 1-14 The range and nuil space of an m x n matrix A.
The rank of A is equal to the number of linearly independent rows of A y and it

is also equal to the number of linearly independent columns of A. If the rank of the
matrix A is as large as possible, then A is said to be of full rank. A square matrix
(m = n) is invertibie if and only if it is of full rank. We denote the inverse of A as
A~\
Certain special matrices have symbols reserved for them. These include the
zero matrix, which is denoted by 0, and the identity matrix, which is denoted by /.
The zero matrix is simply a matrix of Os, while the identity matrix is a square matrix
that has Os everywhere except along the principal diagonal, where it has ls.
Coordinate transformations. To model the motion of a robotic arm, coor-

dinate frames are attached to various parts of the arm and to sensors and objects in
the workspace. In particular, the following right-handed orthonormal coordinate
frames are attached to the links of the robot, starting with link 0 , which is the base,
and ending with link n , which is the tool:
Lk = {jc*, y k, z*}, 0 < k < n (1-5-10)
It is often necessary to transform or map the coordinates of a point with re-
spect to one frame into its coordinates with respect to another frame. The matrix T
is reserved to denote a general coordinate transformation matrix. Typically, T will
have a superscript index or identifier which specifies the source coordinate frame and
a subscript index or identifier which specifies the destination or reference coordinate
frame. For example, T K denotes the coordinate transformation matrix which trans-
forms robot tool coordinates into robot base coordinates. More generally, Tj trans-
forms frame L k coordinates into frame Lj coordinates.
Some authors of robotics texts use either presuperscripts or presubscripts, or
both, such as *7} and KéT, for coordinate transformation matrices (Paul, 1981;
Craig, 1986). We will refrain from using any presuperscripts or presubscripts, in an
effort to simplify the notation somewhat. The price we pay is that in some cases
there may be ambiguity as to whether or not a superscript should be interpreted as
taking a matrix to a power. In these cases we again use parentheses to remove the
ambiguity. Thus (Ti)2 is the square of the matrix Ti, while T\ is the matrix which
transforms frame L2 coordinates into frame L\ coordinates.
Sec. 1-5 Notation 21

Special symbols. A number of special notational symbols are used
throughout the text. In each case they are defined in the text where they first appear
unless they are common, widely accepted symbols. In any event, a list of symbols
can be found in Appendix 3. This list contains brief descriptions of special symbols
used repeatedly throughout the text. Symbols which are used only infrequently, such
as for a temporary variable in a derivation or proof, are not included in this list.
Simple trigonometric expressions appear repeatedly in the analysis of robotic
arms. Expressions involving sums and products of sines and cosines of joint angles
can be shortened and simplified if a standard trigonometric shorthand is employed
as summarized in Table 1-13.
TABLE 1-13 TRIGONOMETRIC *

SHORTHAND
Symbol Meaning
C <f> COS (f>

s<t> sin <¿>
V<f> 1 — eos 4>
c* eos dk
s* sin dk
Q, eos {6k + dj)
Siy sin {dk + dj)
C k-j eos {dk - dj)
Sk -j sin {dk - dj)
Here 0* denotes the joint angle for the kth joint of a robotic arm.Clearly the
notation can be extended in the obvious way to include sums or differences of three
or more joint angles. The joint angle 0* is often replaced by the variable qk, which
denotes the joint variable for the kth joint. That is, qk and 0* can be used inter-
changeably in Table 1-13.
1-6 PROBLEMS
1-1. What is the essential feature that distinguishes soft automation from hard automation?
1-2. Which types of drive technologies would be most suitable for the following applica
tions:
(a) Unload 1000-kg items from a die casting machine
(b) Install integrated circuit chips on a printed circuit board
1-3. What is the minimum number of axes required for a robot to insert and tighten four
nuts (from above) on four vertical bolts as shown in Fig. 1-15? The nuts are available
from a vertical spring-activated parts feeder. Explain why each axis is needed.
1-4. Suppose a cylindrical-coordinate robot has a vertical reach of 480 mm and a vertical
stroke of 300 mm. How far off the floor do parts have to be raised in order to be reach-
able by the robot?
1-5. The base joint of a cylindrical-coordinate robot is driven by a 12-bit digital-to-analog

• •
T
R obot • •
Bolts
N u t feeder Figure 1-15 Nut tightening task.
converter (DAC) and has a swing range of 360 degrees. The radial axis is driven by an
8-bit DAC and has a horizontal reach of 300 mm and a horizontal stroke of 200 mm.
Finally, the vertical axis is driven by a 10-bit DAC and has a vertical reach of 480 mm
and a vertical stroke of 360 mm.
(a) What is the volume of the work envelope?
(b ) What is the vertical precisión?
(c) What is the radial precisión?
(d ) What is the angular precisión about the base?
(e) What is the horizontal precisión?
( f ) What is the total precisión?
1-6. For a spherical-coordinate robot, where within the work envelope is the tool-tip preci
sión the highest?
1-7. For what type of robot is the precisión uniform throughout the work envelope? For
which robots is the vertical precisión uniform?
1-8. An incremental shaft encoder with 2 emitter-detector pairs and 12 slots around the
circumference is used to monitor the angular position of a high-speed motor shaft.The
precisión of the load shaft is measured and found to be 0.05 degrees per count. What is
the gear ratio between the high-speed shaft and the loadshaft?
1-9. Consider the robotic tool shown in Fig. 1-16. Sketch the tool position after each inter-
mediate position of the following YPR operation: yaw 7t / 2 , pitch — 7t / 2 , roll 7 r / 2 .
rotations are taken about the axes of the mobile frame M rather than the fixed frame F,
then the final tool position is the same. Are the intermedíate tool positions the same?
Sec. 1-6 Problem s 23

REFERENCES
A s im o v , I. (1950). /, Robot, Fawcett Publications: Greenwich, Conn.

B oyes , W. E. (1 9 8 5 ). Low-Cost Jigs, Fixtures, and Gages for Limited Production, SME Pub
lications: Dearborn, Mich.
B r o d y , H . (1987). “ U .S . ro b o t m a k e rs try to b o u n c e b a c k ,” High Technology Business, Oc-
to b e r , pp. 18-24.
C r a ig , J. J. (1986). Introduction to Robotics: Mechanics & Control, Addison-Wesley: Read-
ing, Mass.
D o r f , R. C. (1983). Robotics and Automated Manufacturing, Reston: Reston, Va.
Fu, K. S., R. C. G o n z á l e z , a n d C . S. G. L ee (1987). Robotics: Control, Sensing, Vision, and
Intelligence, McGraw-Hill: New York.
G r o o v e r , M. P., M . W eiss, R . N. N a g e l, a n d N. G. O drey (1986). Industrial Robotics, Mc
Graw-Hill: New York.
H e n d r ic k s o n , T., a n d H. S a n d h u (1986). XR-3 Robot Arm, Mark III 8 Axis Controller:
Owner’s Manual, Versión 3.00, Rhino Robots, Inc.: Champaign, 111.
P a u l , R. (1981). Robot Manipulators: Mathematics, Programming, and Control, MIT Press:
Cambridge, Mass.
P ie p e r , D. L. (1968). “The kinematics of manipulators under Computer control,’’ AIM 72,
Stanford AI Laboratory, Stanford Univ.
R obot I nstitu te o f A m er ic a (1982). Worldwide Robotics Survey and Directory, Dearborn,
Mich.
Roth, B. (1983-1984). “A table of contemporary manipulator devices,’’ Robotics Age, July
1983, pp. 38-39; September 1983, pp. 30-31, January 1984, pp. 34—35.
S c m l l in g , R. J., a n d R. W h it e (1990, in press). Robotic Manipulation: A Laboratory Man
ual, Prentice Hall, Englewood Cliffs, N.J.
W o l o v ic h , W. A. (1987). Robotics: Basic Analysis and Design, Holt, Rinehart & Winston:
New York.
24 Robotic Manipulation Chap-1

2
Direct Kinematics:
The Arm Equation
A robotic manipulator can be modeled as a chain of rigid bodies called links. The
links are interconnected to one another by joints, as shown in Fig. 2-1. One end of
the chain of links is fixed to a base, while the other end is free to move. The mobile
end has a flange, or face píate, with a tool, or end-effector, attached to it. There are
typically two types of joints which interconnect the links: revolute joints, which ex
hibit rotational motion about an axis, and prismatic joints, which exhibit sliding mo
tion along an axis.
The objective is to control both the position and the orientation of the tool in
three-dimensional space. The tool, or end-effector, can then be programmed to fol-
low a planned trajectory so as to manipúlate objects in the workspace. In order to
program the tool motion, we must first formúlate the relationship between the joint
variables and the position and the orientation of the tool. This is called the direct
kinematics problem, which we define formally as follows:
Figure 2-1 Robotic manipulator mod

Base eled as a chain of links.
25
Problem : Direct Kinematics. Given the vector of joint variables of a robotic
manipulator, determine the position and orientation of the tool with respect to a co
ordinate frame attached to the robot base.
To develop a concise formulation of a general solution to the direct kinematics

problem, it is helpful to first briefly review some fundamental properties of vector
spaces (Noble, 1969; Schilling and Lee, 1988). This is followed by a discussion of
rotations, translations, and screw transformations between coordinate trames. Four-
dimensional homogeneous coordinates are employed in order to develop a matrix
representation of these operations. A systematic procedure for assigning coordinate
frames to the links of a robotic manipulator is then presented (Denavit and Harten-
berg, 1955), and this leads directly to the arm equation. The solution to the direct
kinematics problem is illustrated using three robotic manipulators: the five-axis
Rhino XR-3, the four-axis Adept One, and the six-axis Intelledex 660. These robots
are used repeatedly as case study type examples throughout the remainder of the
text.
2-1 DOT AND CROSS PRODUCTS
Vectors in n-dimensional space R" can be thought of as arrows emanating from the
origin, as shown in Fig. 2-2 for the case n = 3. The coordinates of the vector are
then synonymous with the location of the tip of the arrowhead.
Figure 2-2 An orthonormal coordinate

frame in R3.
The vectors pictured in Fig. 2-2 are special in the sense that any vector in the
space R 3 can be easily represented in terms of these vectors. To see this, it is useful
to first introduce the notion of an inner product, or dot product, of two vectors.
Definition 2-1-1: Dot Product. The dot product of two vectors x and y in R"
is denoted x • y and is defined:
x*y = 2 xkyk
k= 1
The symbol 4 should be read “equals by definition.” The dot product is also
called the Euclidean inner product in R". The matrix transpose operation can be used
to express the dot product more compactly as
26 Direct Kinematics: The Arm Equation Chap- 2

x • y — x Ty. ( 2 - 1- 1)
Here x T denotes the row vector obtained by taking the transpose of the column vec
tor x. The dot product in R" has all the properties that are characteristic of inner
products in general. In particular, the following properties are fundamental:
Proposition 2-1-1: Dot P roduct. Let { jc, y, z} be vectors R" and let {a,
/?} be real scalars. Then the dot product has the following properties:
1. jc • x > 0
2. jc • jc = 0 <=> jc = 0
3. jc • y ~ y • jc
4.(ax + Py) • z = a ( jc • z) + /3 (y • z)
These properties follow directly from Def. 2-1 -1. Note that the first two properties
indícate that the dot product of a vector with itself is always nonnegative and is zero
only for the zero vector. The third property specifies that the dot product is a com-
mutative function of its two arguments, that is, interchanging the order of jc and y
does not alter the result. Finally, the last and perhaps most important property is that
the dot product is a linear function of its first argument jc.Thus the dot product of
the sum is the sum of the dot products. In view of the commutative property, the dot
product is also a linear function of its second argument y.
The dot product of two vectors is a measure of the orientation between the two
vectors. To see this, it is useful to first examine the concept of orthogonality.
Definition 2-1-2: O rthogonality. Two vectors jc and y in R" are orthogonal

if and only if jc • y = 0 .
For the Euclidean inner product or dot product in Def. 2-1 -1, orthogonal vectors can
be interpreted geometrically in three-dimensional space as perpendicular vectors.
Thus the three vectors shown in Fig. 2-2, which correspond to the three columns of
the identity matrix / , are mutually orthogonal. We therefore refer to them as an or
thogonal set. Not only is the set of vectors in Fig. 2-2 an orthogonal set, it is also a
complete set. A complete orthogonal set is a set of vectors with the property that the
only vector orthogonal to every member of the set is the zero vector.
Definition 2-1-3: Completeness. An orthogonal set of vectors { jc 1, jc2 ,
. . . , jc " } in R" is complete if and only if
y -x k — 0 for 1 < £ < /i y = 0
The number of vectors it takes to form a complete orthogonal set for a vector
space is called the dimensión of the space. Thus R 3 is a three-dimensional space and,
more generally, R ” is an «-dimensional space. The vectors pictured in Fig. 2-2 are
also special in one final sense. The length of each vector is unity, and so we refer to
these vectors as unit vectors. The length or norm of an arbitrary vector in R" can be
defined in terms of the dot product as follows:
Sec. 2-1 Dot and Cross Products 27

Deñnition 2-1-4: Norm. The norm of a vector x in R" is denoted as || x\\ and Prof
is defined: and let 6 I
/ » \ 1/2
||jc|| = ( x - x ) m =
Thus, whe
Thus the norm is an n -dimensional generalization of the absolute valué function. A vectors, ti
set of orthogonal unit vectors is referred to as an orthonormal set. The vectors between t
shown in Fig. 2-2 are an example of a complete orthonormal set, or an orthonormal R 3 and thi
coordinate fram e. It is a right-handed coordinate frame, because when the fingers of Dot
the right hand are curled from the first vector into the second vector, the thumb of the kin
points in the direction of the third vector. One important result that will be useful kinematic
when we examine rotating coordinate frames is the following geometric relationship
between the dot product and the norm:
Proposition 2-1-2: O rientation. Let x and y be arbitrary vectors in R3 and

let 6 be the angle from jc to y. Then:
jc - y = ||*|| ||y || eos 6
Henee the dot product can be interpreted as a measure of the orientation between
two vectors. It is proportional to the cosine of the angle between the vectors, the
constant of proportionality being the product of the lengths of the vectors. It follows
from Prop. 2-1-2 and Def. 2-1-2 that the angle between orthogonal vectors is 7r/2
radians. Henee orthogonal vectors in R3 are perpendicular to one another. COORDINA
There is another operation with vectors which is useful in the analysis of
robotic manipulators, the cross product. Whereas the dot product of two vectors The inne
generates a scalar, the cross product of two vectors generates another vector that can arbitrary
bers of a
be defined as follows:
trary vec
thonorm
Definition 2-1-5: Cross Product. The cross product of two vectors u and t (weighte
in R3 is a vector w = u x v which is orthogonal to u and v in the right-handed sense coordina
and defined as follows:
N
z1 l• 2 z3“ U2 V 3 ~ U3 V 2 De
U\ U2 U3 > = U3 V 1 - U 1 V 3
spect to
JJI l?2 U3_ J _U 1V2 - U2V]_
Here the determinant notation is used in a formal sense to indicate how to evalúate
the cross product. In this case the unit vectors {z\ z2, i 3} are treated as if they were
scalars for the purpose of computing the determinant. For the right-handed or Th
thonormal coordinate frame shown in Fig. 2-2, i 1 = i 2 x z3, i 2 = i 3 x i 1, and na te fra
i 3 = z1 x i 2. Recall that the valué of the dot product u • v depends upon the angle vectors
between the two vectors u and u. Similarly, the length of the cross-product vector frame a
u x t depends on the angle between the two vectors u and o, as can be seen from mental ¡
the following result:
Sec. 2-5
28 Direct Kinematics: The Arm Equation Chap. 2
P ro p o sitio n 2-1-3: C ross P ro d u ct. Let u and t be arbitrar) vectors in R
and let 0 be the angle from u to v. Then:
\\u x 11| = l! u || || v ¡| sin 6
Thus, whereas the dot product is proportional to the cosine of the angle between the
vectors, the size of the cross product vector is proportional to the síne o f the angle
between the vectors. Refer to Fig 2-3 for an iliustration of two vectors u and t in
R 3 and their cross product.
Dot products and cross products are useful in developing concise form ulations
of the kinem atics, statics, and dynamics of robotic manipulators. We begin w ith the
kinem atic analysis.
u Figure 2-3 Cross product of u w ith v
2-2 COORDINATE FRAMES
The inner product, or dot product, is useful because one can employ it to represent
arbitrary vectors in terms of their projections onto subspaces generated by the m em
bers o f a complete orthonorm al set of vectors. In particular, suppose p is an arbi-
trary vector in the space R", and suppose X — { x \ x 2, . . . , jc”} is a complete o r
thonorm al set for R". We can then represent the vector p as a linear com bination
(weighted sum) of the vectors in X where the weighting coefficients are called the
coordinates of p with respect to X.
D efinition 2-2-1: C o o rd in ates. Let p be a vector in R ” and let X — { x \ or2,

, x"} be a com plete orthonorm al set for R \ Then the coordinares of p with re
spect to X are denoted [p]x and are defined implicitly by the equation:
n
p = 2 \p\uk
ife*i
The com plete orthonorm al set X is sometimes called an orthonorm al coordi

nare fram e or an orthonorm al basis for the space R". W hen the coordinate-fram e
vectors are orthonorm al, the coordinates of a vector with respect to that coordinate
frame are particularly easy to compute, as can be seen from the following funda
m ental result:
Sec. 2-2 Coordinate Frames 29

Proposition 2-2-1: O rthonorm al C oordinates. Let p be a vector in R n and
let [p]x be the coordinates of p with respect to an orthonormal coordinate frame
X = { x l, x 2, . . . , jc"}. Then the kth coordinate of p with respect to X is:
lp]i * P * k 1 n
Proof. From Def. 2-2-1, it is sufficient to show that c is the zero vector where
c is the difference or error:
C = p - i í p ■X k) x k
** 1
Using Def. 2-1-2, Def. 2-1-4, and the properties of the dot product summarized in
Prop. 2-1-1, we have for each 1 ^ j ss n: Figure 2-4
spect to the
c •x J = p •x J 2 ( p - x k)x‘
k= l
I
= p • Xj - 2 ( p mXk)(xk • X1) t
k= 1
F
n
t
p - x J - 2 (p ■Xk)hj r
k=l
i
— p • XJ — p ' XJ
l
= 0
Thus c is orthogonal to every member of i g Because X is complete, it follows from
Def. 2-1-3 that c — 0 and thus that Prop. 2-2-1 is valid.
1
We see that the &th coordinate of p with respect to an orthonormal coordinate
r
frame X is simply the dot product of p with the kth member of the set X. The geo-
c
metric interpretation of the relationship between a vector and its coordinates is
t
shown in Fig. 2-4, where, for simplicity, the two-dimensional space R 2 is used.
f
Note how p is decomposed into the sum of its orthogonal projections onto the one-
dimensional subspaces (lines) generated by jc1 and jc2. In view of the simple relation
ship between a vector and its orthonormal coordinates, all coordinate frames used in
the remainder of the text are assumed to be right-handed orthonormal frames unless J
otherwise noted. ]
The solution to the direct kinematics problem in robotics requires that we rep (
resent the position and orientation of the mobile tool with respect to a coordinate
frame attached to the fixed base. This involves a sequence of coordinate transforma-
^ tions from tool to wrist, wrist to elbow, elbow to shoulder, and so on. Each of these
coordinate transformations can be represented by a matrix. To see this, it is useful to
first examine a simple special case, the single-axis “robot” displayed in Fig. 2-5.
Here a single revolute joint connects a fixed link (the base) to a mobile link (the
body). The problem is to represent the position of a point p on the body with respect
to a coordinate frame attached to the base. The coordinates of p in this case will
vary as the joint angle 6 changes. Let F = { / \ / 2, / 3} and M = {m \ m 2, m 3} be or
thonormal coordinate frames attached to the fixed and the mobile links, respec-

Mobile link
Revolute joint
Fixed link
Figure 2-4 Coordinates of p with re Figure 2-5 Single-axis robot.

spect to the coordinate frame X.
tively. Note that m 2 is not shown in Fig. 2-5, since it is obscured by the body in the
position shown, which corresponds to 6 — tt/2. Since p is a point attached to the
body, the coordinates of p with respect to the mobile coordinate frame M will re-
main fixed. However, the coordinates of p with respect to the fixed coordinate frame
F will vary as the body is rotated. From Prop. 2-2-1, the two sets of coordinates of
p are given by:
[p]M - ÍP *w 1, p • m 2, p • m 3]T (2-2-1)

[ pY = [ p - f \ P ' f \ p - ñ T (2-2-2)
The problem then is to find the coordinates of p with respect to F, given the coordi
nates of p with respect to M. This is a coordinate transformation problem. The coor
dinate transformation problem has a particularly simple solution whenthe destina-
tion or reference coordinate frame F is orthonormal, as can beseen from the
following result:
P roposition 2-2-2: C oordinate Transform ations. Let F = { / \ / 2, . . . ,

/" } and M = {m \ m 2, . . . , m n} be coordinate frames for R” with F orthonormal.
Next, let A be the n x n matrix defined by Akj = /* • mj for 1 ^ k, j ^ n. Then for
each point p in R ”:
lp]F = A[p]M
Proof. Since F is an orthonormal coordinate frame, it follows from Prop.
2-1-1, Prop. 2-2-1, and Def. 2-2-1 that for 1 < k < n:
lP)F
> ~ P k- f
= •/*
n
V . fk
= Z l( n
—
p i] f (" * '* / )
Sec. 2-2 Coordínate Frames 31

Thus [p]F = A[p]M where A*y = /* • m; for 1 ^ k, j ^ n.
We cali the matrix A that maps or transforms mobile coordinates into fixed co
ordinates a coordinate transformation matrix. If aj denotes the yth column of A,
then, from Prop. 2-2-1 and Prop. 2-2-2, we find that aj - [mj]F for 1 < j < n.
Consequently, the yth column of the coordinate transformation matrix A is simply the
coordinates of the jth vector of the source coordinate frame M with respect to the
destination coordinate frame F.
Example 2-2-1: Coordínate Transformation
For the two coordinate frames pictured in Fig. 2-5, suppose the coordinates of the
point p with respect to the mobile coordinate frame are measured and found to be
[p]M = [0.6, 0.5, 1.4]r. What are the coordinates ofp with respect to the fixed coordi
nate ffame F with the body in the position shown?
Solution From Prop. 2-2-2, we have:
[p Y = A[p}M
~ f - m ' f ■m 2 P m 0 . 6'
= f 2 •m ' P - m 2 f 2 m 0.5
f - m ' P - m 2 P m 1.4
0 1 0* 0 . 6 '
1 0 0 0.5
0 0 1 1.4
= [-0 .5 , 0.6, 1 4]r
Note that this is consistent with the picture shown in Fig. 2-5.
If we have a coordinate transformation matrix which maps coordinates relative

to one frame into coordinates relative to another frame and instead we want to move
in the reverse direction, we need to find the inverse of the coordinate transformation
matrix. When both the source and the destination coordinate frames are orthonor
mal, finding the inverse of the coordinate transformation matrix is particularly sim
ple, as can be seen from the following useful result:
Proposition 2-2-3: Inverse C oordinate T ran sfo rm atio n . Let F and M be

two orthonorm al coordinate frames in R" having the same origin, and let A be the
coordinate transformation matrix that maps M coordinates into F coordinates as in
Prop. 2-2-2. Then the coordinate transformation matrix which maps F coordinates
into M coordinates is given by A-1, where:
A-1 = A T

Proof. By multiplying both sides of the equation in Prop. 2-2-2 on the Ieft by
A ~ \ we see that A~x maps F coordinates into M coordinates. It remains to verify that
the inverse is just the transpose. Let 1 ^ k, j < n. Since both F and M are or
thonormal, we can use Prop. 2-2-2 and the commutative property of the dot product
as follows:
( A~% = m k- f
lili
== Ajk
= (AT)kj
Since the above equation holds identically for all 1 ^ k %j ¡j /i, Prop. 2-2-3 then
follows.
2-3 ROTATIONS
In order to specify the position and orientation of the mobile tool in terms of a coor
dinate frame attached to the fixed base, coordinate transformations involving both
rotations and translations are required. We begin by investigating the representation
of rotations.
2-3-1 Fundamental Rotations
If the mobile coordinate frame M is obtained from the fixed coordinate frame F by
rotating M about one of the unit vectors of F, then the resulting coordinate transfor
mation matrix is called a fundamental rotation matrix. In the space R3 there are
three possibilities, as shown in Fig. 2-6.
As an illustration, suppose we rotate the mobile coordinate frame M about the

/* axis of the fixed coordinate frame F. Let <f>be the amount of rotation, measured
in a right-handed sense, as shown in Fig. 2-7. Next let R\((t>) be the resulting coordi
nate transformation matrix which maps mobile M coordinates into fixed F coordi
nates. Then, from Prop. 2-2-2:
Sec. 2-3 Rotations 33

The
rotation n
row and t
the diago
Figure 2-7 Rotation of M a b o u t/1 by
the angle
f\ m angle <J>.
for the ¿ti
m P-m 2 p -m 3
Example 2
/?! (4>) m P-m 2 P-m 3 (2-3-1)
Refe
m P-m 2 P -m 3 the f
Since we are rotating about the/ ' axis, it is clear from Fig. 2-7 that/ 1 = m'. Sub- p is i
stituting this into Eq. (2-3-1) and recalling that { p , p , f 3} and { m [, m 2, m ’} are or the c
thonormal sets, it follows that the fundamental rotation matrix (<£) simplifies to: Solu
1 0 0
RÁ<t>) = 0 p - m 2f • m 3
2
(2-3-2)
0 P - m 2 f 3 ■m 3
Next note from Fig. 2-7 that the vectors { / 2, / 3, m 2, m 3} all lie in a plañe or
thogonal to / 1 = m l. Recalling Prop. 2-1-2, the dot product of two vectors is pro-
portional to the cosine of the angle between the vectors. When the vectors are nor
mal (unit length), the dot product is in fact equal to the cosine of the angle between
the vectors. From Fig. 2-7 it is evident that the angle from / 2 to m 2 is <f>, as is the
angle from / 3 to m3. Consequently the diagonal elements in the 2 x 2 submatrix in
Eq. (2-3-2) are equal to eos </>. givei
In order to represent the off-diagonal elements in terms of </>, note from Fig. coor<
2-7 that the angle from f 2 to m3 is t t / 2 + </>. Similarly, the angle from / 3 to m 2 is
Solui
—tt/2 + </>. Using the trigonometric identity for the cosine of the sum (Appendix 1) the i
and Prop. 2-1-2, the general form for the first fundamental rotation matrix then re Prop
duces to:
1 0 0
0 eos (f> —sin 4> (2-3-3)
0 sin (¡> eos 4>
A similar analysis can be used to derive expressions for the second and the
thirdfundamental rotation matrices. In particular, if R2(<£) and R3(</>) represent rota
tions by <f) about the second and third unit vectors of the fixed coordinate frame f f
then:
eos 4> 0 sin (f>
R2Í<f>) i 0 1 0 (2-3-4)
- s in é 0 eos ó Note
34 Direct Kinematics: The Arm Equation Chap. 2 Sec. 2-3

eos <¡> —sin <f> 0
RÁ4>) I sin </> eos 4> 0 (2-3-5)
0 0 1
There is a simple, consistent pattern to the structure of the three fundamental

rotation matrices. The kth row and the Ath column of /?*(<£) are identical to the kth
row and the kth column of the identity matrix /. In the remaining 2 X 2 submatrix,
the diagonal terms are eos <£>, while the off-diagonal terms are ±sin <¡>> where <f>is
the angle of rotation. The sign of the off-diagonal term above the diagonal is ( —1)*
for the kth fundamental rotation matrix.
Example 2-3-1: Rotation

Refer to Fig. 2-7, where the mobile coordinate frame M is rotated about the/ ' axis of
the fixed coordinate frame F. Let <f) — tt/3 radians be the amount of rotation. Suppose
p is a point whose coordinates in the mobile M frame are [p]M = [2, 0, 3]r. What are
the coordinates of p in the fixed coordinate ffame F?
Solution Applying Prop. 2-2-2:
Ip Y = «i (f) [p]M
"1 0 0
= 0 0.5 - 0.866
0 0.866 0.5
= [2 -2.598 i.5 r
Next, suppose q is a point whose coordinates in the fixed coordinate ffame F are
given by [qY = [3, 4, 0]r . What are the coordinates of q with respect to the mobile
coordinate ffame M l
Solution Since the two coordinate frames are orthonormal and have the same origin,
the inverse of the coordinate transformation matrix is just the transpose. Applying
Prop. 2-2-2 and Prop. 2-2-3:
i
"3“
o
= 0 0.5 0.866 4
0 -0.866 0.5 _ 0_
= [3 2 -3 .4 6 4 f
Note that both results are consistent with the picture shown in Fig. 2-7.

2-3*2 Composite Rotations
When a number of fundamental rotation matrices are multiplied together, the pro
duct matrix represents a sequence of rotations about the unit vectors. We refer to
múltiple rotations of this form as composite rotations. Using composite rotations, we
can establish an arbitrary orientation for the robotic tool. Consider the sketch of a
robotic tool shown in Fig. 2-8. Here the mobile coordinate frame M = { m \ m 2,
m 3} rotates with the tool, while the fixed coordinate frame F = { / \ / 2, / 3} is sta-
tionary at the end of the forearm. The three fundamental rotations correspond to
yaw, pitch, and roll. Recall from Chap. 1 that yaw is a rotation about the f l axis,
pitch is a rotation about the f 2 axis, and roll is a rotation about the / 3 axis.

tool.
Each fundamental rotation is represented by a matrix. However, the operation

of matrix multiplication is not commutative, because for two arbitrary matrices A
and B , it is not true that AB = BA. Consequently, the order in which the fundamen
tal rotations are performed does make a difference in the resulting composite rota
tion. Furthermore, once one rotation has been performed, the axes of the two coor
dinate frames are no longer coincident. At this point subsequent rotations of the tool
could be performed about the unit vectors of either the fixed coordinate frame F or
the rotated coordinate frame M. In order to resolve the ambiguities as to how a com
posite rotation matrix should be constructed, the following algorithm is used:
Algorithm 2-3-1: Composite Rotations
1. Initialize the rotation matrix to R = /, which corresponds to the orthonormal

coordinate frames F and M being coincident.
2. If the mobile coordinate frame M is to be rotated by an amount (f>about the kth
unit vector of the fixed coordinate frame F , then premultiply R and /?*(<£).
3. If the mobile coordinate frame M is to be rotated by an amount <¡>about its
own Ath unit vector, then postmultiply R by Rk(4>)-
4. If there are more fundamental rotations to be performed, go to step [2]; else,
stop. The resulting composite rotation matrix R maps mobile M coordinates
into fixed F coordinates.

Thus composite rotation matrices are built up in steps starting with the identity
matrix which corresponds to no rotation at all. Rotations of frame M about the unit
vectors of frame F are represented by premultiplication (multiplication on the left)
by the appropriate fundamental rotation matrix. Similarly, rotations of frame M
about one of its own unit vectors are represented by postmultiplication (multiplica
tion on the right) by the appropriate fundamental rotation matrix.
Since a convention has been adopted for the order of the yaw, pitch, and roll
operations, a general expression for the composite yaw-pitch-roll transformation ma
trix can be obtained. Suppose that the yaw-pitch-roll angles are represented by a
vector 0 in R 3. For notational convenience, let S* = sin 0* and C* = eos 0*. We then
have the following result, which summarizes the composite yaw-pitch-roll tool trans
formation:
Proposition 2-3-1: Yaw-Pitch-Roll Transformation. Let YRP (0) repre

sent the composite rotation matrix obtained by rotating a mobile coordinate frame
M = {m \ m 2, m 3} first about unit vector / ' with a yaw of 0i, then about the unit
vector f 2 with a pitch of 02, and finally about unit vector/ 3 with a roll of 03. The
resulting composite yaw-pitch-roll matrix YPR (0) maps M coordinates into F coor
dinates and is given by:
C2C3 S1S2C3 C ,S 3 C1S2C3 + Si S3

Y P R (0) = C2S3 S1S2S3 4* C1C3 C1S2S3 ~ S1C3
—S2 Si C2 Ci C2
Proof. This is verified by simply applying Algorithm 2-3-1 and using the ex
pressions for the three fundamental rotation matrices. Using the notational short
hand C* = eos dk and S* = sin 0* and Algorithm 2-3-1, we have:
Y P R (0) = R 3(03)/?2(02)Rl(0l)
C3 - s 3 o c2 s2 "1 0 0"
s3 c3 0 o o 0 Ci -S i
o o 1 —s 2 c2 0 Si CiJ
c3 - s 3 0‘ c2 s ,s 2 C1S2
s3 c3 0 o c, -S i
o o 1 - s 2 S1C2 c,c2
C2C3 S1S2C3 — C1S3 C1S2C3 + Si s3
C2S3 S1S2S3 + C,C3 C 1S2S3 — S1C3
—S2 Si C2 C1C2
Example 2-3-2: Yaw-Pitch-RoII

Suppose we rotate the tool in Fig. 2-8 about the fixed axes starting with a yaw of tt/2,
followed by a pitch of - 7 t / 2 and, finally, a roll of tt/ 2 . What is the resulting com
posite rotation matrix?
S e a 2-3 Rotations 37
S o lu tion Applying Prop. 2-3-1:
r —7r
~ } r 2I
\ 2/ !^ 2
~o 0 r
0 “ 1 0
1 0 0
Suppose the point p at the tool tip has mobile coordinates [p]M — [0, 0 , 0 .6 ]r.
Find [p]F follow ing the yaw-pitch-roll transformation.
S olu tion Using Prop. 2-3-1:
Ip Y =RIpY4
i
o
o
“ 0 “
0 - 1 0 0
l 0 0_ _0.6_
= [0.6, 0, Of
Exercise 2-3-1: Yaw-Pitch-Roll. Verify that Example 2-3-2 is correct by

sketching the tool position and the coordinate frames after each fundamental rotation
is performed.
Exercise 2-3-2: Roll-Pitch-Yaw. An alternative way to define the yaw-

pitch-roll transformation is to perform the rotations about unit vectors of the mobile
frame M and in the reverse order: roll, pitch, yaw. This tends to be easier to visual-
ize, particularly when the angles are arbitrary. Show that the resulting composite ro
tation matrix RPY (0), is the same as in Prop. 2-3-1. Is the route taken by the tool
the same?
Exercise 2-3-3: Fundam ental Rotations. Verify that the general yaw-pitch-
roll transformation YPR (0) includes the three fundamental rotations {R\, R2, Ri} as
special cases.
A general rotation of a mobile frame M with respect to a fixed frame F can al-
ways be decomposed into a sequence of three fundamental rotations as in Prop.
2-3-1. It is also possible to represent a general rotation as a single rotation by an an
gle <f>about an arbitrary unit vector u as shown in Fig. 2-9. This is called the equiv-
alent angle-axis representation. For notational convenience, let C</> = eos
S(f> = sin <fi, and V<£ = 1 — eos <f>. The equivalent angle-axis rotation can then be
represented as follows (Fu et al., 1987):
Proposition 2-3-2: Equivalent Angle-Axis. Let F and M be two orthonor

mal coordinate frames in R3 with M initially coincident with F. Let u be a unit vec
tor, and suppose Af is rotated about u by an angle </>in the right-handed sense. Then
the equivalent angle-axis rotation matrix R(<¡), u) which maps M coordinates into F
coordinates is:

Figure 2-9 Equivalent angle-axis rota
tion.
u\ V<j) + Q<¡) U\U2 Wcf) — u$S(f) U\U2 V0 + U2$<t>

/?(<£, u) = U\U2V(j> + w3S<¿> u \ V<¿> + C<j> u2ui\<¡) - WiS$
W1W3V0 — w2S</> M2W3V0 + wiS</> W3V<¿> 4- C<f>
Proof. Let w = [1/1, w2, 0]r denote the orthogonal projection of u onto the/ ’/ 2
plañe as shown in Fig. 2-9. Then:
A = nwti
I (mí + u\)m
Next, let a be the angle from f to m, and let p be the angle from ü to u as shown in
Fig. 2-9. Then, since ||w|| = 1, it follows that:
r - Ux
"a
c - Ul
~K
C/5 = A
sp = w3
We can express the equivalent angle-axis rotation matrix as a composition of

fundamental rotation matrices. First, we rotate the mobile frame M about the/ 3 axis
by an angle of —a . Thus:
R = Ri(-a)
This puts the unit vector u in the f lf 3 plañe. Next we rotate frame M about the / 2
axis by an angle of p . Since the rotation is about an axis of the fixed frame, we pre-
multiply to obtain the composite rotation matrix:
R = R2(P)R>(-a)

This aligns unit vector u with the/ 1 axis. At this point we perform the desired rota
tion by an angle <f>about the unit vector u — f x. Thus the composite rotation matrix
is:
R - Ri(4>)R2(P )R ¿ -a )
Next we reverse the process to restore u to its original position. First we rotate
2-4 HO
It remains to substitute the expressions for the fundamental rotation matrices

from Eqs. (2-3-3) through (2-3-5) into the expression for /?(</>, u) and to perform
the matrix multiplication. The procedure is tedious but straightforward, and the re
sult after simplification using CaC/3 = Wi, SaC/3 = W2, and A2 + u\ = 1 is as
given in Prop. 2-3-2.
The three fundamental rotation matrices represent important special cases of

the equivalent angle-axis rotation matrix, as can be seen from the following exer-
cise:
Exercise 2-3-4: Fundam ental Rotations. Let R(<f>, u) be the equivalent

angle-axis rotation matrix and let Rk{<t>) be the £th fundamental rotation matrix.
Verify that:
R(<t>yfk) = Rk(4>) 1 <3
The equivalent angle-axis representation for a general rotation can also be in-
verted (Paul, 1982). For example, given a rotation matrix /?, suppose we want to
find the axis u and angle <f>. The sum of the diagonal terms of a square matrix is
called the trace of the matrix. If we compute the trace of the matrix R{<¡), u) in
Prop. 2-3-2, then after simplification using trigonometric identities and ||w|| = 1,
the result is trace [/?(<£, u)] — 1 + 2 eos $ . Thus the angle part of the equivalent
angle-axis representation is:
trace (R) — 1
|¡ árceos (2-3-6)
2
Once the angle of rotation is identified over the range (0, 7r), the axis or unit
vector about which the rotation is performed can then be determined. In particular,
if we form differences between off-diagonal terms of /?(<£, u) in Prop. 2-3-2, we
can isolate the components of u as follows:
(2-3-7)

Note that u is well defined for 0 < <f> < 7r, but at <f> = 0 and at <j> = n the ex
pressions for the components of u in Eq. (2-3-7) reduce to 0/0. Thus Eq. (2-3-7) is
not well defined numerically for small valúes of <f>or as (f>approaches tr. For a fur-
ther discussion of these cases, see Paul (1982).
HOMOGENEOUS COORDINATES
Puré rotations are sufficient to characterize the orientation of a robotic tool. How
ever, to characterize the position of the tool relative to a coordinate frame attached
to the robot base, we must use another type of transformation, a translation. Transla-
tions are qualitatively different fforn rotations in one important respect. With rota
tions, the origin of the transformed coordinate frame is the same as the origin of the
original coordinate frame. This invariance of the origin is characteristic of linear
operations in general, and it is this feature that allows us to represent rotations in
three-dimensional space with 3 x 3 matrices. However, the origin of a translated
coordinate frame is not the same as the origin of the original coordinate frame. Con-
sequently, it is not possible to represent a translation in R3 with a 3 x 3 matrix.
2*4-1 Homogeneous Coordinate Frames
Instead we must move to a higher-dimensional space, the four-dimensional space of

homogeneous coordinates. We define the homogeneous coordinates of a point as fol
lows:
Definition 2-4-1: Homogeneous Coordinates. Let q be a point in R3, and

let F be an orthonormal coordinate frame for R3. If cr is any nonzero scale factor,
then the homogeneous coordinates of q with respect to F are denoted [q]F and de
fined:
[q]F = [aqu crq3, cr]r
Thus the homogeneous coordinates of a point q in R3 are represented by a vec-

tor [q]F in four-dimensional space R4. The fourth component cr is a nonzero scale
factor. To recover the original physical three-dimensional vector from its four-
dimensional homogeneous coordinates, we can use q = Ha[qY, where Ha is a 3 x 4
homogeneous coordinate conversión matrix, defined as follows:
=-[/,()] (2-4-1)
cr
Note that the homogeneous coordinates [qY are not unique, because any scale
factor cr ~t~ 0 will yield the same three-dimensional physical vector q. In robotics,
we normally use a scale factor of cr = 1 for convenience. In this case, four-
dimensional homogeneous coordinates are obtained from three-dimensional physical
coordinates by simply augmenting the vector with a unit fourth component. Simi-
larly, three-dimensional physical coordinates are recovered from four-dimensional
homogeneous coordinates by merely dropping the unit fourth component.
Sec. 2-4 Homogeneous Coordinates 41

If a physical point in three-dimensional space is expressed in terms of its ho
mogeneous coordinates and we want to change from one coordinate ffame to an
other, we use a 4 x 4 homogeneous transformation matrix. In general, a homoge
neous transformation matrix T can be partitioned into four sepárate submatrices as
follows:
ro ta tio n p o s itio n
R i P
i
rf J * -
p e rs p e c tiv e scale
Here the 3 x 3 submatrix R in the upper left comer of T is a rotation matrix.

It represents the orientation of the mobile coordinate frame relative to the fixed ref-
erence frame. The 3 x 1 column vector p in the upper right córner of T is a transla
tion vector. It represents the position of the origin of the mobile coordinate ffame
relative to the fixed reference frame. The scalar cr in the lower right córner of T is a
nonzero scale factor which is typically set to unity. Finally, the 1 x 3 row vector rjr
in the lower left córner of T is a perspective vector. For the time being, this vector
will always be rj = 0 . Later, when we examine the use of an overhead camera in
Chap. 8, nonzero perspectives will be used. In terms of a robotic arm, p specifies
the position of the tool tip, R specifies the orientation of the tool, rj specifies a point
of perspective for visual sensing with a camera, and cr is typically set to unity for
standard scaling.
2-4*2 Translations and Rotations
The fundamental operations of translation and rotation can each be regarded as im
portant special cases of the general 4 x 4 homogeneous transformation matrix. To
see this, first consider the question of rotation. Suppose F and M are two orthonor
mal coordinate frames that are initially coincident. If we rotate M by an amount </>
about the kth unit vector of F, then in terms of homogeneous coordinates this opera-
tion can be represented by a 4 x 4 matrix denoted Rot (</>, k), where:
i 0
R ¿ 4>) i 0
Rot (<£, k) = i o (2-4-3)
1
t
o o o ! i
Here Rk(<t>) is simply the fcth fundamental rotation matrix introduced in Sec. 2-3.
We refer to Rot ($, k) as the kth fundamental homogeneous rotation matrix. Com
posite homogeneous rotation matrices are then built up from the fundamental homo
geneous rotation matrices in a manner analogous to the method summarized in Al
gorithm 2-3-1. Note that the translation vector p in each case has been set to zero.

Example 2-4-1: Homogeneous Rotation
Suppose that [q]M ~ [0, 0, 10, l )7 represents the homogeneous coordinates of a point
located 10 units along the third vector of a mobile coordínate frame M Suppose that
initially M is coincident with a fixed coordinate frame F If we rotate the mobile M
ffame by 7r/4 radians about the first unit vector of F, then the resulting homogeneous
coordinate transformation matrix is:
¡ o
m
> 41) i0
: o
Rot W ’ * i
. . j ____ _
0 0 0 i 1 J
’l 0 0 0
7T 7T
0 eos — sin 0
—
4 4
7r 7r
0 sin — eo s — 0
4 4
0 0 0 1
■ 1 0 0 0
0 0.707 -0.707 0
0 0.707 0.707 0
0 0 0 1
Henee the homogeneous coordinates of the point q in the fixed coordinate frame F fol
lowing the rotation are:
I q f = Rot ( j , l)[<?]M
1 0 0 0 Y
0 0.707 0.707 0 0
0 0.707 0.707 0 10
0 0 0 1 1
m _
= [0 -7.07 7.07 l ]5
Finally, the physical coordinates of the point q in the fixed coordinate frame F are
q = [0, -7 .0 7 , 7.07]r.
The pow er o f homogeneous coordinates lies in the fact that matrices can also
be used to represent translations. For example, let F and M be two initially coinci
dent orthonorm al coordinate frames, and suppose we want to transíate the origin of
the mobile coordinate frame M by an amount p k along the fcth unit vector of F for
1 ^ k < 3. Then in term s o f homogeneous coordinates, this operation can be repre-
1
Sec. 2-4 H om ogeneous Coordinates 43
sented by a 4 x 4 matrix denoted Tran (p), where:
1 0 0 P\
0 1 0 P2
Tran (p) = 0 0 1 P> (2-4*4)
«•«
0 0 0 1
We refer to Tran (p) as the fundamental homogeneous translation matrix. Note

that the translation vector is p, while the 3 x 3 rotation matrix R has been set to the
identity matrix /.
Example 2-4-2: Homogeneous Translation

Suppose the homogeneous coordinates of a point are [q]M = [0, 0, 10, l]r, as in Exam-
ple 2-4-1. However, this time suppose we transíate the mobile M coordinate ffame rel
ative to the fixed F coordinate frame by 5 units along the / ' axis and —3 units along
thep axis. Then the homogeneous transformation matrix which represents this opera-
tion is:
' 1 0 0 5
0 1 0 - 3
Tran (p)
0 0 1 0
0 0 0 1
It follows that the homogeneous coordinates of point q relative to the reference frame F
following the translation are given by:
íq\ = Tran (p) [q]M
1 0 0 5 0
0 1 0 - -3 0
0 0 1 0 10
0 0 0 1 1
= [5 -3 10 l]r
Finally, the physical coordinates of the point q in the fixed F coordinate ffame follow
ing the translation are q = [5, —3, 10]r.
Exercise 2-4-1: Inverse T ranslation. Verify that the inverse of the funda
mental homogeneous translation matrix Tran (p) always exists and is given by:
Tran-1 (p) = Tran ( —p)
Exercise 2-4-2: Inverse R otation. Verify that the inverse of the fundamen
tal homogeneous rotation matrix Rot (</>, k) always exists and is given by:
Rot-1 ($, k) = Rot (-< p, k) = Rotr (<f>, k) 1< k ^ 3

2-4-3 Composite Homogeneous Transformations
In the general case, a homogeneous transformation matrix will represent both a rota
tion and a translation of the mobile frame with respect to the fixed ffame. A se-
quence of individual rotations and translations can be represented as a product of
fundamental homogeneous transformation matrices. However, because matrix mul
tiplicaron is not a commutative operation, the order in which the rotations and trans
lations are to be performed is important. Furthermore, the mobile coordinate ffame
can be rotated or translated about the unit vectors of the fixed frame or about its own
unit vectors. The effects of these different operations on the composite transforma
tion matrix are summarized in the following algorithm:
Algorithm 2-4-1: Composite Homogeneous Transformations
1. Initialize the transformation matrix to T = /, which corresponds to the or

thonormal coordinate frames F and M being coincident.
2. Represent rotations and translations using sepárate homogeneous transforma
tion matrices.
3. Represent composite rotations as sepárate fundamental homogeneous rotation
matrices.
4. If the mobile coordinate ffame M is to be rotated about or translated along a
unit vector of the fixed coordinate ffame F, premultiply the homogeneous
transformation matrix T by the appropriate fundamental homogeneous rotation
or translation matrix.
5. If the mobile coordinate ffame M is to be rotated about or translated along one
of its own unit vectors, postmultiply the homogeneous transformation matrix T
by the appropriate fundamental homogeneous rotation or translation matrix.
6. If there are more fundamental rotations or translations to be performed, go to
step 4; otherwise, stop. The resulting composite homogeneous transformation
matrix T maps mobile M coordinates into fixed F coordinates.
Thus composite homogeneous transformation matrices are built up in steps

starting with the identity matrix, which corresponds to coincident mobile and fixed
frames. Rotations and translations of ffame M along the unit vectors of ffame F are
represented by premultiplication (multiplication on the left) by the appropriate fun
damental homogeneous transformation matrix. Similarly, rotations and translations
of frame M along one of its own unit vectors are represented by postmultiplication
(multiplication on the right) by the appropriate fundamental homogeneous transfor
mation matrix.
Example 2-4-3: Composite Homogeneous Transformation
Let F = i / 1» /2, / 3} and M — {m \ m2, m3} be two initially coincident fixed and mo
bile orthonormal coordinate ffames, respectively. Suppose we transíate M along/ 2 by
3 units and then rotate M about/ 3 by 7r radians. Find [m']F after the composite trans
formation.

Solution Using Algorithm 2-4-1:
[mlf =
= Rot (tt, 3) Tran (3i2) [m']M
-1 0 0 0 1 0 0 0 1
0 -1 0 0 0 1 0 3 0
0 0 1 0 0 0 1 0 0
0 0 0 1 0 0 0 1 1
-1 0 0 0 1
0 -1 0 - 3 0
0 0 1 0 0
0 0 0 1 1
[-1 -3 0 If
Thus, in terms of physical coordinates, [ml]F = [—1, —3, 0]r. This is consistent with
the diagram of the two coordinate frames displayed in Fig. 2-10.
f3i k -1
■ 1
/
/
/
/
-------- ►-
Figure 2-10 Coordinate frames for Ex-
ampie 2-4-3.
Exampie 2-4-4: Composite Homogeneous Transformation
Suppose we repeat the transformation of Example 2-4-3 but this time with the order of
the operations reversed. First we rotate M by v radians about/3, and then we transíate
the rotated M by 3 units along/2. Find [ml]F following the composite transformation.
Solution Using Algorithm 2-4-1:
[m1f = T[m']M
= Tran (3i2) Rot (77, 3) [ml]M
1
1 0 0 0 -1 0 0 0
0 1 0 3 0 -1 0 0 0
0 0 1 0 0 0 1 0 0
0 0 0 1 0 0 0 1
1
F
-1 0 0 0 1
0 -1 0 3 0
0 0 1 0 0
0 0 0 1 1
[-1 3 0 l]r
Thus, in terms of physical coordinates, [mlY = [ - 1 , 3 , 0]r . Again, this is consistent
with the diagram of the two coordinate frames displayed in Fig. 2-11. Note that the

J f2 m2 3
Figure 2*11 Coordinate frames for Ex
ample 2-4-4.
final position of the mobile ffame M in Fig. 2-11 differs from that in Fig. 2-10. Henee
the order in which the fundamental transformations are performed is important.
The geometric interpretation of the homogeneous transformation matrix T is

that it provides the position and the orientation of a mobile coordinate frame M with
respect to a fixed reference frame F. In the analysis of robotic manipulators, we will
sometimes also want to express the position and orientation of the fixed frame F in
terms of the mobile frame M. Consequently, we need to find the inverse of the ho
mogeneous transformation matrix T. Because T is no longer a puré rotation between
orthonormal frames, the inverse of T is not simply the transpose. However, when
the perspective vector is 17 = 0 and the scale factor is cr = 1, the inverse of T does
have a particularly simple form, as can be seen from the following useful result:
Proposition 2-4-1: Inverse Homogeneous Transformation. Let T be a ho

mogeneous coordinate transformation matrix with rotation R and translation p be
tween two orthonormal coordinate frames. If 17 = 0 and cr = 1, then the inverse of
lis :
RT ! - R Tp
--i _
0 0 0 1
Proof. Since we have a candidate for T-1, it is simply a matter of evaluating it
to see whether it satisfies the definition of the inverse. The two matrices T and T~x
are partitioned conformably for multiplication, henee the submatrices can be treated
in a manner similar to scalars. Thus:
RT - R Tp R
T-'T =
1------
1
O
O
0 0 0
-
0 ‘
R TR 0
0
0 0 0 1
But since R is a puré rotation between orthonormal coordinate frames, it follows

from Prop. 2-2-3 that R T = R~l. Thus T~]T = /, where T~] is as in Prop. 2-4-1. A
similar analysis reveáis that 7Y _1 = /. Henee the inverse of T is as given in Prop.
2-4-1.

In view of Prop. 2 -4 -1, we can just as easily transform coordinates from the
base to the tool tip as from the tool tip back to the base. There is no need to perform
computationally expensive and potentially inaccurate numerical inverses.
Example 2-4-5: Inverse Homogeneous Transformation

Let F = f / 1, / 2, / 3} and M — {m \ m 2, m3} be fixed and mobile orthonormal coordí
nate frames as shown in Fig. 2-12. Here the homogeneous coordinate transformation
matrix which maps mobile M coordinates into fixed F coordinates is given by:
0 -1 0 0
l 0 0 2
0 0 1 2
0 0 0 1
Find the homogeneous coordinate transformation which maps fixed F coordinates into
mobile M coordinates, and use it to find [ f 2]M.
f2
Figure 2-12 Coordinate frames for Ex-
f1 ampie 2-4-5.
Solution Using Prop. 2-4-1, we have:

21F
[/T = t -' [ f 2]
-i
0 - 1 0 o' o'
1 0 0 2 1
0 0 1 2 0
0 0 0 1 1
- »
0 1 0 —2 0
-1 0 0 0 1
0 0 1 —2 0
0 0 0 1 1
= [-1 0 -2 lf
Thus, in terms of physical coordinates, [ f 2]M = [—1, 0, —2]r . This is consistent with
Fig. 2-12.

2-4-4 Screw Transformations
Certain composite homogeneous transformations arise repeatedly in robotic applica

tions. One is a linear displacement along an axis combined with an angular displace -
ment about the same axis. This type of motion corresponds to a threading or an un-
threading operation and is therefore referred to as a screw transformation.
Definition 2-4-2: Screw Transformation. Let F and M be initially coinci

dent fixed and mobile orthonormal coordinate frames, respectively. If Ai is trans
lated along the kth unit vector of F by a displacement of A and rotated about the kth
unit vector of F by an angle of </>, the resulting composite homogeneous coordinate
transformation matrix is called the kth fundamental screw transformation matrix. It
is denoted Screw (A, <f>, k) and can be expressed:
Screw (A, <¿>, k) = Rot (<£, k) Tran (A/*)
It is clear that the screw transformation matrix includes the fundamental ho
mogeneous rotation matrices as special cases, because Screw (0, <f), k) = Rot (<f>, k)
for 1 ^ k ^ 3 . Similarly, a translation along a unit vector is also a special case, be-
cause Screw (A, 0, k) = Tran (Ai*). As with a physical screw, one can define a pitch
associated with the thread. In the case of homogeneous transformations, the screw
pitch is defined:
P = 27tA threads per unit length (2-4-5)
For a puré rotation, the linear displacement is A = 0. Consequently, a puré ro

tation is a screw with infinite pitch. Similarly, for a puré translation along a unit vec
tor, <f> = 0. Thus a puré translation is a screw with a zero pitch. Note that if p is
positive, then the screw is right-handed, whereas if p is negative it is left-handed.
Refer to Fig. 2-13 for a summary of the relationship between translations, rotations,
and screws. Here each straight line through the origin corresponds to a screw with a
specific pitch. Horizontal lines are puré translations, while vertical lines are puré ro
tations.
The formulation of a screw transformation in Def. 2-4-2 consists of a transla
tion along/* by a displacement of A followed by a rotation about/* by an angle <f>. It
Figure 2-13 Translations, rotations,

and screws.

could instead have been formulated the other way around as a rotation followed by a
translation, as can be seen ffom the following exercise. This is a consequence of the > screw
fact that the rotation and translation operations are being performed along the same bitrar)
axis.
Exercise 2-4-3: Screw Transform ation. Show that the fundamental rota
tion and translation matrices associated with the unit vectors commute. That is,
2-5 LINK CO<
Tran (Ai*).Rot (<£, k) = Rot (</>, k) Tran (Ai*)
Recal 1
Exercise 2-4-4: Inverse Screw Transform ation. Verify that the inverse of either
the screw transformation is again a screw transformation with: assign
equati
Screw-1 (A, <£, k) = Screw (—A, —<£, k)
then t
Example 2-4-6: Screw Transformation physic
Let F — { / \ / 2, / 3} and M = {m \ m 2, m 3} be initially coincident fixed and mobile or
thonormal coordinate frames, respectively. Suppose we perform a screw transforma 2-5-1
tion along a x i s / 2 translating by a distance o f A = 3 and rotating by an angle o f tt/2.
Find [ m 3]F follow ing the screw transformation, and determine the pitch o f the screw. Every
Solution From Def. 2 -4 -2 , w e have: The r<
joint /
[m3]F — Screw ( :3, —, 2 I [m 3]M
— Tran (3i2) Rot í ~ , 2^ [ m 3]M

\ J
1 0 0 0 0 0 1 0 ’o
0 1 0 3 0 10 0 0
0 0 1 0 - 1 0 0 0 1
0 0 0 1 0 0 0 1 1
0 0 1 0 ’o '
0 1 0 3 0
-1 0 0 0 1
0 0 0 1 1
= [ 1 3 0 l]r
T hus, in terms o f physical coordinates, [m 3]F = [1, 3, 0 ]r. This is consistent with the
diagram o f the two coordinate frames displayed in Fig. 2 -1 4 . The pitch o f the screw is
p = j 2 threads per unit length. joint
first
to m
joint
axis
the
para
Figure 2-14 Coordinate frames for Ex- on ti
ampie 2-4-6.
is Vi
50 Direct Kinematics: The Arm Equation Chap. 2 Sec

Exercise 2-4-5: General Screw Transformation. Let Ser (A, <f>, u) denote a
screw with a linear displacement of A and an angular displacement of <f>about an ar
bitrary unit vector u. Find a formulation for Ser (A, <¡>, u) and show that:
Ser (A, <f>, /*) = Screw (A, <f>, k) 1< A< 3
2-5 LINK COORDINATES
Recall that a robotic arm can be modeled as a chain of rigid links interconnected by
either revolute or prismatic joints. The objective of this section is to systematically
assign coordinate frames to each of those links. Once this is done, a general arm
equation which represents the kinematic motion of the links of the manipulator can
then be obtained. We begin by examining certain parameters associated with the
physical design of a robotic arm.
2-5-1 Kinematic Parameters
Every adjacent pair of links is connected by either a revolute or a prismatic joint.

The relative position and orientation of two successive links can be specified by two
jo in t parameters, as shown in Fig. 2 15.
Joint k
Figure 2-15 Joint angle 6 and joint

distance d.
Here joint k connects link k — 1 to link k. The parameters associated with

joint k are defined with respect to z*~\ which is aligned with the axis of joint k. The
first joint param eter, 0k, is called the joint angle. It is the rotation about z *_1 needed
to m ake axis x k~x parallel with axis jc* . The second joint parameter, dk, is called the
jo in t distance. It is the translation along z *-1 needed to make axis x k~' intersect with
axis jc* . Thus 0k is a rotation about the axis of joint k , while dk is a translation along
the axis of joint k. For each joint, it will always be the case that one of these
param eters is fixed and the other is variable. The variable joint param eter depends
on the type o f joint, as indicated in Table 2-1. For a revolute joint, the joint angle dk
is variable and the joint distance dk is fixed. Thus the two links rotate relative to one
Sec. 2-5 Link Coordinates 51

TABLE 2-1 KINEMATIC PARAMETERS
Arm Revolute Prismatic

Parameter Symbol Joint (R) Joint (P)
Joint angle e Variable Fixed

Joint distance d Fixed Variable
Link length a Fixed Fixed
Link twist angle a Fixed Fixed
another about the axis of joint k. For a prismatic joint, the joint distance dk is vari
able and the joint angle 6k is fixed. In this case the two links transíate relative to one
another along the axis of joint k.
Just as there is a joint connecting adjacent links, there is a link between suc-
cessive joints. The relative position and orientation of the axes of two successive
joints can be specified by two link parameters, as shown in Fig. 2-16. Here link k
connects joint k to joint k + 1. The parameters associated with link k are defined
with respect to jc*, which is a common normal between the axes of joint k and joint
k + 1. The first link parameter, aky is called the link length. It is the translation
along ;c* needed to make axis z*-1 intersect with axis z*. The second link parameter,
a*, is called the link twist angle. It is the rotation about jc* needed to make axis z*_I
parallel with axis z*.
Joint k + 1
Link k
w — 5
I
Figure 2-16 Link length a and link

Joint k twist angle a.
Unlike the joint parameters, the two link parameters are always constant and
are specified as part of the mechanical design. For industrial robots, the link twist
angle is usually a múltiple of tt/ 2 radians. Sometimes the axes of joint k and joint
k + 1 intersect, in which case the length of link k is zero. Links of length zero oc-
cur, for example, in robots with spherical wrists, where the last n — 3 axes inter
sect. More generally, robots can be designed to have many of the constant kinematic
parameters equal to zero, in which case they are referred to as kinematically simple
manipulators.
For an «-axis robotic manipulator, the 4n kinematic parameters constitute
the minimal set needed to specify the kinematic configuration of the robot. For
each axis, three of the parameters are fixed and depend on the mechanical design,
while the fourth parameter is the joint variable, as indicated in Table 2-1. For a
rectangular-coordinate robot the first three joint variables are all joint distances,
whereas for an articulated-coordinate robot the first three joint variables are all joint

angles. Between these two extremes are cylindrical, spherical, and SCARA robots,
whose first three joint variables are a combination of joint angles and joint distances.
2-5-2 Normal, Sliding, and Approach Vectors
By convention, the joints and limes of a robotic arm are numbered outward starting
with the fixed base, which is link 0, and ending with the tool, which is link n. For
an n-axis robot there are n + 1 links interconnected by n joints, with joint k con-
necting link Á: — 1 to link k. In order to systematically assign coordinate frames to
the links of an «-axis robotic arm, special attention must be paid to the last link—
the tool, or end-effector. The orientation of the tool can be expressed in rectangular
coordinates by a rotation matrix R = [ r 1, r 2, r 3], where the three columns of R cor
respond to the normal, sliding, and approach vectors, respectively, as shown in Fig.
2-17. The approach vector r 3 is aligned with the tool roll axis and points away from
the wrist. Consequently, the approach vector specifies the direction in which the tool
is pointing. The sliding vector r 2 is orthogonal to the approach vector and aligned
with the open-close axis of the tool. Finally, the normal vector r l is orthogonal to
the plañe defined by the approach and sliding vectors and completes a right-handed
orthonormal coordinate frame.
Figure 2-17 Normal, sliding, and ap

proach vectors of the tool.
Recall that the yaw, pitch, and roll motions are rotations about the normal,
sliding, and approach vectors, respectively. For the purpose of representing tool ori
entation, the origin of the {r1, r 2, r 3} frame can be taken either at the wrist, as in
Chap. 1, or at the tool tip, as in Fig. 2-17. For the remainder of this text we will
assume that the origin of the { r \ r 2, r 3} frame is placed at the tool tip, as in Fig.
2-17, because this simplifies the subsequent representation of tool-tip position.
2-5-3 The Denavít-Hartenberg (D-H) Representation
Denavit and Hartenberg (1955) proposed a systematic notation for assigning right-
handed orthonormal coordinate frames, one to each link in an open kinematic chain
of links. Once these link-attached coordinate frames are assigned, transformations
between adjacent coordinate frames can then be represented by a single standard
4 x 4 homogeneous coordinate transformation matrix. To assign coordinate frames
to the links of the robotic manipulator, let Lk be the frame associated with link k.
That is:

Lk = {jc*, y*, z*} Q< k < n (2-5-1)
Coordinate frame Lk will be attached to the distal end of link k for 0 < k < n. This F
puts the last coordinate frame, L„, at the tool tip. The coordinate frames are assigned
to the links using the the following procedure:
Algorithm 2-5-1: D-H Representation
0. Number the joints from 1 to n starting with the base and ending with the tool
yaw, pitch, and roll, in that order.
1. Assign a right-handed orthonormal coordinate frame Lo to the robot base,
making sure that z° aligns with the axis of joint 1. Set k = 1.
2. Align z* with the axis of joint k + 1.
3. Lócate the origin of Lk at the intersection of the z* and z*-1 axes. If they do not
intersect, use the intersection of zk with a common normal between z k and
z * '1.
4. Select j c * to be orthogonal to both z* and z**1. If z* and z*_1 are parallel, point r
j c * away from z*"1.
5. Select y* to form a right-handed orthonormal coordinate frame Lk.

6. Set k = k + 1. If k < ny go to step 2; else, continué.
7. Set the origin of L„ at the tool tip. Align z n with the approach vector, y n with
the sliding vector, and jc" with the normal vector of the tool. Set k = 1.
8 . Lócate point b k at the intersection of the jc* and z*_1 axes. If they do not inter
sect, use the intersection of jc* with a common normal between jc* and z*-1.
9. Compute 9k as the angle of rotation from jc * 1 to j c * measured about z* \ m
10. Compute dk as the distance from the origin of frame Lk-\ to point bk measured the
along z*"1. Ti*
11. Compute ak as the distance from point b k to the origin of frame Lk measured sigi
along j c * . pie
12. Compute ak as the angle of rotation from z*-1 to z* measured about j c * . reíI
13. Set k ~ k + 1. If k < n, go to step 8; else, stop. » *ur<
For convenience, we will refer to Algorithm 2-5-1 as the D -H algorithm. Note |

that it is essentially a two-pass algorithm in terms of the manipulator. On the first f
pass (steps 0 through 7), a set of n + 1 right-handed orthonormal coordinate frames
{Lo, Li, . . . , L„} is assigned, one to the distal end of each link. On the second pass
(steps 8 through 13), the valúes for the kinematic parameters {0, d, a , a} are deter-
mined. Note that in step 8 axis j c * should always intersect with axis z k~l when
k < n. When k = n y j c * will intersect with z k~l if the last joint is a tool roll joint.
n
2-5-4 An Example: Mícrobot Alpha II
As an example of an application of the D-H algorithm, consider the Alpha II robotic

arm pictured in Fig. 2-18. This is a small table-top robotic arm manufactured by
Microbot, Inc. It is a five-axis articulated-coordinate robotic manipulator that uses
stepper motors for the joint actuators.
H
Ib
54 Direct Kinematics: The Arm Equation Chap. 2 ;
■)
Applying steps O to 7 of the D-H algorithm, we get the link-coordinate dia

gram shown in Fig. 2-19. Here the dotted diagonal line between the origin of ¿3 and
1 the origin of L a indicates that the origins of these two coordinate frames coincide.
They are drawn separated in order to make the diagram more clear. Note that the as-
i signment o f link coordinates dictated by the D-H algorithm is not unique. For exam
ple, the directions of any of the 2 axes could be reversed. Of course, this would also
require reversing the corresponding y axes in order to preserve the right-handed na-
ture of the orthonormal coordinate frames.
Shoulder Elbow Tool pitch
Figure 2-19 Link coordinates of the Alpha II robotic arm.

Next we apply steps 8 to 13 of the D-H algorithm starting with k — 1. Using
Fig. 2-19, we get the set of kinematic parameters for the Alpha II robot shown in
Table 2-2. Since the Alpha II is a five-axis articulated-coordinate robot, the vector
of joint variables is the 5 x 1 vector 0. Note that 9 of the 15 fixed parameters in
Table 2-2 are zero, which makes the Alpha II robot kinematically simple. If the
2-(
origin of the base frame L0 were to be moved up to coincide with the origin of ffame
L\ (allowable under the D-H algorithm), then the kinematic parameters could be
simplified still further with d\ — 0.
TABLE 2-2 KINEMATIC PARAMETERS FOR THE ALPHA II ROBOT
Axis e d a a Home
1 01 215.0 mm 0 —tt/2 0
2 02 0 177.8 mm 0 0
3 03 0 177.8 mm 0 0
4 04 0 0 —7r/2 —re¡2
5 05 129.5 mm 0 0 0
The valúes listed in the last column of Table 2-2 for the joint variable 6 corre
spond to the soft home position pictured in the link-coordinate diagram of Fig. 2-19.
When assigning link-coordinates using the D-H algorithm, it is always easier to
choose a soft home position that corresponds to the joint angles being múltiples of
7r/2radians even if the original picture of the robot is not so configured.
It is evident from inspection of Fig. 2-19 that the origins of the coordinate
frames associated with the tool orientation {L3, L4} coincide at the wrist. This is in
dicad ve of the fact that the Alpha II robot has a spherical wrist. However, it is not a
general spherical wrist, because there is no tool yaw motion. The mechanical design
that enables the Alpha II robot to achieve a pitch-roll spherical wrist is shown in Fig.
2 - 20 .
The wrist design employs a set of three bevel gears {A, B, C} in a “universal”
joint configuration as shown in Fig. 2-20. Gears A and B are driven by sepárate
stepper motors and cables, while gear C is attached to the tool roll axis. If gears A
and B are driven in opposite directions at the same speed, the tool will exhibit a puré
roll motion. However, if gears A and B are driven in the same direction at the same
speed, the tool will exhibit a puré pitch motion with the tool tip moving “into” and
Pitch axis
Roll axis
• P
Tool
Figure 2-20 Pitch-roll spherical wrist
using bevel gears.

“out o f” the page. More complex motions involving simultaneous pitch and roll
movement can also be achieved. It is clear that the pitch and roll axes intersect at the
wrist.
2-6 THE ARM EQUATION
Once a set of link coordinates is assigned using the D-H algorithm, we can then
transform from coordinate ffame k to coordinate frame k - 1 using a homogeneous
coordinate transformation matrix. By multiplying several of these coordinate trans
formation matrices together, we arrive at a composite coordinate transformation ma
trix which transforms or maps tool coordinates into base coordinates. This com
posite homogeneous coordinate transformation matrix is called the arm matrix.
2-6-1 The Arm Matrix
There are four steps involved in constructing the homogeneous transformation ma
trix which maps frame k coordinates into frame k — 1 coordinates. Each of these
steps is associated with one of the four kinematic parameters summarized in Table
2-1. To determine the transformation matrix, we successively rotate and transíate
coordinate frame k — 1 to render it coincident with coordinate frame k. Using steps
8 to 12 of the D-H algorithm, this involves the four fundamental operations summa
rized in Table 2-3.
TABLE 2-3 TRANSFERRING FRAME k - 1

TO FRAME K
Operation Description
1 Rotate Lk- Xabout z k~l by 0k.

2 Transíate Lk~\ along z k~x by dk.
3* Transíate Lk-\ along x k~] by ak.
4 Rotate L*-i about x k~l by a k.
From Fig. 2-15 we see that the effect of operation 1 is to make axis j c * ' 1 paral-
lel to axis jc * . Following operation 2, axis jc * -1 is then aligned (collinear) with axis
jc * . Next, from Fig. 2-16 we observe that operation 3 ensures that the origins of
frames Lk~i and Lk coincide. Finally, the effect of operation 4 is to align axis z * '1
with axis z*. Thus the two coordinate frames Lk- 1 and Lk are coincident at this point.
The four fundamental operations summarized in Table 2-3 each can be interpreted as
a rotation or a translation of L*-i along one of its own unit vectors. It follows from
A lgorithm 2-5-1 that we can postmultiply I by the four fundamental homogeneous
transformation matrices to get the composite homogeneous transformation matrix
associated with Table 2-3. We see from Table 2-3 that operations 1 and 2 together
form a screw transformation along axis z * '1. In view of Exercise 2-4-3, operations 3
and 4 also form a screw transformation, in this case along axis j c * ' 1. Consequently,
the composite homogeneous transformation summarized in Table 2-3 can be ex-
Sec. 2-6 The Arm Equation 57

pressed as a composition oí two screw transformations, as follows:
71-1(0*, dky ak> a k) = Screw (dky 0*, 3) Screw (a*, a*, 1) (2-6-1)
Here 71-1 denotes the transformation from coordinate frame k to coordinate
frame k — 1. In general, T denotes a homogeneous coordinate transformation, the
superscript being the index of the source coordinate frame and the subscript being
the index of the destination, or reference, coordinate frame. As a consequence of the
systematic notation for assigning link coordinates in the D-H algorithm, a single
general expression for transforming between adjacent coordinate frames can be ob
tained. In particular, using Eq. (2-6-1) and the expression for the screw transforma
tion in Def. 2-4-2, we arrive at the following result, which employs the shorthand
notation, Sx = sin jc and Cx = eos jc:
Proposition 2-6-1: Link-Coordinate Transform ation. Let {Lo, Li, . . . ,

L n} be a set of link-coordinate frames assigned by Algorithm 2-5-1, and let [q]k and
[q]k~1 be the homogeneous coordinates of a point q with respect to frames Lk and
L*~i, respectively. Then, for 1 < k < n, we have [q]k~] = Tk-\[q]k, where:
cek —Ca¿ S6k Sa* S6k ak C0k
sek QockC6 k - S akC6 k ak Sdk
0 Sak Cak dk
0 0 0 1
Particular cases of the transformation matrix of Prop. 2-6-1 include 7o, which
maps shoulder coordinates into base coordinates; 71, which maps elbow coordinates
into shoulder coordinates; Ti, which maps wrist or tool yaw coordinates into elbow
coordinates, and so forth. If instead we want to move from the base toward the tool,
we need to use the transformation 7*-1, which is the inverse of 71-1.
Exercise 2-6-1: Inverse Link-Coordinate Transform ation. Utilize Prop.

2-4-1 and Prop. 2-6-1 to show that the following transformation maps link k - 1
coordinates into link k coordinates:
C$k S 6k 0 ~ak
-C a kS 6 k C ak C6 k Sak - d kSak
7T1=
SakS 6 k - S a kCOk Cak dk Ccc/t
0 0 0 1
Recall that three of the kinematic parameters that appear in the coordinate
transformation matrix 71-1 are constant, while the remaining parameter is the joint
variable. The joint variable will sometimes be 6 k (for revolute joints) and sometimes
be dk (for prismatic joints). Rather than treat the two cases separately, we introduce
a joint type parameter, defined as follows:
A joint k revolute
( 2 - 6 - 2)
joint k prismatic
Thus & is a binary-valued function of the joint type. With the aid of & we can then

define the kth joint variable qk as follows:
qk = + (1 - §j<f* (2-6-3)
It is clear that & has the effect of selecting either 0* or dk, as appropriate. In
general, the kth homogeneous link-coordinate transformation matrix Tí-1 is a func
tion of qk for 1 ^ k ^ n. For articulated-coordinate robots, we have q = 0, but for
all other robots q is a mixture of joint angles and joint distances. To solve the direct
kinematics problem, we must determine the position and orientation of the tool rela
tive to a coordinate ffame attached to the base. The transformation from tool coordi
nates to base coordinates is obtained by a succession of coordinate transformations
starting at the tool tip and working backward, one frame at a time, to the base. In
particular, if TSSl represents a transformation from tool-tip coordinates (link n) to
base coordinates (link 0), then:
TBUq) = T xo{qx)T]{q2) \ • • T U (q n) « Tn0(q) (2-6-4)
Note from the notation in Eq. (2-6-4) that when a sequence of coordinate
transformations is performed, the coordinate transformation algebra is such that di
agonal índices cancel. Consequently, for a composite transformation, the super
script on the rightmost factor is the identifier of the source coordinate frame, while
the subscript on the leftmost factor is the identifier of the destination, or reference,
coordinate ffame.
In order to compute the arm matrix, it is often helpful to partition the problem
at the wrist. This effectively decomposes the problem into two smaller subproblems,
one subproblem associated with the major axes used to position the tool, and the
other subproblem associated with the minor axes used to orient the tool. Here we
break the expression for the arm matrix into two factors at the wrist, or third axis, as
follows:
T£S(q) = Th(qx)T 2M 2)T\{q2) = Tl{q) (2-6-5)
TOJiq) » n ( q A)T\{qs) • • • Tnn-Mn) | V&q) (2-6-6)
For a general six-axis robot, Ttiqe) maps tool tip coordinates into roll coordi
nates, Tí(q5) maps roll coordinates into pitch coordinates, and maps pitch co
ordinates into yaw coordinates or wrist coordinates. Thus the composite transforma
tion Tz{q) maps tool tip coordinates into wrist coordinates. Similarly, T$q3) maps
wrist coordinates into elbow coordinates, T 2\(q2) maps elbow coordinates into shoul
der coordinates, and T l0(q$ maps shoulder coordinates into base coordinates. Thus
the composite transformation To(q) maps wrist coordinates into base coordinates.
Once the two wrist-partitioned factors are obtained, the general arm matrix is then
computed by forming their product, as follows:
TS£(q) = TERq 1, q2, q*)TSUqA, q5, . . . , qn) (2-6-7)
Again, notice that the diagonal identifiers in the composite transformation can
cel. If one interchanges the subscript and superscript of a given transformation, this
changes the direction of the transformation. Consequently, interchanging the sub
script and the superscript is equivalent to inverting the transformation matrix, as in
the following example:

TSSf(q) « [ r ¿ £ ( q ) ] - ' 12-6-8)
Recall from Prop. 2-4-1 that taking the inverse of a 4 x 4 homogeneous
transformation matrix is a simple matter which involves transposing a 3 x 3 matrix
and multiplying a 3 x 3 matrix by a 3 x 1 vector.
2-6-2 The Arm Equation
The reason for factoring the arm matrix into the product of two matrices partitioned
at the wrist is that closed-form expressions for the two factors can be obtained rela-
tively easily. At this point, the two matrices can be multiplied together and a closed-
form expression for the en tire arm matrix can then be obtained. Once an expression
for the arm matrix is available, we can then substitute it into the following matrix
equation, called the arm equation:
R(q) ¡ piq)
tool
r base (<?) = (2-6-9)
0 0 0 ! 1
For each valué of the joint vector q , the arm matrix T'SSU^) can be evaluated.
The 3 x 3 upper left submatrix R(q) specifies the orientation of the tool, while the
3 x 1 upper right submatrix p(q) specifies the position of the tool tip. The three
columns of R indicate the directions of the three unit vectors of the tool frame with
respect to the base frame. Similarly, p specifies the coordinates of the tool tip with
respect to the base frame. The solution of the direct kinematics problem in Eq.
(2-6-9) is shown schematically in Fig. 2-21.
Figure 2-21 Position and orientation of

the tool in base coordinates.
2-6-3 An Example: Microbot Alpha II
As an example of an arm matrix, consider the Alpha II robotic arm, whose link-
coordinate diagram was developed in Fig. 2-19. First we find closed-form expres
sions for the two wrist-partitioned factors of the arm matrix. We begin by computing
the wrist coordinates relative to the base. Using the notation C* É eos qk and S* =
sin qk and some trigonometric identities from Appendix 1, we have from Prop.
2-6-1 and Table 2-2:

c, 0 0 ' Q -S2 c3 -s3 0 fl3C3*
Ór*
-S i
<3
N
r wrist
base = T 0l T } T l = Si 0 c, 0 s2 c2 0 2 2 a S s3 c3 0 O S 3 3
0 -■1 0 di 0 0 1 0 0 0 i 0
0 0 0 1 0 0 0 1 0 0 0 1
c,c2 ~CiS2 -S i a2C,C2 C3 - S 3 0 C
13
C 3
- , -s,s2
S1C2 c1 «2S1C2 S 3 c3 0 üySx
-s2 ~c2 0 di — a S 2 2 0 0 1 0
0 0 0 1 0 0 0 1
C,C23 ~"CiS 23
___ I
5, Ci(fl2C2 4- a3C23)
S .C 23 S iS 23 ( S.(tf C2 2 + a3c23) (2-6-10)
-S 2 3 C 23 0 di — a 2S 2 Ü3S23
. 0 0 0 1
Here, to simplify the notation, we have used C*, 4 eos (qt + q,) and S*, 4
sin ( qk+ q¡).
Next, to compute the tool coordinates relative to the wrist, we have:
tool n r4 tt5
rwrist í 3 •* 4
c4 0 —s4 0 C5 -Sí 0 0
s4 0 c4 0 s5 c5 0 0
0 --1 0 0 0 0 1 di
0 0 0 1 0 1
r0
0
.
c4c5 c4s5 -s4 — d s S 4

s4c5 -s4s5 c:4 d $ Q 4 ( 2 - 6 - 11)
—Ss -Cs 0 0
0 0 0 1
Note that the final expression for depends only on the last two joint vari
ables. Similarly, the final expression for | ¡ f | | depends only on the first three joint
variables. To find the arm matrix, we multiply the two wrist-partitioned factors to
gether. The product, after simplification using the trigonometric identities in Ap-
pendix 1, and after substitution using the numerical valúes for d and a in Table 2-2,
is the following arm matrix:
C1C234C5 4- S1S5 ~ C jC 234S5 4 - — 234
s ,c 5 CiS C ,( 117 .8 C2 + 177 8 . 23
C - 9 6 . 5 S234)
S1C234C5 - c,s5 S1C234S5 — CiC5 —SiS234 S ,( 177 . 8 C2 + 117 . 8 C23 - 9 6 .5 S234)
—S234C5 S 234Ss -C i* 2 1 5 .9 - 177 . 8S2 - 177 . 8 S23 - 9 6 .5 C234
0 0 0 1
(2 - 6 - 12)
Here the notation Cijk is short for eos (qt + qj 4- qk) and the notation Sijk is short for
sin (q¡ + qj + <?*). Note that the approach vector r 3 and position vector p are inde
pendent of the tool roll angle qs, as expected.

c, 0 -s, 0' [ c * - S 2 0 q2Q2 C, —s3 O ¿J3C3
wrist
7
base = ThT\T\ = Si 0 c, 0 Sz c 2 0 a2S2 s 3 c3 0
A Ú3S3
0 *1 0 di ¡ 0 0 1 0 0 0 1 0
0 0 0 1 j .0 0 0 1 .0 0 O 1
‘ c,c2 - C 1S2 - s , tf2C,C2 c , ~s3 0 fl3c 3
= S1 C2 —s ,s 2 c 1 ¿j2SiC2 s3 c 3 0 £J3S3
“ S2 —c 2 0 di ~ a2$ 2 0 0 1 0
0 0 0 1 0 0 0 1
C1C23 "~CiS23 s, Ci(a2c 2 + ÚE3C23)
StC23 -s,s23 c Si(a2C2 + a3C23)
( 2 - 6 - 10)
-S» ~C23 0 di a2S2 ■ U3S23
L 0 0 0 l
simplify the notation, we have used Ckj = eos (qk + < 1j)
A
* ^ Hi)m
Next, to compute the tool coordinates relative to the wrist, we have:
rtool
wrist
'T’ 4 'T’S
i 3 -»
/3
c4 0
-s4 0'
1
C
00
1
Xn
A
Ul
s4 0
c4 0 s5 c5 0 0
0 -1
0 0 0 0 1 d5
0 0
0 1 ° 0 0 1
C4C5 c4s5 —s4 —Í/5S4
s4c5 -s4s5 c4 É/5C4 ( 2 - 6 - 11)
-s5 ~c5 0 0
0 0 0 1
Note that the final expression for r ^ 'sl depends only on the last two joint vari
ables. Similarly, the final expression for rgS? depends only on the first three joint
variables. To find the arm matrix, we multiply the two wrist-partitioned factors to
gether. The product, after simplification using the trigonometric identities in Ap-
pendix 1, and after substitution using the numerical valúes for d and a in Table 2-2,
is the following arm matrix:
C1C234C5 + S1S5 C1C234S3 + s ,c 5 - C 1S234 C,(117.8C2 + 177.8C23 - 96.5S234)
S1C234C5 - c ,s 5 —S1C234S5 —c ,c 5 —S1S234 S,(177.8C2 + II7.8C23 ~ 96.5S234)
—S234C3 S234S5 —C234 215.9 - 177.8S2 - 177.8S23 —96.5C234
0 0 0 1
( 2 - 6 - 12)
Here the notation Cljk is short for eos {q¡ + qj + q¡) and the notation Sijk is short for
sin (<?, + q} + qk). Note that the approach vector r 3 and position vector p are inde
pendent of the tool roll angle <75, as expected.

2-7 A FIVE-AXIS ARTICULATED ROBOT (RHINO XR-3)
In this section we perform a detailed kinematic analysis of a five-axis articulated

robot with tool pitch and roll motions. This class of robots includes both the Rhino
XR-3 educational robot and the Alpha II industrial robot as special cases. The Rhmo
XR-3 robot is shown in Fig. 2-22.
Figure 2-22 Five-axis articulated robot (Rhino XR-3). (Courtesy of Rhino

Robots, Inc., Champaign, IL. Robotic arm developed and manufacturad in the
USA.)
The physical structure of the Rhino XR-3 is such that the base motor is
mounted vertically and remains fixed, while the shoulder, elbow, and tool pitch mo-
tors are mounted horizontally on the body, which rotates about the base axis. The
tool roll and tool closure motors are smaller motors mounted in the hand.

2-7-1 The Link-Coordinate Diagram
We begin by constructing a link-coordinate diagram that is based on the Denavit-

Hartenberg algorithm. Applying steps 0 to 7 of the D-H algorithm to the Rhino XR-
3 robot, we get the diagram of link coordinates shown in Fig. 2-23.
Elbow Tool pitch
Figure 2-23 Link coordinates of a five-

axis articulated robot (Rhino XR-3).
Next we apply steps 8 to 13 of the D-H algorithm starting with k = 1. Using

Fig. 2-23, this yields the set of kinematic parameters shown in Table 2-4. Since this
is an articulated-coordinate robot, the vector of joint variables is q = 0. The valúes
of q listed in the last column of Table 2-4 correspond to the soft home position pic-
tured in the link-coordinate diagram of Fig. 2-23.
Note that ds represents the tool length, which may vary from robot to robot de-
pending upon which type of tool is installed. The valúes for the joint distances d and
link lengths a of the Rhino XR-3 robot are as follows (Hendrickson and Sandhu,
1986):
d = [260.4, 0.0, 0.0, 0.0, 171.5]T mm (2-7-1)
a = [0.0, 228.6, 228.6, 9.5, 0.0]r mm (2-7-2)
Here ds = 171.5 mm corresponds to the standard fingers. Optional fingers of
different lengths are available. If the robot is mounted on the optional aluminum
base, then di = 279.4 mm.
Next consider the problem of computing the arm matrix of the robot in Fig. 2-23.
First we partition the problem at the wrist to generate two simpler subproblems. Us
ing the notation C* A eos and S* A sin #*, we have from Prop. 2-6-1 and Table
2-4:
Sec. 2-7 A Five-Axis Articulated Robot (Rhino XR-3) 63

r wrist T 1T2 t 3
base i O* m 2
c, 0 -Si 0* c 2 - s 2 0 02 C2 c 3 - s 3 0 03 C3
Si 0 c, 0 s2 c 2 0 02 S2 S3 c3 0 03 S 3
0 -•1 0 di 0 0 1 0 0 0 1 0
0 0 0 1 0 0 0 1 .0 0 0 1
c ,c 2 -c,s2 -s Si Ci C2 c 3 - s•3 0 03 C3
s ,c 2 -S iS 2 c, 02Si C2 s3 c*3 0 03 S3
—S2 -C a 0 d\ Ü2 S2 0 0 1 0
0 0 0 1 0 0 0 1
CIC23 —Ct S23 s 1 C\{a2 C2 + 03C23)
S,C23 —Si S23 c\ S,(02C2 + 03C23)
(2-7
“ S23 C 23 0 d\ — a2 S2 — 03S23
0 0 0 1
Here, to simplify the notation, we have used C*, = eos (<7* + qj) and S*,
= sin {qk + qj). The transformation TbS? provides the position and orientation of
the wrist frame relative to the base frame. Since there is no tool yaw motion for this
five-axis robot, the wrist frame corresponds to the tool pitch frame As a partial
check of the expression for Tbas?, we can evalúate the transformation matrix at the
soft home position. From Table 2-4, this is q — [0, —ir /2, 7r / 2, 0, —7r / 2]r, which
yields:
1 0 0 03
0 0 1 0
r£Se(home) = 0 -1 0 d\ + a2 (2-7-4)
0 0 0 1
TABLE 2-4 KINEMATIC PARAMETERS OF A

FIVE-AXIS ARTICULATED ROBOT
Axis 0 d a a Home
1 dx 0 —tt/ 2 0
2 0 a2 0 —tt/ 2
3 <?3 0 as 0 7t/2
4 0 a4 —tí ¡ 2 0
5 ?5 d5 0 0 —7t/2
Inspection of Fig. 2-23 reveáis that this is consistent with the link-coordinate dia
gram. The first column of the rotation matrix indicates that, in the soft home posi
tion, axis x 3 of the tool pitch frame has coordinates (1, 0, 0) in the base frame. Con
sequently, it is pointing in the same direction as base axis x°. The second column of
R indicates that axis y 3 of the tool pitch ffame has coordinates (0, 0, - 1 ) in the base
ffame, in which case it is pointing in the opposite direction of base axis z°. The last

column o f R specifies that axis z 3 of the tool pitch ffame has coordinates (0, 1, 0)
relative to the base. Thus 2 3 points in the same direction as base axis y°. Finally, the
position vector p indicates that the coordinates of the origin of the tool pitch frame
L3 relative to the base frame Lo are (a3, 0, di + a2). Thus L3 is located a distance <23
in ffont o f the base and d\ + a 2 above the base. This is consistent with Fig. 2-23.
Next we determine the position and orientation of the tool relative to the wrist:
I
wnst
c4 0 -S 4 CÍ4 C4 c5 -s, 0
s4 0 C4 Ü4 S4 s5 C5 0
0 --1 0 0 0 0 1 ¿5
0 0 0 1 .0 0 0 1
C4C5 “C4 S 5 -s4 Cl4 C4 - ¿/5S4
S4C5 -S4S5
n1 c4 Q4 S4 + d$ C4 (2-7-5)
-s5 0 / 0
0 0 0 1
Note that the final expression for the position and orientation of the tool relative to
the wrist depends only on the last two joint variables, as expected. At this point we
could again make a partial check by evaluating this transformation at the home posi
tion. This should provide the position and orientation of frame L5 relative to ffame
L3. It remains to find the position and orientation of the tool relative to the base.
This is obtained by multiplying the two wrist-partitioned factors together. After sim
plification using the trigonometric identities in Appendix 1, we get the following
final result, which is the solution of the direct kinematics problem for the five-axis
articulated robot in Fig. 2-23:
Proposition 2-7-1: Rhino XR-3. The arm matrix Tb^Áq) for the five-axis
articulated-coordinate robot whose link-coordinate diagram is given in Fig. 2-23 is:
Cj C 234C 5 + Si s 5 -"C 1 C 234S 5 + S 1 C 5 -C , S 234 ' C 1 (í*2 C 2 ÍZ3 C 23 + ¿I4 C 234 ~ ds S 234)
Si C 234C 5 “ c , s 5 “ S 1 C 234S 5 — c , c 5 - 'S 1 S 234 J S 1(í?2C2 + ÍÍ3 C 23 + <24C 234 ~ S 234)
S 234 Cs S 234S 5 —C 234 | d\ — ¿I2 S 2 — fl3 S23 ~ ÍI4 S 234 ~ ds C 234
1
0 0 0 i 1
Note how the final expression for the position and orientation of the tool rela
tive to the base now depends on all of the kinematic parameters. We can make a
partial check by evaluating the arm matrix at the soft home position, which ffom
Table 2-4 is q = [0, —tt/2 , 7t/2, 0, —tt/2]7. This yields:
0 1 0 #3 + Ü4
1 0 0 0
rgSiChome) = 0 0 -1 d[ + a 2 - d5
—
.0 0 0 1

Inspection of Fig. 2-23 reveáis that this is consistent with the link-coordinate dia
gram. The first column o f the rotation matrix indicates that, in the soft home posi
tion, tool axis x 5 (the normal vector) points in the same direction as axis y ° of the
base. The second column o f R indicates that tool axis y 5 (the sliding vector) points
in the same direction as axis x° o f the base. The last column o f R specifies that tool
axis z 5 (the approach vector) points in the opposite direction from axis z° of the
base. Finally, the position vector p indicates that the origin o f the tool coordinate
ffame (the tool tip) is located a distance ai + ¿z4 in front of the base and
di 4* a 2 - di above the base when the robot is in the soft home position.
The solution to the direct kinematics problem in Prop. 2-7-1 is generic, in the
sense that all o f the nonzero joint distances and link lengths appear as explicit
parameters. Consequently, other robotic arms having a similar kinematic
configuration but different valúes for these kinematic parameters can also be ana-
lyzed using this general expression for the arm matrix. O f course, not all five-axis
articulated robots will fit this pattern. The Alpha II industrial robot is one that does,
as can be seen from the following example.
Example 2-7-1: Alpha II Robot

As an example of the use of the generic form of the five-axis articulated arm matrix,
consider the Alpha II robot, which is an articulated-coordinate robot that is sometimes
used in wafer-handiing applications in the semiconductor manufacturing industry. From
Table 2-2, the link twist angles are the same as those in Table 2-4. However, we have
the following joint distances and link lengths for the Alpha II robotic arm:
d = [215.9, 0.0, 0.0, 0.0, 96.5]r mm

a = [0.0, 177.8, 177.8, 0.0, 0.0]r mm
It is clear from these joint distances and link lengths that the Alpha II robot, unlike the
Rhino XR-3 robot, has a spherical wrist (a4 = d4 = 0). Thus the arm matrix for the
Alpha II is somewhat simpler than that for the Rhino XR-3. In particular, if we evalú
ate the generic expression for the arm matrix in Prop. 2-7-1 using the joint distances
and link lengths for the Alpha II, the result is:
Ci C;>34 C5 “f" S, s 5 c5
Cj C234 S 5 + Si —Cj S 234 I! C i(177.8C2 + 177.8C23 “ 9 6 . 5 S 234)
Si Ci34 C5 C, s5 —Si C234 S 5 — C, c3 S{ S234 !i S,(177.8Cj + I 7 7 . 8 C23 - 96.5S»,)
“ S234 C 5 S234 S 3 “ C234 ! 215.9 - 177.8S2 - 177.8S23 — 96.5C234
|
1
1
0 0 0 : 1
This is precisely the same result as obtained in Sec. 2-6, where an independent analysis
of the kinematics of the Alpha II robot was performed. Thus, in terms of its kinematic
structure, the Rhino XR-3 robot includes the Alpha II robot as a special case. As a par
tial check on this expression for the arm matrix for the Alpha II robot, we can evalúate
it at the soft home position, which, from Table 2-2, corresponds to q - [0, 0, 0,
—7t/2, 0]r . For these valúes of the joint variables, we have the following transforma
tion matrix. Inspection of Fig. 2-19 reveáis that this is indeed consistent with the link-
coordinate diagram.
Direct Kinematics: The Arm Equation Chap. 2

0 0 1 485.1
0 -1 0 0
TOUhome) = 1 0 0 215.9
0 0 0 1
2-7-3 Joint Coupling
It is apparent from the generic expression for the arm matrix in Prop. 2-7-1 that the
angles 023 = 02 + 03 and 0234 = 02 + 03 + 04 appear repeatedly. From Fig. 2-23 we
see that the angle 023 can be thought of as a global elbow angle. It is the elbow angle
measured relative to the work surface or jc°y° plañe. When 023 = 0, the forearm is
horizontal and pointing straight out, whereas when 023 = —7r/2, it is vertical and
pointing straight up. Similarly, the angle 0234 can be thought of as a global tool pitch
angle. It is the tool pitch angle measured relative to the work surface or x°y° plañe.
When 0234 = 0, the tool is vertical and pointing straight down, whereas when
0234 = — t t / 2 it is horizontal and pointing straight out.
The mechanical linkages for some robots, for example, the Rhino XR-3, are
designed in such a way as to couple the movement of the shoulder, elbow, and tool
pitch joints through a parallel bar mechanism. When the shoulder motor is activated
to produce a rotation of A02 = j8, the elbow moves simultaneously to produce a
counteracting rotation of A03 = —jS. In this way the global elbow angle 023 remains
fixed when the shoulder motor is activated. Consequently, the orientation of the
forearm relative to the base coordinate frame is unaffected by the shoulder motor.
Similarly, when the elbow motor is activated to produce a rotation of A03 = y, the
tool pitch moves simultaneously to produce a counteracting rotation of A04 = —y.
Thus the global tool pitch angle 0234 remains fixed when either the shoulder motor or
the elbow motor is activated. This means that the orientation of the tool relative to
the base coordinate frame is unaffected by the shoulder and the elbow motors. These
coupling effects can be useful when controlling the robot at the joint level. For ex
ample, if a liquid-filled container is carried by the tool, it will not spill on account of
tipping when either the shoulder motor or the elbow motor is activated.
The kinematic analysis of a robotic arm using the D-H algorithm is based upon
independent control of the joints with no coupling assumed between them. When
mechanical coupling does exist, it can be removed at the software level by prepro-
cessing the joint movement command A0 with a coupling matrix C and a precisión
matrix P. This produces a command vector A/i that specifies the number of encoder
counts to be sent to each motor starting with the base and ending with the tool roll.
For the Rhino XR-3 robot, the encoder counts required to produce a movement of
A0 degrees in joint space can be formulated as follows:
-1
Pi 0 0 0 0 1 0 0 0 0
0 P2 0 0 0 0 1 0 0 0
Ah 0 0 0 0 0 1 1 0 0 A0 = P-'C A0 (2-7-7)
0 0 0 p4 0 0 1 1 1 0
0 0 0 0 Ps 0 0 0 0 1

Here A8 is a vector with five components which specifies the desired change in the
five joint angles expressed in degrees. The coefficient matrix C is a coupling matrix.
Note the ls beiow the diagonal for the elbow command A/j3 and the tool pitch com
mand A/14. These off-diagonal terms effectively remove the coupling introduced by
the parallel bar mechanism. The matrix P~x is the inverse of a diagonal precisión
matrix. Here the p ’s specify the precisión of each joint expressed in degrees per en
coder count. From Table 1-11, we have:
p = [0.2269, 0.1135, 0.1135, 0.1135, 0.3125]7 degrees per count (2-7-8)
Note that the terms p3 and p4 appearing in Eq. (2-7-7) have negative signs preceding
them. This is because positive directions for the 0’s in the link-coordinate diagram
in Fig. 2-23 do not necessarily correspond to positive encoder counts in the Rhino
movement command (Hendrickson and Sandhu, 1986). Given the way the Rhino
XR-3 robots are wired at the factory, the base, shoulder, and tool roll motors have
positive encoder counts corresponding to positive joint angles, but the elbow and
tool pitch motors have negative encoder counts corresponding to positive joint an
gles; henee the minus signs to compénsate. Of course, the Rhino XR-3 hardware
could be modified by reversing the power leads to the motors and remounting
(flipping) the optical encoder disks. However, this is really not necessary, since the
introduction of minus signs in the software achieves the same result.
The vector Ah specifies the proper number of encoder hole counts to send to
each of the five motors, starting with the base and ending with the tool roll, in order
to achieve a move of A6 degrees in joint space. A matrix formulation of A/i is used
in Eq. (2-7-7) to highlight the interaxis coupling, which manifests itself through off-
diagonal terms in C. For a software implementation of Eq. (2-7-7), clearly it would
be more efficient to premultiply the matrix expression for Ah by hand to produce five
sepárate scalar equations for the individual components of Ah.
2-8 A FOUR-AXIS SCARA ROBOT (ADEPT ONE)
Another important type of robotic manipulator is the four-axis horizontal-jointed

robot, or SCARA robot. Examples of robots which belong to this general class in
clude the Adept One robot, the IBM 7545 robot, the Intelledex 440 robot, and the
Rhino SCARA robot. For illustration purposes we examine the Adept One robot,
shown in Fig. 2-24. The treatment, however, is generic, and consequently the
method used is applicable to SCARA robots in general.
The Adept One robot is unique in that it was the first commercial robot to im-
plement a direct drive system for actuation. No gears or other mechanical power
conversión devices are used. Instead, high-torque low-speed brushless DC motors
are used to drive the joints directly. This eliminates gear friction and backlash and
allows for clean, precise, high-speed operation.
2-8-1 The Unk-Coordinate Diagram
To construct the link-coordinate diagram, we apply steps 0 to 7 of the D-H al

gorithm. The resulting set of link coordinates for the SCARA class of robots is
shown in Fig. 2-25.

Figure 2-24 Four-axis SCARA robot (Adept One). (Courtesy of Adept Technol
ogy, Inc., San José, CA.)
Vertical
Elbow extensión
Figure 2-25 Link coordinates of a

four-axis SCARA robot (Adept One).
Base
Sec. 2-8 A Four-Axis SCARA Robot (Adept One) 69

In this case the vector of joint variables is q = [0*, 02, ¿fe, ¿fe]7. The first two
joint variables {ft, 62 } are revolute variables which establish the horizontal compo
nent of the tool position p. The third joint variable, ¿fe, is a prismatic variable which
determines the vertical component of p. Finally, the last joint variable, 04, is a revo
lute variable which Controls the tool orientation R. Applying steps 8 to 13 of the
D-H algorithm yields the kinematic parameters shown in Table 2-5.
Note that 7 of the 12 constant parameters in Table 2-5 are zero, which makes
the SCARA robot kinematically simple. Indeed, of all the practical industrial robots,
the SCARA robotis perhaps the simplest to analyze. Thevalúes for the joint dis
-
tances d and linklengths a of the Adept One robot are asfollows:
d = [877, 0.0, ¿fe, 200]r mm (2-8-1)
a = [425, 375, 0.0, O.Of mm (2-8-2)
Here a tool length of ¿fe = 200 mm has been assumed. Note that a specific valué for
¿fe has not been specified, because ¿fe is a joint variable in this case. The valúes for ¿fe
range from 0 to 195 mm, which implies a vertical stroke of 195 mm.
Next we compute the arm matrix Tb£?e(q) for the SCARA robot. Since there are only
four axes, we will not perform the intermedíate step of partitioning the problem at
the wrist. Instead we compute the arm matrix directly in one step, as follows:
7 tool
base
—
~
T i t 2 t 3 t 4
¿ 0 i l ¿ 2 ¿ 3
C, Si 0 üy C y c2 -s2 0 0,2 C2 1 0 0 0 c4 •s4 o o

S, -C y 0 ay Si s2 c2 0 Ü2S2 0 1 0 0 s4 c4 o o
0 0 -1 dy 0 0 1 0 0 0 1 q3 o 0 1 ¿fe
0 0 0 1 0 0 o 1 0 0 0 1 o o o 1
"C 1 -2 S i- 2 0 üyC 1 0 0 0' c4 - s 4 o o
S i —2 - C i - -2 0 ¿*iS 0 1 0 0 s4 c4 o o
0 0 -1 di 0 0 1 qi o 0 1 ¿fe
.0 0 0 1 0 0 0 1 o o o 1
" Ci —2 Si- ■2 0 üy C c4 -s4 0 0
S ,—2 - C i - -2 0 ay S s4 c4 0 0 (2-8-3)
0 0 -1 1 0 0 1 d
0 0 0 0 0 0 1
After multiplication and simplification using trigonometric identities from Ap-

pendix 1, we get the following arm matrix for a SCARA robot. Since the nonzero
joint distances d and link lengths a have been left as explicit parameters, the follow-
70 Direct Kinematics: The Arm Equation Chap *

OF A FOUR-AXIS SCARA ROBOT
Axis e d a a Home
1 d, a. ir 0
iébt <72 0 a2 0 0
3 0 é> 0 0 ISO 6
4 <?4 ¿4 0 0 7r/2
ing expression for the arm matrix is generic and applies to a variety of SCARA class
robots:
Proposition 2-8-1: Adept One. The arm matrix Tb¡¡c(q) for a four-axis
SCARA robot whose link-coordinate diagram is given in Fig. 2-25 is:
L _
U
0 Q\ C , + 0 2 C 1 -2
N
S i.-2 -4
1
1
00
-Ci - 2 -4 fli S i + 0 2 S 1 -2
(S
0
•'t
1
1
rtool
base
__
0 0 -1 di - <73 “ d4
0 0 0 1
Here the notation C 1- 2-4 denotes eos (<71 — q 2 — q*)> and similarly, S 1- 2-4 de
notes sin (<?i — qi — <74). Note that the approach vector is fixed at r3 = —i 3, inde
pendent of the joint variables. This is characteristic of SCARA robots, which are de
signed to manipúlate objeets from directly above. They are often used in light
assembly tasks such as the insertion of components into circuit boards.
Example 2-8-1: Adept One
Find the position of the tool tip of the Adept One robot when the joint variables are
q = [7t/4, -7r/3, 120, 7r/2]r.
Solution Using Prop. 2-8-1 and Eqs. (2-8-1) and (2-8-2), we have:
p = [a,C, + ¿Z2C1- 2, ai Sj + a2S,_2, d\ - qi - dA]r
= [0.707a, - 0.259a2, 0.707a, + 0.966a2, d, - d4 - 120f
= [203.4, 662.7, 557.0]7 mm
2-9 A SIX-AXIS ARTICULATED ROBOT (INTELLEDEX 660)
As a third example, we perform a kinematic analysis of a six-axis articulated indus

trial robot, the Intelledex 660 manipulator, shown in Fig. 2-26. The Intelledex 660
robot is a high-precision light-assembly industrial robot that uses closed-loop stepper
motor control to achieve a tool-tip repeatability of ±0.001 in. (±0.0254 mm). This
type of robot finds use in a variety of assembly and material-handling applications,
including wafer- and photomask-handling operations in clean rooms in the semicon
ductor manufacturing industry.
Sec. 2-9 A Six-Axis Articulated Robot (Intelledex 660) 71

Figure 2-26 Six-axis articulated robot (Intelledex 660T). (Courtesy o f Intelledex. Inc.,
Corvallis, OR )
The physical structure of the Intelledex 660 robot is quite unique. It is a full
six-axis robot» yet there are only two joints out near the end of the arm , a tool pitch
joint (“outboard") and a tool roll joint (“ tool spin”). Even though the wrist has only
two joints, arbitrary orientations of the tool are possible. Tool yaw motion is
achieved indírectly through a unique two-axis shoulder design. The first axis of the
shoulder rolls the entire arm , while the second axis achieves a pitch-type motion in
the plañe selected by the shoulder roll. Thus, to implement a puré horizontal yaw
m otion o f the tool, one can first roll the shoulder by 7r/2 radians and then activate
the tool pitch joint. By eñectively relocating the tool yaw joint in the shoulder, a re-
duction in the weight out at the end of the arm is realized. In some robots, such as
the M icrobot Alpha II and the GE P-50 robot, weight reduction at the end of the arm
is achieved by placing the actuators for the distal joints back near the base of the
robot and then running cables or chains up through the arm . The virtue of the In-
telledex design is that there is no need for mechanical power transmission through
the arm . Instead, the yaw motion is achieved indirectly through software.

2-9-1 The Link-Coordinate Diagram
First we construct a link-coordinate diagram based on the Denavit-Hartenberg al

gorithm . Applying steps 0 to 7 of the D-H algorithm to the Intelledex 660 robot, we
get the diagram of link coordinates shown in Fig. 2-27.
Elbow Tool pitch
Figure 2-27 Link coordinates of a six-axis articulated robot (Intelledex 660T).
Next we apply steps 8 to 13 of the D-H algorithm starting with k = 1. Using

Fig. 2-27, this yields the set of kinematic parameters shown in Table 2-6. Since this
is an articulated-coordinate robot, the vector of joint variables is q — 6 . The valúes
listed in Table 2-6 for 6 correspond to the soft home position pictured in the link-
coordinate diagram of Fig. 2-27.

OF A SIX-AXIS ARTICULATED ROBOT
Axis e d a a Home
1 <71 dx 0 7t/2 7t/2

2 <72 0 0 tt/2 — tt / 2
3 <73 0 a3 0 it /2
4 <74 0 Oa 0 0
5 <7s 0 0 tt/2 7t/2
6 <?6 de 0 0 0
Here 11 of the 18 fixed parameters are zero. In this case de represents the tool
length, which may vary from robot to robot, depending upon which type of tool is
installed. The valúes for the joint distances d and link lengths a of the Intelledex 660
robot are as follows:

[373.4, 0.0, 0.0, 0.0, 0.0, 228.6f mm (2-9-1)
a [0.0, 0.0, 304.8, 304.8, 0.0, O.Of mm (2-9-2)
Here d* — 228.6 mm corresponds to the standard pneumatic gripper. If, instead, the
small servo tool or some other tool is installed, then the calibration technique dis-
cussed in the operator’s manual should be used to measure d(>.
Next consider the problem of computing the arm matrix of the robot in Fig. 2-27.
First we partition the problem after the third axis, which, in this case, is the elbow
joint. Using Prop. 2-6-1 and Table 2-6, we have:
elbow 'T'l nr2 nr>3
7base * 0 í 1 -* 2
c, 0 s, 0 c 2 0 s2 0“ - s 3 0 #3C3
l
u
---------
fO
'
s, 0 -Ct 0 s2 0 -c2 0 s 3 c 3 0 03S3

0 1 0 di 0 1 0 0 0 0 1 0
0 0 0 1 0 0 0 1 0 0 0 1
c,c2 s, c,s2 o c3 —s3 0 03 C3
s,c2 -c, s,s2 0 s3 c3 0 0 3 S3
s2 o —c 2 d. 0 0 1 0
o o o 1 .0 0 0 1
Ci C2 C3 + Si S3 C1 C2 S3 + s,c3 C1 S2 (CiC2c3 + s,s3W
S1 C2 C3 - C1 S3 S1 C2 S3 - c,c3 s,s2 (S1 C2 C3 ~ Ci 83)0 3
S2 C3 -S 2 S3 -c 2 d\ + S2C303
o o o 1
(2-9-3)
The transformation TbíST provides the position and orientation of the elbow
frame L3 relative to the base frame L0. As a partial check of the expression for JáS",
we can evalúate it at the soft home position. From the last column of Table 2-6, this
is q = [tt/2, — tt/I, 7t / 2 , 0, 7t / 2 , 0]r, which yields:
1 0 0 03
0 0 -1 0 R
Lbibsew(home) = 0 1 0 di (2-9-4)
0 0 0 1
Inspection of Fig. 2-27 reveáis that this specification of the position and orientation
of ffame L3 relative to ffame Lo is consistent with the link-coordinate diagram.
Next we determine the position and orientation of the tool relative to the el
bow:
74 Direct Kinematics: The Arm Equation Chap. 2 S

rssu | n n n
C4 -s4 0
L-----
u
0
Oa Q a s5 0 c6 s6 o o
40
S4 c4 0 s4 s5 0 “ Cs 0 s6 c6 0 o
0 0 1 0 0 1 0 0 o o 1 do
0 0 0 1 .0 0 0 1 o o O 1
C45 0 S45 CI4 c4 C. - ■s6 0 o
S45 0 C45 Ü4 S4 s6 c6 0 0
0 1 0 0 0 0 1 do
0 0 0 1 0 0 0 1
(2-9-5)
C45 c 6 “ C 45 s 6 •S45 c¡4 C4 -f s,15 6
S4 5 C6 "“S45 s6 - c45 ¿24 s4 - G \sd o
s6 c6 0 0
0 0 0 1
Note that the final expression for the position and orientation of the tool rela
tive to the elbow depends only on the last three joint variables, as expected. At this
point we could again make a partial check by evaluating this transformation at the
soft home position. This should provide the position and orientation of frame Lo rel
ative to frame L3. We still need to find the position and orientation of the tool rela
tive to the base. This is obtained by multiplying the two elbow-partitioned factors
together. After considerable simplification using the trigonometric identities in Ap-
pendix 1, we get the following final result, which is the solution of the direct kine
matics problem for the six-axis Intelledex 660 robot:
Proposition 2-9-1: Intelledex 660. The position p(q) and orientation R(q)
of the tool ffame relative to the base frame for the six-axis articulated-coordinate
robot whose link-coordinate diagram is given in Fig. 2-27 are:
Cl C3 + ¿Í4C34 ■
+■¿¿6 S345) + S i(fl3S 3 + #4S 34 ~ C 34s)
P= S 1 C2O33 C3 + Ú4C34 + d(,Sj4s) —Ci(<33 S 3 + Q4S 34 “ ¿6 C 34s)
d\ + 82(03 C 3 + 04 C34 + í?6 S345)
(C 1C 2 C 3 4 5 + S i S 345) C 6 + C j S2 Sg C 1S 2 C 6 “ (C1C2C345 + S 1 S345)Sg •“ S1C345 + C i C2 S 345
R = (S i C2 C 345 ~ C i S 345)C6 + S( S2 Só S 1S 2 C 6 — (SiC2C345 ~ C , S 345)S 6 C1C345 + S1C2S345
S2 C 345 C é ~ C2 Sg ~ C2 Cñ S2C345S6 S2 S345
It is evident that the expressions for the approach vector r 3(q) and the tool-tip
position p (q ) are independent of the tool roll angle q6, as expected. We can make a
partial check of the final expression for the arm matrix by evaluating it at the soft
home position, which corresponds to q = [tt/2 , - n / 2 , 7t/2, 0, 7t/2, 0]r. This
yields:

" 0 0 1 03 + 04 + dé
0 -1 0 0
7b£Uhome) = 1 0 0 dx
0 0 0 1
Note that this is consistent with the link-coordinate diagram in Fig. 2-27. The first
column of the rotation matrix indicates that, in the soft home position, tool axis jc6
(the normal vector) points in the same direction as axis z° of the base. The second *
column of R indicates that tool axis y 6 (the sliding vector) points in the opposite di
rection from axis y° of the base. The last column of R specifies that tool axis z6 (the
approach vector) points in the same direction as axis x° of the base. Finally, the po
sition vector p indicates that the origin of the tool coordinate ffame (the tool tip) is
located a distance <a3 + a4 + d 6 in front of the base and d\ abo ve the base when the
robot is in the soft home position.
2-10 PROBLEMS
2-1. Which of the kinematic parameters are variable for a revolute joint? Which are vari
able for a prismatic joint?
2-2. Consider the single-axis robot in Fig. 2-28 shown in the home position, which corre
sponds to 0 = 7r/2. Suppose the point p on the mobile link has coordinates
[p]M = [0.5, 0.5, 2.0f.
(a) Find an expression for R(0), the coordinate transformation matrix which maps mo
bile M coordinates into fixed F coordinates as a function of the joint variable 6 .
(b) Use R( 6 ) to find [p]F when 0 = 7r and when 0 = 0.
Figure 2-28 Single-axis robot.
2-3. Consider the following coordinate transformation matrix, which represents a funda
mental rotation. What is the axis of rotation (1, 2, or 3), and what is the angle of rota- i
tion?
0.500 0 -0.866'
R = 0 1 0
0.866 0 0.500
76 Direct Kinematics: The Arm Equation Chap. *

2-4. Consider the robotic tool shown in Fig 2-29. Suppose we yaw the tool by tt about / ‘,
then pitch the tool by - t t /2 about / 2, and finally roll the tool by t t /2 a b o u t /\
(a) Sketch the sequence of tool positions after each of the yaw, pitch, and roll move-
ments.
(b) Find the transformation matrix T which maps tool coordinates M into wrist coordi
nates F following the sequence of rotations.
(c) Find [p]Fy the location of the tool tip p in wrist coordinates F following the se
quence of rotations, assuming that the tool-tip coordinates in terms of the tool
frame are [p]M = [0, 0, 0.8]r.
2-5. Do Prob. 2-4 but with the yaw, pitch, and roll of the tool performed in reverse order
about the unit vectors of the M coordinate frame rather than about the unit vectors of
the F coordinate frame.
Tool
Roll
G f nr
Yaw Pitch

m tool.
2-6. Consider the following composite homogeneous coordinate transformation matrix T:
R ! p
0 0 0 |1
Assuming that R and p represent a rotation and a translation of the mobile frame about
its own unit vectors, which of the following sequence of operations does the composite
transformation T represent? That is, what is the implicit order of the two fundamental
operations?
(a) T represents a rotation of R followed by a translation of p?
(b) T represents a translation of p followed by a rotation of /??
(c) T represents neither in general?
2-7. Find a numerical example which shows that the order in which a rotation and a transla
tion are performed does affect the final relationship between two initially coincident
coordinate frames. Specify the coordinate transformation matrices, and sketch the se
quence of positions.
2-8. Homogeneous coordinate transformation matrices can be used to represent scaling op
erations as well as rotations and translations. Consider the following fundamental ho
mogeneous scaling matrix:
c1 0 0 0
0 C2 0 0
Scale (c, cr) 4 0 0 C3 0
_0 0 0 a
Sec. 2-1 0 Problem s

77
(a) Show that when a * 0 and * - | l . i, l) r, Scale (c, cr) scales all three physical
coordinates by 1ja He re ¡r is referred to as a global scale factor
(b) Show that Scale (c l) wales the Árth physical coordínate by ck for 1 s k s 3.
Here c is referred to a s a vector of local scale factors.
(c) Show that when cr * 0 and c = [cr, cr, tr]7, the local and global scale factors can
cel one another, in which case multiplication by Scale (c. cr) has no effect on the
physical coordinates.
2-9. Consider the Ummation PUMA 200 manipulator, shown in Fig. 2-30. This is a six-axis
articulated robot with a roll-pitch-roll type of spherical wrist. Assign link coordinates
using the first half of the D-H algorithm Label the diagram with a 's and d's as appro
priate.
Figure 2-30 Unimation PUMA 200 robot. (Courtesy of Westinghouse Automa

tion División, Pittsburgh, PA.)
2-10. Use the second half of the D-H algorithm to fill in the kinematic parameters for the
PUMA 200 robot in Table 2-7, consistent with your link-coordinate diagram from
Prob. 2-9. Indicate which parameters are the joint variables.
TABLE 2-7 P U M A 200 KINEM ATIC PARAMETERS
Axis 6 d a a Home
1
2
3
4
5
6
78 D irect K inem atics: T h e A rm Equation Chap. 2

2-11. Compute the first factor of the arm matrix T££(q) for the PUMA 200 robot using your
table of kinematic parameters from Prob. 2-10.
2-12. Compute the second factor of the arm matrix Tl£ tv(q) for the PUMA 200 robot using
your table of kinematic parameters from Prob. 2-10.
2-13. Compute the arm matrix Tl^ ( q ) for the PUMA 200 robot using your wrist-partitioned
factors from Prob. 2-11 and Prob. 2-12.
2-14. An alternative way to specify the tool orientation is the Z-T-Z Euler angle representa
tion. The mobile frame M and fixed frame F start out coincident. Frame M is first ro
tated about m 3 by an angle of a; it is then rotated about m2 by an angle of j8; and
finally, frame M is again rotated about m3 by an angle of y. Find the composite Z-K-Z
Euler angle rotation matrix RZYZ(a y /3, y) which maps mobile M coordinates into fixed
F coordinates.
2-15. Consider the United States Robots Maker 110 manipulator shown in Fig. 2-31. This is
a five-axis spherical coordinate robot with a pitch-roll spherical wrist. Assign link co
ordinates using the first half of the D-H algorithm. Label the diagram with a ’s and d 's
as appropriate.
Figure 2-31 United States Robots Maker 1 10 robot. (Courtesy of United States Robots,
King o f Prussia, P A .)
2-16. Use the last half of the D-H algorithm to fill in the table of kinematic parameters for
the Maker 110 in Table 2-8, consistent with your link-coordinate diagram for Prob.
2-15. Indicate which parameters are the joint variables.
2-17. Compute the first tactor of the arm matrix T\£f(q) for the Maker 110 robot using your
table of kinematic parameters from Prob. 2-16.
2-18. Compute the second factor of the arm matrix TOUg) for the Maker 110 robot using
your table of kinematic parameters from Prob. 2-16.
Sec. 2-10 Problem s 79

TABLE 2-8 MAKER 110 KINEMATIC PARAMETERS
Axis 6 d a a Home
1
2
3
4
5
2-19. Compute the arm matrix 7b£¡.(#) for the Maker 110 robot using your wrist-partitioned
factors from Prob. 2-17 and Prob. 2-18.
REFERENCES
Craig, J. (1986). Introduction to Robotics: Mechantes and Control, Addison-Wesley: Read-

ing, Mass.
D e n a v it, J., and R. S. H a rte n b e rg (1955). “A kinematic notation for lower-pair mechanisms
based on matrices,” A J. Appl. Mech., June, pp. 215-221.
Fu, K. S., R. C. G o n z á le z , and C. S. G. L e e (1987). Robotics: Control, Sensing, Vision, and
Intelligence, McGraw-Hill: New York.
H e n d ric k so n , T., and H. S a n d h u (1986). XR-3 Robot Arm, Mark I I I 8 Axis Controller:
Owner’s Manual, Versión 3.00, Rhino Robots, Inc.: Champaign, 111.
N o b le , B. (1969). Applied Linear Algebra, Prentice-Hall, Inc.: Englewood Cliffs, N.J.
Paul, R. P. (1982). Robot Manipulators: Mathematics, Programming, and Control, MIT
Press: Cambridge, Mass.
S c h illin g , R. J. and H. L e e (1988). Engineering Analysis: A Vector Space Approach, Wiley:
New York.
80 Direct Kinematics: The Arm Equation Chap- 2

3
In v e rse Kinematics:
Solving
The Arm Equation
In Chap. 2 we developed a procedure for determining the position and orientation of

the tool of a robotic manipulator given the vector of joint variables. We now exam
ine the inverse problem of determining the joint variables given a desired position
and orientation for the tool. The inverse kinematics problem is important because
manipulation tasks are naturally formulated in terms of the desired tool position and
orientation. This is the case, for example, when external sensors such as overhead
cameras are used to plan robot motion. The information provided by the camera is
not in terms of joint variables; it specifies the positions and orientations of the ob-
jects that are to be manipulated.
The inverse kinematics problem is more difficult than the direct kinematics
problem because a systematic closed-form solution applicable to robots in general is
not available. Moreover, when closed-form Solutions to the arm equation can be
found, they are seldom unique. The ways in which múltiple Solutions arise are inves-
tigated. A compact representation of tool position and orientation called the tool-
configuration vector is then introduced. Solutions to the arm equation for four
generic classes of robots are presented. Representative members of these classes in
clude: the five-axis Rhino XR-3, the four-axis Adept One, the six-axis Intelledex
660, and a three-axis planar articulated arm. Chapter 3 concludes with an example
o f an application of inverse kinematics to plan motion in a robotic work cell using
information available from an overhead camera.
3'1 THE INVERSE KINEMATICS PROBLEM
The key to the solution of the direct kinematics problem was the Denavit-Hartenberg
(D-H) algorithm, a systematic procedure for assigning link coordinates to a robotic
manipulator. Successive transformations between adjacent coordinate frames, start-
81
ing at the tool tip and working back to the base of the robot, then led to the arm ma
trix. The arm matrix represents the position p and orientation R of the tool in the \
base frame as a function of the joint variables q, as shown in Fig. 3-1.
<*3 (94.......9„}
Figure 3-1 Tool conñguration as a

function of joint variables.
For convenience, we will refer to the position and orientation of the tool col-
lectively as the configuration of the tool. The solution to the direct kinematics prob
lem can be expressed in the following form, where R represents the rotation and p
represents the translation of the tool ffame relative to the base frame:
R !
T'S&iq) I (3-1-1)
0 0 0 I 1
The solution to the direct kinematics problem is useful because it provides us with a
relationship which explicitly shows the dependence of the tool configuration on the
joint variables. This can be utilized, for example, in determining the size and shape
of the work envelope. In particular, suppose Q represents the range of valúes in R"
that the joint variables can assume. We refer to Q as the joint-space work envelope of
the robot. Typically, the joint-space work envelope is a convex polyhedron in Rn of
the following general form:
Q= q{ GR": q mm < <
Here q mn and # raax are constant vectors in Rm which represent joint limits and C is a
joint coupling matrix. For the simplest special case, the joint coupling matrix is
C = /; in this case q f in and qf™ represent lower and upper limits, respectively, for
joint k. More generally, if the joint coupling matrix C has some off-diagonal terms,
then Eq. (3-1-2) represents constraints on linear combinations of joint variables. In
this case the number of inequality constraints m may exceed the number of variables
n. The set of positions in Cartesian space R3 that are reachable by the tool tip can be
investigated by examining the valúes of p(q) as q ranges over Q. Similarly, the set of
orientations achievable by the tool can be determined by examining the valúes of
R(q) as q ranges over the joint-space work envelope Q.
Perhaps the most important benefit provided by the solution of the direct kine
matics problem is that it lays a foundation for solving a related problem, the inverse
82 Inverse Kinematics: Solving the Arm Equation Chap. 3

kinematics problem. The vector of joint variables q is restricted to the subset Q of
R". We refer to the vector space R" in this case as joint space. Similarly, the tool-
configuration parameters {/?, p) can be associated with a subset W of R 6. We refer to
the vector space R 6 in this case as tool-configuration space. Tool-configuration space
is six-dimensional because arbitrary configurations of the tool can be specified by us
ing three position coordinates ( p i, /?2, p 3) together with three orientation coordi
nates (yaw, pitch, roll). Solving the direct kinematics problem is equivalent to
finding the mapping from joint space to tool-configuration space, while solving the
inverse kinematics problem is equivalent to finding an inverse mapping from tool-
configuration space back to joint space. A pictorial summary of the relationship be
tween the two forms of the kinematics problem is shown in Fig. 3-2.
Direct {p. R}
kinematics
equations
Joint space Tool-configuration

space R 6
I nverse {p. R}
kinematics
equations
Figure 3-2 Direct and inverse kinematics.
The inverse kinematics problem is more difficult than the direct kinematics
problem, because no single explicit systematic procedure analogous to the D-H al
gorithm is available. As a result, each robot or generically similar class of robots has
to be treated separately. However, the solution to the inverse kinematics problem is
much more useful. Indeed, a key to making robots more versatile lies in using feed-
back from external sensors such as tactile arrays and visión. External sensors supply
information about part location and part orientation directly in terms of
configuration-space variables. It is then necessary to use this information to deter
mine appropriate valúes for the joint variables to properly configure the tool. Thus
we must find a mapping from a tool-configuration space input specification into a
joint-space output specification. This is the inverse kinematics problem. We charac
terize this problem more formally as follows:
Problem : Inverse Kinematics. Given a desired position p and orientation R

for the tool, find valúes for the joint variables q which satisfy the arm equation in
Eq. (3-1-1).
The solution to the inverse kinematics problem is useful even when external
sensors are not employed. A case in point is the problem of getting the end-effector
or tool to follow a straight-line path. Certain types of assembly, welding, and sealing
operations require that straight-line paths be followed or at least closely approxi-
mated. A straight-line trajectory is naturally formulated in tool-configuration space.
It is then necessary to find a corresponding trajectory in joint space which will pro-
Sec. 3-t The Inverse Kinematics Problem 83

duce the straight-line motion of the tool tip. This is an important special case of the
inverse kinematics problem.
3-2 GENERAL PROPERTIES OF SOLUTIONS
Although a general solution has been obtained for the direct kinematics problem, the
details of an explicit solution to the inverse kinematics problem depend upon the
robot or the class of robots being investigated. However, there are certain character
istics of the solution that hold in general.
3-2-1 Existence of Solutions
First let us examine conditions under which Solutions to the inverse kinematics prob
lem exist. Clearly, if the desired tool-tip position p is outside its work envelope, then
no solution can exist. Furthermore, even when p is within the work envelope, there
may be certain tool orientations R which are not realizable without violating one or
more of the joint variable limits. Indeed, if the robot has fewer than three degrees of
freedom to orient the tool, then whole classes of orientations are unrealizable. To ex
amine this issue further, consider the following more detailed formulation of the
arm equation:
Su Rn Rn j P\
Ri\ R 22 Rn ¡ P2
T g£ (q) = (3-2-1)
R“
i\ R 32 R 33 1 P3
o o o ¡ 1
Since the last row of the arm matrix is always constant, the arm equation constitutes
a system of 12 simultaneous nonlinear algebraic equations in the n unknown compo
nents of q.However, the 12 equations are by no meansindependent of one another.
Since R is acoordinate transformation matrix representing a purérotation from one
orthonormal frame to another, it follows from Prop. 2-2-3 that R~x = R T. But if
R TR — / , then the three columns of R form an orthonormal set. The mutual orthog
onality puts three constraints on the columns of R , namely:
r 1• r 2 = 0 (3-2-2)
r 1• r 3 = 0 (3-2-3)
r r3 = 0 (3-2-4)
These constraints come from the off-diagonal terms of R TR gj I. The diagonal terms
dictate that each column of R must also be a unit vector. This adds three more con
straints to the columns of R:
r = 1 1< k < 3 (3-2-5)
Thus the 12 constraints implicit in the arm equation actually represent only six inde
pendent constraints on the n unknown components of the vector of joint variables q.
If we are to have a general solution to the inverse kinematics problem, one for which

a q can be found which generates an arbitrary tool configuration, then the number of
unknowns must at least match the number of independent constraints. That is:
General manipulation ^ n > 6 (3-2-6)
This lower bound on the number of axes n is a necessary but not sufficient condition
for the existence of a solution to the inverse kinematics problem when arbitrary tool
configurations are specified. Clearly, the tool position must be within the work en
velope of the robot, and the tool orientation must be such that none of the limits on
the joint variables are violated. Even when these additional constraints on the valúes
of p and R are satisfied, there is no guarantee that a closed-form expression for a so
lution to the inverse kinematics problem can be obtained (Pieper, 1968).
The general strategy for solving the inverse kinematics problem simplifies
somewhat when the robot has a spherical wrist. To see this, consider the case of an
n-axis robot where 4 < n < 6. Suppose the last axis is a tool roll axis, and suppose
the robot has a spherical wrist, which means that the n — 3 axes at the end of the
arm all intersect at a point. For this class of robots, the inverse kinematics problem
can be decomposed into two smaller subproblems by partitioning the original prob
lem at the wrist. Given the tool tip position p and tool orientation /?, the wrist posi
tion p wnst can be inferred from p by working backward along the approach vector:
pwnst « p __ dy (3-2-7)
Here the joint distance dn represents the tool length for an «-axis robot as long as the
last axis is a tool roll axis. The approach vector r 3 is simply the third column of the
rotation matrix R. Once the wrist position p wnst is obtained from {p, i?, d„}y the first
three joint variables {«q\, q2, #3} that are used to position the wrist can then be ob
tained from the following reduced arm equation:
(3-2-8)
The fourth column of 7*2? represents the homogeneous coordinates of the origin of
the wrist ffame L3 relative to the base ffame Lo. Since the wrist coordinates depend
only on the joint variables {#1, q2, ^3}, these joint variables of the major axes can be
solved for separately using Eq. (3-2-8). Once the major axis variables {qx, q2> qy)
are found, their valúes can then be substituted into the general arm equation in Eq.
(3-2-1) and it can be solved for the remaining tool orientation variables {q4,
3-2-2 Uniqueness of Solutions
The existence of a solution to the inverse kinematics problem is not the only issue
that needs to be addressed. When Solutions do exist, typically they are not unique.
Indeed, múltiple Solutions can arise in a number of ways. For example, some robots
are designed with n axes where n > 6. For these robots, infinitely many Solutions to
the inverse kinematics problem typically exist. We refer to robots with more than six
axes as kinematically redundant robots, because they have more degrees of ffeedom
than are necessary to establish arbitrary tool configurations. These extra degrees of
Sec. 3-2 General Properties of Solutions 85

freedom add flexibility to the manipulator. For example, a redundant robot might be
commanded to reach around an obstacle and manipúlate an otherwise inaccessible
object. Here some of the degrees of freedom can be used to avoid the obstacle while
the remaining degrees of freedom are used to configure the tool. Consider, for ex
ample, the top view of a redundant SCARA robot shown in Fig. 3-3. Since only two
joints are needed to establish the horizontal position of the tool, the second elbow
joint of the SCARA robot is redundant.
Redundant
elbow
Figure 3-3 Reaching around an obsta

cle with a redundant robot.
Even when a robot is not kinematically redundant, there are often circum-
stances in which the solution to the inverse kinematics problem is not unique. Sev-
eral distinct Solutions can arise when the size of the joint-space work envelope Q is
sufficiently large. As a case in point, consider the articulated-coordinate robot
shown in Fig. 3-4. If the limits on the range of travel for the shoulder, elbow, and
tool pitch joints are sufficiently large, then two distinct Solutions exist for the simple
task of placing the tool out in ffont of the robot.
Elbow up
Figure 3-4 Múltiple Solutions with a

nonredundant robot.
We refer to the two Solutions in Fig. 3-4 as the elbow-up and the elbow-down
solution. In tool-configuration space the two Solutions are identical, because they
produce the same p and R, but in joint space they are clearly distinct. Typically the
elbow-up solution is preferred, because it reduces the chance of a collision between
the links of the arm and obstacles resting on the work surface.

3-3 TOOL CONFIGURATION
In order to develop a solution to the inverse kinematics problem, the desired tool
configuration must be specified as input data. Thus far we have assumed that the tool
configuration is represented by the pair {p> /?}, where p represents the tool position
relative to the base and R represents the tool orientation relative to the base. Speci-
fying tool-tip position with a translation vector p is a natural and convenient tech-
nique. However, specifying tool orientation with a rotation matrix R is, at best, awk-
ward, because two-thirds of the information that must be provided is redundant. In
this section we introduce an alternative representation of tool configuration that is
more compact.
3-3-1 Tool-Configuration Vector
Consider the tool orientation information provided by the approach vector or last
column of R. The approach vector effectively specifies both the tool yaw angle and
the tool pitch angle, but not the tool roll angle, because the roll angle represents a
rotation about the approach vector. Here, for convenience, the terms yaw, pitch, and
roll are being used to refer to rotations of the tool about the normal vector, sliding
vector, and approach vector, respectively. It would be more precise to refer to them
as turn, tilt, and twist angles (Whitney, 1972).
Given that the approach vector specifies the tool orientation except for the tool
roll angle, some method must be found to augment the approach vector with the roll
angle information. If we simply append the roll angle, this yields a total of four com
ponents, which is not a minimal representation of tool orientation. Instead, note
from Eq. (3-2-5) that the approach vector r 3 is a unit vector specifying a direction
only. The size or length of the approach vector can be scaled by a positive quantity
without changing the specified direction. This is the key to encoding the tool roll in
formation into the approach vector to produce a minimal representation of orienta
tion.
To recover the tool roll angle from a scaled approach vector, we must use an
invertibie function of the roll angle q„ to scale the length of r 3. The range of travel
of the tool roll joint is specified in the joint-space work envelope Q in Eq. (3-1-2).
Often this range is bounded because of the presence of cables running from the wrist
to sensors mounted at the tool tip. However, even when qn is unbounded, the follow
ing positive, invertible, exponential scaling function can be used:
/((?„) = e x p — (3-3-1)
7T
Other scaling functions might also be used, but the function/is sufficient be-
cause/(g,,) > 0 for all qn and f ~ l(qn) is well defined. If we scale the approach vector
r 3 by f { q n), this yields a compact representation of tool orientation. We then com
bine this representation of orientation with the tool tip position and refer to the re
sulting vector in R6 as a tool-configuration vector.
Sec. 3-3 Tool Configuration 87

Definition 3-3-1: Tool-C onfiguration Vector. Let p and R denote the posi
tion and orientation of the tool frame relative to the base frame where qn represents
the tool roll angle. Then the tool-configuration vector is a vector w in R6 defined:
w1 P
A A
w _[exp(<?„/7r)r}]_
Since the tool-configuration vector has only six components, it is a minimal

representation of tool configuration in the general case. The first three components,
w 1 = p , represent the tool-tip position, while the last three components,
w 2 = [exp (#rt/7r)]r3, represent the tool orientation. The tool roll angle can be eas-
ily recovered from the tool-configuration vector w, as can be seen from the follow
ing exercise:
E xercise 3-3-1: Tool Roll. Show that the tool roll angle qn can be obtained
from the tool-configuration vector w as follows:
qn = ir ln (w4 + W5 + h>6)1/2
The tool-configuration vector w provides us with a more convenient way to

specify the desired tool position and orientation as input data to the inverse kinemat
ics problem. Furthermore, the structure of the tool-configuration vector can reveal
useftil qualitative information about the types of positions and orientations achiev-
able by the tool. To see this, we examine a five-axis articulated robot.
3-3-2 Tool Configuration of a Five-Axis Articulated Robot
Many industrial robots have five axes; some have only four. Even though these ma
nipulators are not general in the sense of being able to generate arbitrary tool
configurations, they are still quite versatile. However, when robotic work cells em
ploy ing these robots are planned, the layout of the part feeders and other fixtures
must be constrained so they do not require the flexibility of a full six-axis robot. As
an illustration, consider a five-axis articulated-coordinate robot which has no tool
yaw motion. In this case, the part feeders might be arranged concentrically around
the robot base as shown in Fig. 3-5.
The part feeders can be tilted at any angle y that is consistent with the range of
global tool pitch angles realizable by the robot. However, the part feeders must be
oriented in such a way as to be aligned with radial lines or spokes emanating from
the base axis. This is because with the tool yaw angle fixed as in Fig. 3-5, the ap
proach vector of the tool is constrained to lie in a vertical plañe through the base
axis. This constraint can be summarized algebraically in terms of the components of
the arm matrix as follows:
R u p i = Rupx (3-3-2)
Here the ratio of the first two components of the approach vector r 3 must equal the
ratio of the first two components of the position vector p. In geometric terms, the

Figure 3-5 Part presentation for a five-
OA
axis articulated robot.
tool can manipúlate objects from above, from the front, from the back, and even
from below consistent with work-envelope constraints and limits on the range of tool
pitch angles. However, the five-axis robot shown in Fig. 3-5 cannot manipúlate ob
jects from the side. That is, the orthogonal projection of the approach vector onto
the work-surface or jt°y° plañe cannot have any component that is tangent to a circle
of radius || p || centered at the base axis. Another way to view the constraint on tool
orientation is to substitute Eq. (3-3-2) in the general expression for the tool-
configuration vector in Def. 3-1-1. This yields the following locus of tool
configurations, where /3 is a constant:
w = [wi, w2, vv3, /3wi, /3w2, w6]t (3-3-3)
Thus, for example, if the tool tip is directly in front of the robot (w2 = 0), the ap
proach vector must point forward (vv5 = 0). More generally, the ratio of the hori
zontal components of the tool-tip position, wi/w2, must equal the ratio of the hori
zontal components of the approach vector, w4/w 5. It is evident by inspection of the
formulation in Eq. (3-3-3) that this robot has only five degrees of freedom, because
components vv4 and w5 cannot be specified independently of wi and w2.
3-3-3 Tool Configurations of a Four-Axis SCARA Robot
The most important special case for manipulating parts corresponds to approaching
parts from directly above. This includes picking up a part from a horizontal work
surface and placing it down on a horizontal work surface. As an example of a robot
designed for these types of operations, consider the four-axis horizontal-jointed or
SCARA robot shown in Fig. 3-6. Here the tool yaw and tool pitch angles are fixed
in such a way that the tool always points straight down.
Recall that the five-axis robot of Fig. 3-5 had an approach vector r 3 that was
restricted to lie in a vertical plañe through the base axis. The geometry of a SCARA
robot removes one more degree of freedom and restricts the approach vector to lie
along a vertical line within that plañe, namely:
r3 = -i3 (3-3-4)
The four-axis SCARA robot in Fig. 3-6 is a minimal configuration for a gen-
eral-purpose robot. The first three axes are the major axes used to position the tool
tip, while the fourth axis is a minor axis used to orient the tool by varying the slid-
Sec. 3-3 Tool Configuration 89

□ J
rn
Figure 3-6 Part presentation for a four-axis SCARA robot.
ing vector. If Eq. (3-3-4) is substituted in the general expression for the tool-
configuration vector in Def. 3-3-1, this yields:
w p i, p 2, p 3, 0, 0, -ex p (3-3-5)
The four degrees of freedom that are available are evident from inspection of Eq. (3-
3-5), where we see that two of the components of the tool-configuration vector are
always zero. The SCARA robot is a particularly simple robot in terms of analysis,
yet it is a practical robot for many applications.
3-4 INVERSE KINEMATICS OF A FIVE-AXIS ARTICULATED ROBOT

(RHINO XR-3)
The solution to the inverse kinematics problem can be approached either numeri-
cally or analytically. The numerical approaches (Uicker et al., 1964) employ search
techniques to invert the tool-configuration function w (q) by solving the following op-
timization problem, where w E R6 is the desired valué for the tool configuration:
Minimize: v{q) = ||w(<?) — vv ||
subject to: q E Q
If a q E Q can be found such that v(q) = 0, then clearly q is a solution to the in
verse kinematics problem. To solve the optimization problem, a sequence of in-
creasingly accurate approximate Solutions {q°, q \ . . .} is constructed starting from
an initial guess q°. If the initial guess is sufficiently cióse to an actual solution, the
sequence may converge fairly rapidly. However, for a general q°, the sequence will
converge only to a local minimum of v(q), a minimum for which v(q) may or may
not be zero. The virtue of the numerical approach lies in the fact that it is general, in
the sense that the same basic algorithm can be applied to many different manipula
tors; only w(q) and Q vary. However, the drawbacks are that it is a comparatively
slow technique, and furthermore it produces only a single, approximate solution for
each initial guess q°.

The second basic approach to solving the inverse kinematics problem is to ex*
ploit the specific nature of the direct kinematic equations to determine an analytical
or closed*form expression for the solution. These exact methods are considerably
faster than the numerical approximation methods, and they can be used to identify
múltiple Solutions. The main drawback of the analytical methods is that they are
robot-dependent. Consequently a sepárate analysis has to be performed for each
robot or generic class of robots (Pieper, 1968).
We ¿Ilústrate the analytical solution technique by applying it to several different
types of robots. The first class of robots we consider are the five-axis vertically-
jointed or articulated robots with tool pitch and tool roll motion. This class of robots
includes the Rhino XR-3 robot, the Microbot Alpha II robot, and several other com-
mercial robotic manipulators as special cases. The link-coordinate diagram for a
five-axis articulated robot was previously developed as Fig. 2-23 in Chap. 2. For
convenient reference, we redisplay it here as Fig. 3-7.
Elbow Tool pitch
Figure 3-7 Link coordinates of a five-

axis articulated robot (Rhino XR-3).
The solution to the inverse kinematics problem starts with the expression for
the tool-configuration vector w(q), which can be obtained from the arm matrix de
veloped in Chap. 2. From Prop. 2-7-1 and Def. 3-3-1, the tool-configuration vector
for the five-axis articulated arm is:
C i( # 2C 2 + 0 3 C 23 + 04C234 ~ ^5 S234)
S i(# 2C 2 + ÍÍ3C23 "F ÍÍ4C234 ““ ^58234)
d
\ —CI2S2 ~
~Q3S23~
~ 04S234 ~ ¿/5C234
w (q) (3-4-1)
- [ e x p ^ s / tt)]C 1S 234
-[e x p (g 5/ 7r ) ] S iS 234
—[ e x p ( ^ / t t ) ] C 234
Since this is an articulated robot, q = 0. Recall that the notation Cy* is short
for eos (q, + qj + qk) and similarly Sy* denotes sin (q¡ + q} -F <?*). In order to ex-
tract the jo in t variable vector q from the nonlinear tool-configuration function w(q ),
Sec. 3-4 Inverse Kinematics of a 5-Axis Articulated Robot (Rhino XR-3) 91

we appiy row operations to the components of w and exploit various trigonometric
identities from Appendix 1 in order to isolate the individual joint variables.
3-4-1 Base Joint
Since there is no yaw motion, the easiest joint variable to extract is the base angle qu
Inspection of the expressions for w\ and W2 in Eq. (3-4-1) reveáis that they have a
factor in common. If we divide wi by Wi, this factor caneéis, and we are left with
S i/C i. Thus the base angle is simply:
#i = atan2 (w2, wj) (3-4-2)
The function atan2 in Eq. (3-4-2) denotes a four-quadrant versión of the arc-
tan function. The atan2 function allows us to recover angles over the entire range
[ 7r, 7r]. This function is used repeatedly in subsequent derivations. It can be im-
plemented with the more common two-quadrant arctan function as indicated in
Table 3-1, where the notation sgn represents the signum, or sign, function. That is,
sgn (a) retums —1, 0 , or 1 depending upon whether a is negative, zero, or positive,
respectively.
TABLE 3-1 FOUR-QUADRANT ARCTAN FUNCTION
Case Quadrants atan2 ( y , jc)
x > 0 1 ,4 arctan ( y / x )
x = 0 1 ,4 [sgn (y)] t t / 2
x < 0 2 ,3 arctan { y / x ) + [sgn (y)] tt
3-4-2 Elbow Joint
The elbow angle #3 is the most difficult joint variable to extract, because it is strongly
coupled with the shoulder and tool pitch angles in a vertical-jointed robot. We begin
by isolating anintermediate variable, #2:34, called the global tool pitch angle. Here
#234 = #2 + <73+ <74 is the tool pitch angle measured relative to the work surface or
x°y° plañe. Inspection of the last three components of w in Eq. (3-4-1) reveáis that
—(C1W4 + Sih>5) /( —vv6) = S234/C 234. Since the base angle #1 is already known, the
global tool pitch angle can then be computed using:
#234 = atan2 [ - ( C 1W4 + Si\v5), ~w 6] (3-4-3)
Once the shoulder angle # 2 and elbow angle #3 are known, the tool pitch angle
#4 can be computed from the global tool pitch angle #234. In order to isolate the
shoulder and elbow angles, we define the following two intermediate variables:
b\ = C1W1 + Siw2 - ¿Z4C234 + ¿/5S234 (3-4-4)
b2 ~ d\ — {Z4S234 ~ ds C234 ~ W3 (3-4-5)
Note that b\ and b2 are constants whose valúes are known at this point because
#1 and #234 have already been determined. If we take the expressions for the compo-
nents of w in Eq. (3-4-1) and substitute them in the expressions for b\ and b2y this

yields:
b\ = a2C2 + ÍÍ3C23 (3-4-6)

b2 = 122S2 4* 1Z3S23 (3-4-7)
We are now left with two independent expressions involving the shoulder and
elbow angles; the coupling with the tool pitch angle has been removed. The elbow
angle can be isolated by computing ||¿»||2. Using trigonometric identities from Ap-
pendix 1 , we find, after some simplification, that:
|| ¿7 1|2 = a\ + IdidiC* + al (3-4-8)
Thus the cosine of the elbow angle has been determined. If we now solve Eq.
(3-4-8) to recover #3 , this results in two possible Solutions, as illustrated previously
in Fig. 3-4. The first one is the elbow-up solution, and the second one is the elbow-
down solution:
43 = ± árceos M - ai - a) (3 -4 -9 )
In view of Eq. (3-4-9), the solution to the inverse kinematics problem for a
five-axis articulated-coordinate robot is not unique. In most instances, we use the
elbow-up solution, because this keeps the elbow joint further away from the work
surface. Indeed, for some commercial robots, this is the only solution allowed,
because of joint limit restrictions contained in Q.
3-4-3 Shoulder Joint
To isolate the shoulderangle q2, we return to the expressions in Eqs. (3-4-6) and
(3-4-7),which represent b\ and b2 in terms of the shoulder andelbowangles. If we
expand C23 and S23, using the cosine of the sum and sine of the sum trigonometric
identities, and rearrange the terms, this yields:
bx = (a2 + a3C3)C2 - (a3S3)S2 (3-4-10)
b2 = (a2 + a 3C3)S2 + (<z3S3)C2 (3-4-11)
Since the elbow angle q3 is already known, Eqs. (3-4-10) and (3-4-11) consti-
tute a system of two simultaneous linear equations in the unknowns C2 and S2. If we
use row operations to solve this linear system, the result is:
„ _ (a2+ a,C,)bi
v-2 — li b\\2
_ (a2 + aiCi)b2 - a 3S3í>! ,, .

S2 = ------------ ¡r^ii------------ (3-4-13)
Since we have expressions for both the cosine and the sine of the shoulder an
gle, we can now recover the shoulder angle over the complete range [—tt, 7r] using
the atan2 function. In particular:
q2 = atan2 [(a2 + a3 C3)¿>2 — a 3S3¿?i, (a2 + a 3C3)¿?i -f a 3 S3¿?2] (3-4-14)
Sec. 3-4 Inverse Kinematics of a 5-Axis Articulated Robot (Rhino XR-3) 93

3-4-4 Tool Pitch Joint
8
I
The work for extracting the tool pitch angle #4 is already in place. We know the
shoulder angle #2, the elbow angle #3, and the global tool pitch angle #234. Thus:
#4 = #234 - #2 - #3 (3-4-15)
3-4-5 Tool Roll Joint qr5
The final joint variable is # 5 , the tool roll angle. This can be recovered from the last
three components of w, as indicated previously in Exercise 3-3-1. In this case, we
have:
#5 = 7r In (h>4 + + w l ) í/2 (3 -4 -1 6 )
Recall that the tool-configuration vector w is constructed from the position

vector p and a scaled versión of the approach vector r 3. In certain cases, it may be •
useful to specify the rotation matrix R and then compute the tool roll angle q5 from
it. To see how this might be done for the five-axis articulated robot, recall from
Prop. 2-7-1 that the rotation matrix is:
CiC234C5 + S 1S5 C 1C234S5 + S1C5 C 1S234

R{q) S1C234C5 ~ C 1S5 S,C 234S5 - C ,C 5 - S , S 234 (3 -4 -1 7 )
—S234C5 S234S5 ““C234
On the surface, it appears that the easiest way to recover #5 from R is to use the
last row of R , where R32A —-^31) = S5/C 5. However, the ratio R31/R 32 is undefined
whenever the global tool pitch angle # 2 3 4 is a múltiple of t t . Rather than pursue this
approach further by decomposing it into special cases, we instead consider the first
two components of the normal vector r l. Taking Si times the first component and
subtracting Ci times the second component, the result is:
11 1*V21 (3 -4 -1 8 )
From Eq. (3-4-18) we can compute #5 over the range [ — t t / 2 , t t / 2 ] . Since we !

want a general expression for the tool roll angle, one that is valid over the entire
range [ — t t , t t ] , we also need to determine C5. Here we consider the first two com
ponents of the sliding vector r 2. Taking Si times the first component and subtracting
Ci time the second component, the result is:
C5 — Si R n Ci1R
A.22 (3 -4 -1 9 )
Given both the cosine and the sine of the tool roll angle, we can now evalúate
#5 over the complete range [ — t t , t t ] using the atan2 function. In particular, we
have:
#5 — atan2 (Si/?n C 1R21» S 1R12 Ci/?22) (3 -4 -2 0 )
94 Inverse Kinematics: Solving the Arm Equation Chap-3

3-4-6 Complete Solution
The complete inverse kinematics solution for a five-axis articulated robot with tool
pitch and tool roll motion is summarized in Algorithm 3-4-1. Note that this solution
applies to both the Rhino XR-3 robot and the Alpha II robot as long as appropriate
valúes for the joint distance vector d and link length vector a are used. In either
case, the parameter d$ represents the tool length, which may vary depending upon
which of the various types of optional fingers are used. The two expressions for <?3
specified by the plus and minus signs represent the elbow-up and elbow-down Solu
tions, respectively.
Algorithm 3-4-1: Inverse Kinematics of a Five-Axis Articulated

Robot (Rhino XR-3)
The solution to the inverse kinematics problem outlined in Algorithm 3-4-1

assumes that the tool-configuration vector w is supplied as input data. If instead
the tool position and orientation {/?,#} are supplied directly, then w ¡ |
[ p i , p i , p s , #13, #23, # 3 s F can be used to compute {^i, q2, #3, q*} from Algorithm
3-4-1 and q 5 can be computed from # using Eq. (3-4-20). A third possibility is to
use something called a reduced tool-configuration vector w as input to the inverse
kinematics algorithm. For an «-axis robot with n < 6, a reduced tool-configuration
vector is a vector in R” rather than R6. This way each of the components of w can be
independently specified. In the case of a five-axis articulated robot with tool pitch
and tool roll motion, the following is an example of a reduced tool-configuration
vector
w = [ p u p 2 i P3, <?234, qsY (3-4-21)
See. 3-4 Inverse Kinematics of a 5-Axis Articulated Robot (Rhino XR-3) 95

Here the tool-tip position p is supplied and two parameters describing tool orienta
tion are added. The first is the global tool pitch angle q234 = qi + <73 + q4, which is
the tool pitch measured relative to the work surface or jc°y° plañe. The second is the
tool roll angle q$. If the reduced tool-configuration vector vv is used as input data,
then Algorithm 3-4-1 simplifies with #234 = vv4 and q$ = vv5.
Exercise 3-4-1: Home Position. For the five-axis articulated robot in Fig.
3-7, use Algorithm 3-4-1 to find q when the robot is in the following position,
called the home position. Does your answer agree with the last column of Table 2-4?
w = [¿*3 + 04, 0, d\ + a2 — dsy 0, 0, —0.607]r
Exercise 3-4-2: Zero Position. For the five-axis articulated robot in Fig. 3-7,
use Algorithm 3-4-1 to find q when the robot is in the following position, called
the zero position:
w = [a2 + 03 + 04, 0, d\ — ú?5, 0, 0, —l]r
Exercise 3-4-3: Máximum Reach. For the five-axis articulated robot in Fig.
3-7, use Algorithm 3-4-1 to find q when the robot is in the following positions:
1. A máximum horizontal reach position

2. A máximum vertical reach position
3-5 INVERSE KINEMATICS OF A FOUR-AXIS SCARA ROBOT

(ADEPT ONE)
Another important class of robotic manipulators for which a generic solution to the
inverse kinematics problem can be obtained is the four-axis horizontal-jointed
robot, or SCARA robot. Examples of robots which belong to this class include
the Intelledex 440 robot, the IBM 7545 robot, and the Adept One robot. The link-
coordinate diagram for a four-axis SCARA robot was previously developed as Fig.
2-25 in Chap. 2. For convenient reference, we redisplay it as Fig. 3-8.
Unlike the vertical-jointed robot, the vector of joint variables in this case is
not 6 but is instead q = [0i, 02, <¿3, 04]T. The arm matrix for the SCARA robot has
been previously computed in Chap. 2. From Prop. 2-8-1 and Def. 3-3-1, we get the
following expression for the tool-configuration vector for the SCARA robot:
01C1 1 c
"“t2C\ - 2
tfiSi + 02S1-2
d\ ~ q^ ~ d4
w{q) (3-5-1)
0
0
-e x p { q j tt)

Vertical
Elbow extensión
Figure 3-8 Link coordinates of a four-

axis SCARA robot (Adept One).
Recall that C 1 -2 is short for eos {q\ — q2) and S 1 -2 denotes sin (^1 - q2). Note
from the last three components of w that the approach vector r 3 is fixed at r 3 = - t 3.
This is characteristic of SCARA-type robots in general, which always approach parts
directly fromabove. In order to extract the joint angle vector q from the nonlinear
tool-configuration function w(q), we again apply row operations to the components
of w and exploit trigonometric identities from Appendix 1 in order to isolate the in
dividual joint variables.
3-5-1 Elbow Joint
We can extract the elbow angle q2 from the first two components of the tool-
configuration vector w. Note that w] + w2 represents the square of the radial com
ponent of the tool position, a quantity which should be independent of the base
angle q\. If we compute w? + W2 from Eq. (3-5-1), then after simplification using
trigonometric identities the result is:
h>i + W2 — a\ + 2a\a2C2 + <22 (3-5-2)
Thus the cosine of the elbow angle has been determined in terms of known con-
stants. If we now solve Eq. (3-5-2) to recover q2, this results in two possible Solu
tions. The first one is called the left-handed solution, and the second one is the
right-handed solution:
w\ + wi - a\ - al
q2 = ± árceos --------------- (3-5-3)
2a\a2
Sec. 3-5 Inverse Kinematics of a 4-Axis SCARA Robot (Adept One) 97

It follows from Eq. (3-5-3) that the solution to the inverse kinematics problem
for a four-axis SCARA robot is not unique. The motivation for the terms left-handed
and right-handed can be seen in Fig. 3-8. If the robot is viewed from above, then,
when q 2 > 0 , the robotic arm curls in from the left, whereas when q2 < 0, it curls
in from the right.
3-5-2 Base Joint
The base angle q i9 which might also be thought of as a horizontal shoulder angle,
can now be obtained from q2. First we take the expressions for and w2 in Eq.
(3-5-1), and expand the C 1-2 and S i-2 terms using the cosine of the difference and
sine of the difference trigonometric identities in Appendix 1, After rearranging the
coefficients, this yields:
(ai + a 2C2)Ci + (a2S2)Si = wi (3-5-4)
( - a 2S2)C, + (a, + a2C2)Si = w2 (3-5-5)
Since the elbow angle qi is already known, this is a systemof two simultaneous lin
ear equations in the unknowns Ci and Si. If we use row operations to solve this lin
ear system, the result is:
c - fl2S2Wi + {a\+ a2C2)w2 n ,
(a2S2)2 + (fll + a2C2)2 {' ' )
C =ÍS + Ü2^ Wl ~ fl2S2W2 , .

(a2S2)2 + (ax + a 2 C 2)2 { )
Since we have expressions for both the cosine and the sine of the base angle, we can
again recover the base angle over the complete range [ —77*, 77] using the atan2 func
tion:
q\ = atan2 [a2S2wi + (a\ + a 2 C 2)w2, (ai + a2C2)wi — a2S2w2] (3-5-8)
3-5-3 Vertical Extensión Joint
The prismatic joint variable q3 is associated with sliding the tool up and down along
the tool roll axis. In a SCARA-type robot, the vertical component of the tool mo
tion is uncoupled from the horizontal component, as can be seen from the expres
sion forwin Eq. (3-5-1). Extracting the prismaticjoint variable #3 is therefore a
simple matter:
q 3 = di - d 4 - w3 (3-5-9)
3-5-4 Tool Roll Joint
The final joint variable is the tool roll angle q4. Inspection of the last component of
w reveáis that q4 can be easily recovered from it as follows:
q4 = 77 ln |u>6| (3-5-10)
1
As an alternataive to specifying vv, one could instead specify the rotation ma
trix R and then compute the tool roll angle <74 from it. To see how this might be done
for the four-axis SCARA robot, recall from Prop. 2-8-1 that the rotation matrix is:
C1-2-4 S i- 2-4 0
R(q) S1-2-4 “C l- 2-4 0 (3-5-11)
0 0 -1
The angle <71- 2-4 is called the global tool roll angle. It is the tool roll angle
measured relative to the base frame of the robot. Note from the normal vector in Eq.
(3-5-11) that R21/R11 = S1-2-4/C1-2-4. Thus the global tool roll angle can be com
puted from R as follows:
<7 1 - 2 -4 = atan2 (R21, /fu ) (3-5-12)
Once the global tool roll angle is known, the tool roll angle <74 can be com
puted directly from it because q\ and <72 are also known:
<74 — <71 “* #2 — <71-2-4 (3-5-13)
3. 5.5 Complete Solution
The complete inverse kinematics solution for a four-axis SCARA-type robot is sum
marized in Algorithm 3-5-1. The fixed joint distances d and the link lengths a will
depend on the particular robot. These parameters can be obtained from specification
sheets supplied by the robot manufacturer. The two expressions for q 2 specified by
the plus and minus signs represent the left-handed and right-handed Solutions, re
spectively.
Algorithm 3-5-1: Inverse Kinematics of a Four-Axis SCARA Robot

(Adept One)

is based on the assumption that the tool-configuration vector w is supplied as input
data. As an alternative, the pair {p, R} can be specified. In this case
Sec. 3-5 Inverse Kinematics of a 4-Axis SCARA Robot (Adept One) 99

w = Cpi* Pi» P3, # 13, # 23, # 33]r can be used to compute {#i, qi, #3} from Algorithm
3-5-1 and then # 4 can be computed from R using Eqs. (3-5-12) and (3-5-13). A
third possibility is to again use a reduced tool-configuration vector vv as input to the
inverse kinematics algorithm. In the case of a four-axis SCARA robot, the follow
ing vector in R4 is an example of a reduced tool-configuration vector:
vv = [ p u p 2 i P3, q i- 2 - 4]T (3-5-14)
Here the tool-tip position p is supplied and one additional parameter describing tool
orientation is added. It is the global tool roll angle # 1 - 2 -4 which is the tool roll mea
sured relative to the base frame. When vv4 = 0 , the jc axes of the tool and the base
point in the same direction and positive valúes for vv4 correspond to counterclock-
wise tool rotations as seen from above. If the reduced tool-configuration vector is
used as input data, Algorithm 3-5-1 can be used to compute {q 1, #2, #3}, and #4 is
computed from Eq. (3-5-13) with # 1- 2-4 = vv4.
Exercise 3-5-1: M inimum Reach. For the four-axis SCARA robot, suppose
the upper arm and the forearm are of equal length (02 = a\). Find the joint variables
# when the tool tip reaches back and just touches the base axis, as follows:
w = [0, 0, 0, 0, 0, - l ] r
Exercise 3-5-2: M áximum Reach. For the four-axis SCARA robot, find the
joint variables # when the radial extensión of the arm is máximum, as follows:
vv = [ai + a2 , 0, 0, 0, 0, —l]r
3-6 INVERSE KINEMATICS OF A SIX-AXIS ARTICULATED ROBOT

(INTELLEDEX 660)
As a third example of computing an inverse kinematic solution, consider the six-axis

Intelledex 660 articulated robot. The link-coordinate diagram for the Intelledex 660
was previously developed as Fig. 2-27 in Chap. 2. For convenient reference, it is re-
displayed here as Fig. 3-9.
Since this is an articulated robot, # = 0. The solution to the inverse kinemat
ics problem starts with the expression for the tool-configuration vector vv(#), which
can be obtained from the arm matrix developed in Chap. 2. From Prop. 2-9-1 and
Def. 3-3-1, the tool-configuration vector for the Intelledex 660 robotic arm is:
C i C 2 (C3ÚE3 + C34Ú4 + S 345*4 ) + S i( S 3 f l3 + S34CZ4 — C345 d¿)
S l C 2 ( C 3<23 + C34ÍZ4 + S 345*4) ~ C j (S3ÍZ3 + S34Ú4 — C 345¿4 )
di + S2(C3fi3 + C34Ú4 + S ^ d t )
w(q) (3-6-1)
[e x p ( # 6 / ,w)]("—S i C 3 4 5 + C 1C 2 S 3 4 5 )
[e x p (# 6 / f r ) ] ( C i C 345 + S 1C 2S 34 5 )
[e x p (# 6 / ' 7r ) ] S 2 S 345
Recall that the notation Cljk is short for eos (#, -f q¡ + #*) and, similarly, S,>*

Figure 3-9 Link coordinates of a six-axis articulated robot (Intelledex 660T).
denotes sin (#, + q¿ + qk). We also make use of the expression for the rotation ma
trix R (q) developed in Prop. 2-9-1:
( C 1 C 2 C 345 + S iS 345)C 6 + C 1 S2 S6 C 1 S2 C 6 — (CJC 2C 345 + SjS345)S6 1 —S 1C 345 + C 1 C 2 S345
R = (S 1 C 2 C 345 ~ C j 8345)06 + S 1 S2 S6 S 1 S2C 6 ~ (S 1 C 2C 345 — C |S 345)S 6 1 C 1 C 345 + S 1 C 2 S 345
S 2 C 345C 6 — C 2 S6 —C 2C 6 “ S2 C 345S6 828345
(3-6-2)
In order to extract the joint variable vector q from the tool-configuration vector
w(q) and the rotation matrix R{q), we apply row operations to vv and R and exploit a
host of trigonometric identities in order to isolate the individual joint variables.
The Intelledex 660 robot is quite complex, and in order to develop closed-form ex
pressions for the joint variables, we assume that the tool configuration is available
both in the form of the tool-configuration vector vv and the translation and rotation
pair {p, /?}. From Exercise 3-3-1, the tool roll angle q6 can be recovered from the
last three components of the tool-configuration vector vv as follows:
#6 = tt ln (vv4 + vv2 + w l ) 1/2 (3-6-3)
3-6-2 Shoulder Roll Joint
Given the tool roll angle, the shoulder roll angle #2 can be computed from the third
components of the normal vector r ‘ and the sliding vector r 2. If we formúlate
# 31S6 + # 32C6, this causes the common term involving C234 to drop out. The re-
Sec. 3-6 Inverse Kinematics of a 6-Axis Articulated Robot (Intelledex 660) 101
mainder, after simplification using Cl + S | = 1, is simply —C2 . Consequently, we
can solve for <72 as follows:
q2 = ±árceos ( - R 3iS6 — RnCe) (3-6-4)
Clearly, the solution is not unique. Recall from Table 2-6 that in the home po
sition, q2 = —7 t / 2 . We therefore choose the negative solution and restrict our atten-
tion to angles in the range
—7r < q2 < 0 (3-6-5)
The solution corresponding to q2 < 0 is called the left-handed solution, be-
cause, from Fig. 2-26, it corresponds to the upper arm being on the left side of the
small shoulder and forearm. If the shoulder is rolled by t t to produce angles in the
range 0 < q2 < t t , this places the arm in the right-handed configuration. The end
points {—7T, 0} are not included in Eq. (3-6-5), for reasons which will become ap-
parent as we proceed.
3-6-3 Base Joint
To extract the base angle <71 , we make use of the first and second components of the
normal vector r 1 and the sliding vector r 2. First we compute the quantity
/?uS 6 + R 2 2 C 6 . The result, after simplification using the identity S i + C l = 1, is:
/?n Só “F /^nCó == C1S2 (3- 6-6)
Recall from Eqs. (3-6-4) and (3-6-5) that S2 is a known nonzero quantity. But
q6 is also known, from Eq. (3-6-3), and henee we have isolated Ci. To recover the
base angle over the entire range [ —7r, 7r], we must also isolate Si. This can be
achieved with the second components of the normal and sliding vector. Again using
the trigonometric identity S i + C l = 1, we find:
-R21 Só "1" R22C6 == S1S2 (3- 6-7)
If we now divide Eq. (3-6-7) by Eq. (3-6-6), the nonzero factor S2 caneéis
and we are left with an expression for S 1 /C 1 . Thus the base angle qx over the range
[ 7r, 7r] is:
I
q x — atan2 (/?2iSó + R 2 2 C$y RuSó + /^nCé) (3-6-8) 1
3-6-4 Elbow Joint
Perhaps themost difficult joint angle to extract is the elbow angle <74, which is tightly
bound to the shoulder pitch angle <73 and the tool pitch angle <75. We can remove the
effeets of the tool pitch angle q 5 by partitioning the problem at the wrist, as sug-
gested in Eqs. (3-2-7) and (3-2-8). If we subtract from the tool tip position p ,
this yields the wrist position relative to the base. To simplify the subsequent equa
tions, we also subtract d \i 3 and denote the result as b. That is,
b = p — d¿r3 — d \i3 (3-6-9)

The first two components of the intermediate variable b specify the x and y co
ordinates of the wrist ffame L4 relative to the base ffame Lo. The last component of b
specifies the vertical distance from the wrist frame L4 to the shoulder ffame L2. If we
substitute for p and r 3 in Eq. (3-6-9), we get the following expressions for the com
ponents of b:
b\ = CiC2(C3a 3 + C34ÍÍ4) + Si(S3tf3 + 834^4) (3-6-10)
b 2 = SiC2(C303 + C3404) “ Ci(S3fl3 + S3404) (3-6-11)
bi = S2(C303 + C34fl4) (3-6-12)
Note how the effects of the tool pitch angle qs have been removed from the
original expression for p. The quantity || b || represents the straight-line distance from
the shoulder joint to the wrist joint. Because the elbow joint is the only joint be
tween the shoulder and the wrist, ||¿>|| should depend exclusively on q4. If we com
pute \\b\\2,the result after simplification using various trigonometric identities in
Appendix 1 is:
\\b ||2 — + 2a3tf4C4 + al (3-6-13)
Thus ||¿?|| depends exclusively on q4y as expected. Solving Eq. (3-6-13) for q4 then
yields:
g4 = ±arccos M . 7. ? L ~ * (3-6-14)
2¿Z3Ú4
Again we see that the solution is not unique. The positive solution returns
angles in the range [0 , ir] and is referred to as the elbow-down solution, while the
negative solution, which produces angles in the range [ - 7r, 0], is the called the el
bow-up solution.
3-6-5 Shoulder Pitch Joint
The shoulder pitch angle q3 is also a challenge to extract. The basic approach is to
find two independent linear equations in S3 and C3. These can be obtained from the
expressions for b now that the angles q i9 q2, and q4 are known. First, note from Eqs.
(3-6-10) and (3-6-11) that S1&1 — C ib2 — S3a3 + S34a4- If we use the sine of the
sum trigonometric identity for S34 and rearrange the coefficients, this yields:
Sibi — C\b 2 — (ú3 + C4a4)S3 + (S4a4)C3 (3-6-15)
To produce a second equation involving the unknowns S3 and C3, we note ffom
Eq. (3-6-12) that &3/S 2 = C3«3 + If we use the cosine of the sum trigono
metric identity for C m and rearrange the coefficients, the result is:
= ( - S 4a4)S3 + (flj + C4a4)C3 (3-6-16)

S'2
Note that the expression in Eq. (3-6-16) involves división by S2. This is possi-
ble because S2 # 0 in view of Eq. (3-6-5). We can now solve Eqs. (3-6-15) and (3-
6-16) simultaneously using row operations, and the result is:
Sec. 3-6 Inverse Kinematics of a 6-Axis Articulated Robot (Intelledex 660) 103
S 4 íu (S j6 i — C i bz) (ay + C U a J b s / S i
c, = (3-6-17)
(S^ ) 2 + (¿23 + C4ÚÍ4)2
(<23 + C4úu)(Si b\ — C i¿^2) — S4cuby/S2
S3 (3-6-18]
(S4 CÍ4)2 + (a? + C404)2
Given the expressions for C3 and S3, the atan2 function can now be used to
solve for the shoulder pitch angle over the entire range [ —tt. tt]. This yields:
S4Ú4b}
#3 = atan2 (ay + C4a4)(Si/?i - C i62) —
S2
(a3 + C4a4)^3
S4fl4(S,¿>» “ C lW + (3-6-19)
S¡
3-6-6 Tool Pitch Joint
The only joint variable that remains to be isolated is the tool pitch angle #5. We de
termine #5 indirectly by first computing the sum angle #345 using the R matrix. Ob
serve from Eq. (3-6-2) that Ryy = S2S345. Because S2 is known and nonzero, this
means that #345 is known over the interval [ —tt/2 , tt/2 ] . To extend the solution to
the entire interval [ —tt, 7r], we must also isolate C345. This can be achieved by using
the remaining components in the third row of R. Note from Eq. (3-6-2) that
/?3iCó — JR32S6 = S2C345. It then follows that the sum angle #345 is:
#345 — atan2 (Ryy, RyiQe ~~ RyiSe) (3 -6 -2 0 )
The work for extracting the final joint variable is now in place because qy, #4,
and #345 are all known. Recalling that #345 = #3 + #4 + #5, it follows that:
#5 — q345 “ qy~ #4 (3 -6 -2 1 )
The complete inverse kinematics solution for the six-axis Intelledex 660 robot is
summarized in Algorithm 3-6-1. Recall that this solution is valid only for the left-
handed configuration with —tt < #2 < 0. The right-handed solution can be
obtained by simply changing the sign of the árceos term in the expression for #2.
However, in either case, Algorithm 3-6-1 is not valid when the shoulder roll angle
#2 is a múltiple of tt, because it is assumed that S2 ^ 0. This assumption is explicit
in the expression for the shoulder pitch angle qy and implicit in the expressions for q\
and #345.
Exercise 3-6-1: Home Position. For the six-axis Intelledex 660 robot, use
Algorithm 3-6-1 to find q when the robot is in the following position, called the
home position. Does your answer agree with the last column of Table 2-6?
w = [ay + a 4 + de, 0 , d \ , 1, 0 , 0 ] r
0 0 f
/? = 0 -1 0
1 0 0

Algorithm 3-6-1: Inverse Kinematics of a Six-Axis Articulated
Robot (Intelledex 660)
3-7 INVERSE KINEMATICS

OF A THREE-AXIS PLANAR ARTICULATED ROBOT
As a final example of computing the inverse kinematics of a robot, consider the

three-axis planar articulated robot shown in Fig. 3-10. For a robot which operates in
a plañe, there are only three degrees of freedom for the tool: two translational mo
tions within the plañe, and one rotational motion about an axis orthogonal to the
plañe.
To solve the inverse kinematic equations of the three-axis planar manipulator,
we must first derive the direct kinematic equations. The application of steps 0 to 7
Figure 3-10 A three-axis planar articu

lated robot.
Sec. 3-7 Inverse Kinematics of a 3-Axis Planar Articulated Robot 105

of the D-H algorithm to the three-axis planar articulated robot yields the link-
coordinate diagram shown in Fig. 3-11.
y3i
\
Tool

three-axis planar articulated robot.
Next, the application of steps 8 to 13 of the D-H algorithm results in the kine
matic parameters shown in Table 3-2.

OF A THREE-AXIS PLANAR ARTICULATED
ROBOT
A xis 0 d a a Home
1 <l\ 0 a\ 0 7r/3
2 <72 0 a2 0 —7f/3
3 <73 d* 0 0 0
Since the robot is an articulated robot, the vector of joint variables is q = 0.

Using Table 3-2 and Prop. 2-6-1, the arm matrix for the three-axis articulated robot
can be computed as follows:
n n Ti
c, -s, 0 «,C,: c 2 - s 0 02 C2 c3 -S i 0 0
s, C, 0 tfiSi s 2 c 2 0 02S2 s3 Cl 0 0
0 0 1 0 0 0 1 0 0 0 1 di
0 0 0 1 .0 0 0 1 .0 0 0 1
Cn - s 12 0 0,C1+ Cl2C\2 c3 -s 3 0 0'
Sl2 Cj2 0 (3iSi + a 2 S12 s3 c 3 0 0
0 0 1 0 0 0 1 di
0 0 0 1 0 0 0 1
Cl23 Si23 0 ajCi + CliCn
S|23 C,23 0 OiSi + C12 S 12 (3-7-1)
0 0 1
0 0 0 1

Exercise 3-7-1: Planar «-Axis Manipulator. Use the pattern in 7o, 7o, and
7o to write down, by inspection, the arm matrix of an «-axis planar articulated
robot, where n is arbitrary.
At this point we combine the tool-tip position and tool orientation into the
more compact tool-configuration vector w. Although the tool-configuration space
could be regarded as three-dimensional in this case, to facilítate comparisons with
other robots we use the general case w — (vv1, vv2), where w 1 = p and
w2 = [exp ( q i/ir ^ r 3. From inspection of the arm matrix in Eq. (3-7-1), we see that
the tool-configuration function of the three-axis planar articulated robot is:
üi C\ 4 ¿Z2C12
tfiSi 4 ¿22S12
di
w(<?) = (3-7-2)
0
0
exp f a / n )
The planar nature of the three-axis robot is evident ffom Eq. (3-7-2), since
W<?), W4(<?), and w5(q) are all constant. If the tool length d$ is zero, then the tool tip
moves in the jc°y° plañe. More generally, the tool tip p moves in a plañe parallel to
the x°y° plañe at a distance dj above it.
In order to extract the joint angle vector q from the nonlinear tool-
configuration function w(q), we again apply row operations to the components of w
and exploit trigonometric identities from Appendix 1 in order to isolate the individ
ual joint variables.
3-7-1 Shoulder Joint
First, note from Fig. 3-11 that the radial distance between the base frame Lo and
the wrist frame L2 depends only on the shoulder angle < 72. In particular, from Eq.
(3-7-2) we have:
vv? -f W2 = (fliCi + Ü2C12)2 + (tfiSi 4 - ¿Z2S12)2
= a \ C \ + l a xa 2C i C n + 02 CÍ2 + u?S2 + 2 a i a 2 S \ S n 4- a l S h
= a j + 2¿zi¿^(CnCi S12S1) 4 * Ü2
— ai + 2¿z]Q2 C2 + a\ (3-7-3)
We can now solve Eq. (3-7-3) for the shoulder angle. Isolating C2 and applying the
árceos function to both sides yields the follow ing two Solutions:
w2, + wl - a 2, - a 2 ^
q2 = ± á rceo s --------------- (3-7-4)
2ü i ü 2
As was the case with the SCARA robot, the first solution is called the left-handed
solution and the second is the right-handed solution.
Sec. 3-7 Inverse Kinematics of a 3-Axis Planar Articulated Robot 107

3-7-2 Base Joint
Given the shoulder angle q 2 , the base angle can now be determined from Eq.
(3-7-2). If the expressions for C 12 and S12 are expanded using the cosine of the sum
and sine of the sum trigonometric identities and the coefficients of Ci and Si are col-
lected, this yields:
(a\ + ¿z2C2)Ci — (a2S2)Si = w\ (3-7-5a)
(a2S2)Ci + (a\ + a2C2)Si = w2 (3-7-5b)
Thus we have two linear equations in the unknowns Ci and Si. These equations can
be solved simultaneously using elementary row operations, and the result is:
(a\ +-r tf2C2)wi
u2K^2jyv\ ~r
+ au2a2w2
2 S 2 w2 ,
Ci = -----;-----7 ^ ,■ ,—7 - 7 (3-7-6a)
(ai + a 2 C2) 2 + (a2 S2):
_ (fifi + a2 C2)w2 - a2 S 2 w\ t .
Si — 7 n \2 * / c~vT (3-7-6b)
(«i -I- a 2 C2)¿ + (a 2s 2y
Note that if a 2 — a\, then the numerators and the denominators of the expressions
for Ciand Sigo to zero when q2 = tt. This corresponds to a workspace singularity
with thetool roll axiscollinear with the base axis. If we avoid this singularity, then
both Ci and Si are known and the base angle q\ can be recovered over the complete
range [ — t t , t t ] using:
q\ = atan2 [(ai + a2 C2)w2 — a2 S 2 w\, (a\ + a 2 C2)w 1 + íz2S2w2] (3-7-7)
From Eq. (3-7-1), the tool roll angle q3 can be obtained directly from the last com
ponent of the tool-configuration vector as follows:
q3 = tt ln vv6 (3-7-8)

I
The complete inverse kinematics solution for the three-axis planar articulated robot
in Fig. 3-11 is summarized in Algorithm 3-7-1. Recall that the two expressions for
q 2 specified by the plus and minus signs represent the left-handed and right-handed
Solutions, respectively.
assumes that the tool-configuration vector vv is supplied as input data. If instead the
pair {/?, i?} is specified, then w = [ p x, p u p 3, R i3, R23, R33]T can be used to com
pute qx and q 2 using Algorithm 3-7-1. From Eq. (3-7-2), the tool roll angle can then
be computed from the global tool roll angle <7123, as follows:
q 3 = atan2 (Ru , R 2 1) - q\ ~ qi (3-7-9)

Algorithm 3-7-1: Inverse Kinematics of a Three-Axis Planar
Articulated Robot
q2 = ±árceos --------- --------

2d^3^
I
q3 = ir In w6
Exercise 3-7-2: Right Angle. For the three-axis planar articulated robot in
Fig. 3-11, find the joint variables q when the first two links form a right angle, as
follows:
w = [a2, ai, <¿3, 0, 0, l]r
3-8 A ROBOTIC WORK CELL
The technique for transforming between tool coordinates and base coordinates using
homogeneous coordinate transformation matrices can also be applied more generally
to coordinate transformations between various stations in a robotic work cell. As an
illustration, consider the single-robot work cell shown in Fig. 3-12. Here a part in
the shape of a block has been extracted, say, from a feeder, and placed in the view-
ing area of an overhead camera. The visión system inspeets the part and then in-
forms the robot controller to pick up the part and place it in either a pass bin or a
reject bin, depending upon the outeome of the inspection process.
Base xb
Figure 3-12 A single-robot work cell.
Sec. 3-8 A Robotic Work Cell

In order to perform a kinematic analysis of this robotic work cell, let TISU»
represent the position and orientation of the part as seen by the camera. For ex
ampie:
0 “ 1 0 0
y p art _
-1 0 0 -5
* cam era (3-8-1)
0 0 -1 19
0 0 0 1
Next let TcSera represent the position and orientation of the base of the robot as
seen by the camera. In this case suppose that:
0 -1 0 15
base __
1 0 0 25
7camera (3-8-2)
0 0 -1 20
0 0 0 1
Consider the problem of finding the position and orientation of the part rela
tive to the base of the robot. Using homogeneous coordinate transformation ma
trices, recalling that diagonal identifiers cancel, and using Prop. 2-4-1, we have:
n art 'T ’c a m e r a 'T ’p a r t
ase * base * camera
— ( TTbase \ —1 'T’part
V-* camera/ •» camera
i
0 -1 0 15' 0 -1 0 0
-1 0 0 25 -1 0 0 -5
0 0 -1 20 0 0 -1 19
0 0 0 1 0 0 0 1
0 -1 0 25' 0 -1 0 o'
-1 0 0 15 1 0 0 -5
0 0 -1 20 0 0 -1 19
0 0 0 1 0 0 0 1
1 o o 30
0 1 o 15 (3-8-3)
o o 1 1
F
0 0 0 i 7
Thus the position of the part relative to the base ffame of the robot is
p = [30, 15, l]7. Furthermore, the three axes of the part frame point in the same
directions as the corresponding axes of the base ffame, as is indicated by the fact
that the rotation submatrix is R — /. This is consistent with the picture in Fig. 3-12.
Next suppose we want to have the robot reach down and pick up the part di
rectly ffom above by grasping it on the front and back faces. To determine the re-

quired rotation matrix R - [x\ y \ z'], first note from Fig. 3-12 that to pick up the
part from above we need an approach vector which points down. In terms of base
coordinates, this corresponds to r 3 = —i3. Next, to grip the object on the front and
back faces, we need a sliding vector of r 2 = ± i2. Finally, to complete the right-
handed orthonormal coordinate frame, it is necessary that r l = T / 1. Thus there are
two possible Solutions for the rotation matrix R, one being:
”1 0 0“
R = 0 -1 0 (3-8-4)
0 0 -1
We can now put the position and orientation information together to develop
the tool configuration needed to grasp the part. In particular, the arm matrix which
places the tool in the position p and orientation R to pick up the part is:
1 0 0 30
ynool 0 -1 0 15
base (<?) = (3-8-5)
0 0 -1 1
---------
0 0 0 1
Alternadvely, we can specify the more compact tool-configuration vector w{q)
needed to pick up the part. Here the required tool roll angle qn must be determined.
Suppose the robot in Fig. 3-12 is a Rhino XR-3. Then, from Eq. (3-4-2), we have
q\ — atan2 (10, 25) = 21.8 degrees, and from Eqs. (3-8-4) and (3-4-20):
q$ = atan2 (Si Ru — Q1 R 2 1 , S1R12 ~ C1R22)
= atan2 (Si, Ci)
= 21.8 degrees (3-8-6)
This is consistent with Fig. 3-7, where the robot is pictured in the soft home posi
tion, which corresponds to q — [0, —t t / 2 , t t / 2 , 0, — t t / 2 ] t . From Def. 3-3-1, the
tool-configuration vector w(q) consists of the position vector p augmented by the
approach vector r 3 scaled by exp (^s/tt). Thus, in this case, the tool configuration
needed to grasp the block from above on the front and back faces is:
w = [30, 15, 1, 0, 0, -1.129]r (3-8-7)
To actually command the robot to reach down and pick up the part, we now
need to know the valué of the joint variables q that will produce the configuration of
the tool specified in Eq. (3-8-7). At this point, Algorithm 3-4-1 would be used to
determine the vector of joint variables q for the five-axis articulated robot.
Exercise 3-8-1: Focus Camera. Find the arm matrix needed to have the
robot in Fig. 3-12 reach up and focus the camera by twisting the lens using tool roll
motion.
Sec. 3-8 A Robotic Work Ceíl 111

3-9 PROBLEMS
3-1. Suppose a six-axis articulated robot has the following limits on the range of travel of its
joint variables:
— 37r/4 IJ qi < 3 ir /4 —tt/2 < ^4 < 7r/2
-tt/4 < < 37r/4 —ir/2 ^ #2 + #3 + <75 ^ ?r/2
—ir /2 ^ q2 + qi ^ tt / 2 — tt ^ q 6 < tt
Find the joint limit vectors {^min, ^ m*x} and the joint coupling matrix C that character
ize the joint-space work envelope Q of this robot.
3-2. Suppose the tool of a five-axis articulated robot is put into a singular configuration in
which the tool roll axis is collinear with the base axis as shown in Fig. 3 -1 3 . Show
that there are an infinite number of Solutions to the inverse kinematics problem for this
particular tool configuration. Is this a kinematically redundant robot?
Figure 3-13 A singular tool

configuration.
3-3. Consider the link-coordinate diagram for the four-axis cylindrical-coordinate robot
with tool roll motion shown in Fig. 3-14. Here the vector of joint variables is
q = [0i, ¿2, ¿3, 04]r , and the kinematic parameters are as specified in Table 3-3. Find
the arm matrix 7b£l(</) for this cylindrical-coordinate robot.
3-4. Find the tool-configuration vector w{q) for the cylindrical-coordinate robot in Fig.
3-14.
3-5. Solve the inverse kinematics equations for the cylindrical-coordinate robot in Fig.
3-14. Do parts (a) through (d) in whichever order is easiest.
(a) Find the base angle #i(w).
(b) Find the vertical extensión qi(w).
(c) Find the radial extensión q^w).
(d) Find the tool roll angle #4(w).
3-6. Find a reduced tool-configuration vector vv for the cylindrical-coordinate robot in Fig.
3-14.
3-7. For the cylindrical-coordinate robot in Fig. 3-14, use the rotation matrix R{q) to find
the tool roll angle <?4.

four-axis cylindrical robot.
TABLE 3 -3 KINEMATIC PARAMETERS
OF THE CYLINDRICAL ROBOT
Axis 0 d a a Home
1 0 0 0 0
2 tt / 2 <?2 0 7r/2 d
3 0 43 0 tt/2 r
4 ¿4 0 0 TT
3-8. Consider the link-coordinate diagram for the five-axis spherical-coordinate robot with
tool pitch and tool roll motion shown in Fig. 3-15. Here the vector of joint variables is
q — [0i, 02, d3, 04, 05]r , and the kinematic parameters are as specified in Table 3-4.
Find the arm matrix 7ba°e(#) for this spherical-coordinate robot.
3-9. Find the tool-configuration vector w(q) for the spherical-coordinate robot in Fig. 3-15.
3-10. Solve the inverse kinematics equations for the spherical-coordinate robot in Fig. 3-15.
Do parts (a) through (e) in whichever order is easiest.
(a) Find the base angle qi(w).
(b) Find the shoulder angle qiiyv).
(c) Find the radial extensión q3{w).
(d) Find the tool pitch angle qi(w).
(e) Find the tool roll angle q¡M-
3-11. Find a reduced tool-configuration vector w for the spherical-coordinate robot in Fig.
3-15.
3-12. For the spherical-coordinate robot in Fig. 3-15, use the rotation matrix R(q) to find the
tool roll angle q5.
3-13. Recall from Chap. 2 that the composite yaw-pitch-roll transformation can be repre
sented by the following rotation matrix YPR. Solve for the yaw, pitch, and roll angles
0 G R3 in terms of the components of YPR.
Sec. 3-9 Problems 113

C2C3 S .S jC j - C ,S 3 C ,S 2C3 + S ,S 3'
YPR (0) = C2S3 S,S2S3 + C ,C j C ,S 2S3 - s ,c 3

—S2 SIC2 C1C2
Figure 3-15 Link coordinates of a five-axis spherical robot.
OF THE SPHERICAL ROBOT
Axis e d a a Home
1 <7i dx 0 —
77/2 0
2 <?2 0 0 7t/2 77/2
3 77 ?3 0 77/2 r
4 <74 0 0 — 77/2 — 77/2
5 ?5 ds 0 0 tt/ 2
3-14. Suppose the joint-space work envelope Q for the four-axis cylindrical robot is of the
following form. Show that there are two Solutions to the inverse kinematics problem in
this case by completing Table 3-5. The second solution should be expressed in terms of
the first solution. For example, a joint variable might be negated, or 7r might be added
to a joint variable.
—T T < q x <TT —H < <73 < H
0 < q2 < V —7r < <y4 < 7r
TABLE 3-5 MULTIPLE SOLUTIONS
OF THE FOUR-AXIS CYLINDRICAL ROBOT
Vertical Radial
Solution Base Extensión Extensión Tool Roll
1 <¡2 <?3 44
2

3-15. Suppose the joint-space work envelope Q for the five-axis spherical robot is of the fol
lowing form. Show that there are two Solutions to the inverse kinematics problem in
this case by completing Table 3-6. To simplify the problem somewhat, you can assume
that q2 = 0 corresponds to the arm pointing straight up.
— TT < q x < 7T
TT ^ — TT
0 < q3 < H
— TT < < TT
— TT < qs < 7T
TABLE 3-6 MULTIPLE SO LUTIO NS

OF THE FIVE-AXIS SPHERICAL ROBOT
Radial Tool Tool

Solution Base Shoulder Extensión Pitch Roll
1 <72 <73 <?4 <75
2
REFERENCES
P ie p e r , D. L. (1968). “The kinematics of manipulators under Computer control,” AIM 72,

Stanford AI Laboratory, Stanford Univ.
U i c k e r , J. J., J. D e n a v it , and R. S. H a k t en b er g (1964). “An iterative method for the dis
placement analysis of spatial mechanisms,” Trans. ASME J. Appl. Mech., Vol. 31, Ser. E,
pp. 309-314.
W h it n e y , D. E. (1972). “The mathematics of coordinated control of prosthetic arms and ma
nipulators,” J. Dynamic Systems, Measurement and Control, Vol. 122, pp. 303-309.
115
Sec. 3-9 Problems

Robotica 1 3 Fotos2 PDF

Uploaded by

Copyright:

Available Formats

Robotica 1 3 Fotos2 PDF

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

Robotica 1 3 Fotos2 PDF

Uploaded by

Copyright:

Available Formats

FUNDAMENTAIS

1-1 AUTOMATION AND ROBOTS

1-1 AUTOMATION AND ROBOTS

Figure 1-1 An industrial robot. (Courtesy of Intelledex, Inc., Corvallis, OR.)

tasks, they are referred to as examples of soft automation. A qualitative comparison

2 Robotic Manipulation Chap. 1

Figure 1-1 An industrial robot. (Courtesy of Intelledex, Inc., Corvallis, OR.)

tasks, they are referred to as examples of soft automation. A qualitative comparison

2 Robotic M anipularon Chap. 1

Figure 1-2 Relative cost-eífectiveness

Definition 1-1-1: Robot. A robot is a software-controllable mechanical de-

1-2 ROBOT CLASSIFICATION

In order to refine the general notion of a robotic manipulator, it is helpful to classify

1-2-1 Drive Technologies

See. 1-2 Robot Classification 3

1-2-2 Work-Envelope Geometries

The end-effector, or tool, of a robotic manipulator is typically mounted on a flange

TABLE 1-1 TYPES OF ROBOT JOINTS

Type Notation Symbol Description

Revolute R Rotary motion about an axis

Prismatic P Linear motion along an axis

4 Robotic Manipulation Chap. 1

Robot Axis 1 Axis 2 Axis 3 Total revolute

Figure 1-3 Cartesian robot.

If the first joint of a Cartesian-coordinate robot is replaced with a revolute

Figure 1-4 Cylindrical robot.

Sec. 1-2 Robot Classification 5

Figure 1-5 Spherical robot.

Like a spherical-coordinate robot, a SCARA robot (Selective Compliance As-

7 Figure 1-6 SCARA robot

6 Robotic Manipulation Chap ‘

TABLE 1-3 TYPES OF ROBOT MOTION CONTROL

Control method Applications

Material handiing 24.4

Traditional applications of material handiing, welding, and spray painting and

8 Robotic Manipulation Chap. 1

1-4 ROBOT SPECIFICATIONS

1-4-1 Number of Axes

Axes TVpe Function

1 -3 Major Position the wrist

Sec. 1-4 Robot Specifications 9

1-4-2 Capacity and Speed

1-4-3 Reach and Stroke

10 Robotic Manipulation Chap. I

1-4-4 Tool Orientation

Figure 1-9 Yaw, pitch, and roll of

Sec. 1-4 Robot Specifications 11

TABLE 1-8 YAW, PITCH, AND ROLL MOTION

Operation Description Axis

12 Robotic Manipulation Chap. 1

1-4-5 Repeatability, Precisión, and Accuracy

Repeatability is another important design characteristic. Repeatability is a measure

TABLE 1-9 HORIZONTAL AND VERTICAL PRECISION

Robot Horizontal Vertical

Cartesian Uniform Uniform

Sec. 1-4 Robot Specifications 13

Figure 1-11 Horizontal precisión of a

14 Robotic Manipulation Chap. 1

Figure 1-12 Slotted disk with 2-bit

Figure 2-1 Robotic manipulator mod

The com plete orthonorm al set X is sometimes called an orthonorm al coordi

Figure 2-4 Coordinates of p with re Figure 2-5 Single-axis robot.