NAVAL POSTGRADUATE SCHOOL Monterey, California

THESIS
OBSTACLE AVOIDANCE CONTROL FOR THE REMUS AUTONOMOUS UNDERWATER VEHICLE by Lynn Renee Fodrea December 2002

Thesis Advisor:

Anthony Healey

Approved for public release; distribution is unlimited

THIS PAGE INTENTIONALLY LEFT BLANK

REPORT DOCUMENTATION PAGE

Form Approved OMB No. 0704-0188

Public reporting burden for this collection of information is estimated to average 1 hour per response, including the time for reviewing instruction, searching existing data sources, gathering and maintaining the data needed, and completing and reviewing the collection of information. Send comments regarding this burden estimate or any other aspect of this collection of information, including suggestions for reducing this burden, to Washington headquarters Services, Directorate for Information Operations and Reports, 1215 Jefferson Davis Highway, Suite 1204, Arlington, VA 22202-4302, and to the Office of Management and Budget, Paperwork Reduction Project (0704-0188) Washington DC 20503. 1. AGENCY USE ONLY (Leave 2. REPORT DATE 3. REPORT TYPE AND DATES COVERED blank) December 2002 Masters Thesis 4. TITLE AND SUBTITLE Obstacle Avoidance Control for the 5. FUNDING NUMBERS REMUS Autonomous Underwater Vehicle N0001401AF00002 6. AUTHOR (S) Lynn Fodrea 8. PERFORMING ORGANIZATION 7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) REPORT NUMBER Naval Postgraduate School Monterey, CA 93943-5000 9. SPONSORING / MONITORING AGENCY NAME(S) AND ADDRESS(ES) 10. SPONSORING/MONITORING AGENCY REPORT NUMBER Office of Naval Research

800 North Quincy Street Arlington, VA 22217-5660

11. SUPPLEMENTARY NOTES The views expressed in this thesis are those of the author and do not reflect the official policy or position of the U.S. Department of Defense or the U.S. Government. 12a. DISTRIBUTION / AVAILABILITY STATEMENT Approved for public release; Distribution Unlimited. 13. ABSTRACT (maximum 200 words) 12b. DISTRIBUTION CODE

Future Naval operations necessitate the incorporation of autonomous underwater vehicles into a collaborative network. In future complex missions, a forward look capability will be required to map and avoid obstacles such as sunken ships. This thesis examines obstacle avoidance behaviors using a forwardlooking sonar for the autonomous underwater vehicle REMUS. Hydrodynamic coefficients are used to develop steering equations that model REMUS through a track of specified points similar to a real-world mission track. Control of REMUS is accomplished using line of sight and state feedback controllers. A two-dimensional forward-looking sonar model with a 120 horizontal scan and a 110 meter radial range is modeled for obstacle detection. Sonar mappings from geographic range-bearing coordinates are developed for implementation in MATLAB simulations. The product of bearing and range weighting functions form the gain factor for a dynamic obstacle avoidance behavior. The overall vehicle heading error incorporates this obstacle avoidance term to develop a path around detected objects. REMUS is a highly responsive vehicle in the model and is capable of avoiding multiple objects in proximity along its track path.
14. SUBJECT TERMS Obstacle avoidance, REMUS, Underwater vehicle, AUV 15. NUMBER OF PAGES 79 16. PRICE CODE 20. LIMITATION OF ABSTRACT UL

17. SECURITY CLASSIFICATION OF REPORT Unclassified

NSN 7540-01-280-5500

18. SECURITY CLASSIFICATION OF THIS PAGE Unclassified

19. SECURITY CLASSIFICATION OF ABSTRACT Unclassified

Standard Form 298 (Rev. 2-89) Prescribed by ANSI Std. 239-18

THIS PAGE INTENTIONALLY LEFT BLANK

ii

Approved for public release; distribution is unlimited

OBSTACLE AVOIDANCE CONTROL FOR THE REMUS AUTONOMOUS UNDERWATER VEHICLE

Lynn Fodrea Lieutenant, United States Navy B.S., U.S. Naval Academy, 1998

Submitted in partial fulfillment of the requirements for the degree of

MASTER OF SCIENCE IN MECHANICAL ENGINEERING from the NAVAL POSTGRADUATE SCHOOL December 2002

Author:

Lynn Fodrea

Approved by:

Anthony J. Healey Thesis Advisor

Young W. Kwon Chairman, Department of Mechanical Engineering

iii

THIS PAGE INTENTIONALLY LEFT BLANK

iv

ABSTRACT

Future Naval operations necessitate the incorporation of autonomous underwater vehicles into a collaborative network. In future complex missions, a forward look

capability will be required to map and avoid obstacles such as sunken ships. This thesis examines obstacle avoidance behaviors using a forward-looking sonar for the autonomous underwater vehicle REMUS. Hydrodynamic coefficients are

used to develop steering equations that model REMUS through a track of specified points similar to a real-world mission track. Control of REMUS is accomplished using line of A two-dimensional

sight and state feedback controllers.

forward-looking sonar model with a 120 horizontal scan and a 110 meter radial range is modeled for obstacle detection. Sonar mappings from geographic range-bearing coordinates are developed for implementation in MATLAB simulations. The product of bearing and range weighting functions form the gain factor for a dynamic obstacle avoidance behavior. The overall vehicle heading error incorporates this obstacle avoidance term to develop a path around detected objects. REMUS is a highly responsive vehicle in the model

and is capable of avoiding multiple objects in proximity along its track path.

THIS PAGE INTENTIONALLY LEFT BLANK

vi

TABLE OF CONTENTS

I.

INTRODUCTION ............................................1 A. BACKGROUND .........................................1 B. MOTIVATION .........................................2 C. OBSTACLE AVOIDANCE FOR AUTONOMOUS UNDERWATER VEHICLES ...........................................3 D. PATH PLANNING ......................................4 E. SCOPE OF THIS THESIS THE REMUS VEHICLE ...........7 F. THESIS STRUCTURE ...................................8 STEERING MODEL .........................................11 A. GENERAL ...........................................11 B. EQUATIONS OF MOTION IN THE HORIZONTAL PLANE .......11 C. HYDRODYNAMIC COEFFICIENTS .........................16 D. VEHICLE KINEMATICS ................................20 E. VEHICLE DYNAMICS ..................................20

II.

III. CONTROL METHODS AND ARCHITECTURE .......................21 A. GENERAL CONTROL THEORY ............................21 B. REMUS CONTROL ARCHITECTURE ........................23 C. SLIDING MODE CONTROL ..............................23 D. LINE OF SIGHT GUIDANCE ............................25 IV. OBSTACLE AVOIDANCE MODEL ...............................29 A. THE REMUS SEARCH PATH .............................29 B. SONAR MODEL .......................................30 C. HEURISTICS ........................................31 VEHICLE SIMULATION .....................................35 A. BASIC SINGLE POINT OBSTACLE AVOIDANCE .............35 B. MULTIPLE POINT OBSTACLE AVOIDANCE .................39

V.

VI. CONCLUSIONS AND RECOMMENDATIONS .........................45 A. CONCLUSIONS .......................................45 B. RECOMMENDATIONS ...................................46 APPENDIX A ..................................................49 LIST OF REFERENCES ..........................................59 INITIAL DISTRIBUTION LIST ...................................63

vii

THIS PAGE INTENTIONALLY LEFT BLANK

viii

LIST OF FIGURES

Figure 1. REMUS VEHICLE ......................................7 Figure 2. Local and Global Coordinate System (From: Marco and Healey, 2001) .................................12 Figure 3. Track Geometry and Velocity Vector Diagram ........26 Figure 4. Typical REMUS Search Path .........................29 Figure 5. Forward-look Sonar Model ..........................30 Figure 6. Bearing Weighting Function ........................32 Figure 7. Range Weighting Function ..........................33 Figure 8. Block Diagram System Dynamics .....................34 Figure 9. Single Point Obstacle Run (On Path) ...............35 Figure 10. Single Point Obstacle Run: Rudder/Heading/oa.....36 Figure 11. Single Point Obstacle Run (Off Path) .............36 Figure 12. Rudder/Heading/oa (Off Path).....................37 Figure 13. Figure-Eight Obstacle Run ........................38 Figure 14. Rudder/Heading/oa Figure-Eight...................38 Figure 15. Multiple Single Point Obstacle Run ...............40 Figure 16. Rudder/Heading/oa................................41 Figure 17. Multiple Point Obstacle Run ......................41 Figure 18. Multiple Point Obstacle Run: Rudder/Heading/oa...42 Figure 20. Alternate Range Weighting Function ...............43 Figure 21. Range Weighting Function Dynamics Comparison .....44

ix

THIS PAGE INTENTIONALLY LEFT BLANK

LIST OF TABLES

Table 1. Table 2.

REMUS Functional and Physical Characteristics ......8 REMUS Hydrodynamic Coefficients for Steering ......19

xi

THIS PAGE INTENTIONALLY LEFT BLANK

xii

ACKNOWLEDGEMENTS

I would like to thank my thesis advisor, Professor Anthony J. Healey, for his expert insight, direction, and assistance during the development of this work. His

ability to see beyond my thoughts and assumptions allowed this product to become what it is. I would also like to thank CDR Bill Marr, an AUV team member, friend, and counselor who has continued to be there to listen to and advise me on career, academic, and social concerns. Finally, I would like to thank my fianc, Major Todd Woodrick, USAF, for listening to me talk through my

thoughts, for being there to challenge me, and for being my truest fan and support over the last year and a half.

xiii

THIS PAGE INTENTIONALLY LEFT BLANK

xiv

I.
A. BACKGROUND United States

INTRODUCTION

naval

warfare

strategy

is

constantly One of

evolving and adapting to our ever-changing world.

the most foreign and complex areas of naval warfare that requires a myriad of resources to explore and classify is that of the underwater world. Operations in the littoral With increased Amphibious and an increased

environment

need for Force Protection of our nations ports, it is critical to be able to characterize the undersea

battlefield and an enemys coastal defenses.

Recently, the

undersea battlefield has undergone considerable change with the advent of improved mines, submarine quieting, and other littoral threats. It has often been said that the best way to combat threats in a specific environment is to use assets in the same medium. A major area of development for combating

this complex undersea battlefield from the surf zone to the shallow (UUV). forces water UUVs by regime not is the Unmanned Underwater to from our the Vehicle military hostile a more

only the

increase human but

safety

removing

swimmer also

minefield

environment,

they

provide

maneuverable asset in the random and turbulent waters of the littorals. The UUV Mission Priorities, as outlined in

the Organic Off-board Mine Reconnaissance CONOPS, include programs undersea that will extend knowledge the and control of of the

battle

space

through

employment

covert

sensors capable of operating reliably in high-risk areas. The CONOPS states that there are four basic mission areas 1

for

which

the

utility mine

of

unmanned

undersea

systems

was

substantiated:

warfare,

surveillance,

intelligence

collection, and tactical oceanography. and reliability during these missions,

To ensure success it is imperative This

that the UUVs used are capable of obstacle avoidance.

thesis will focus on obstacle avoidance arguments for a specific type of UUV known as the Autonomous Underwater Vehicle (AUV). respect to AUVs are unmanned, independent craft with and control and require no external

power

interface. These well-known Vehicles.

AUVs appeal to the underwater community in that

they are able to: Provide their own power Provide data storage capabilities Make decisions sensors capabilities based alone on inputs them or from onboard their

set

apart Remotely

from

counterparts,

ROVs

Operated

ROVs are not only tethered, but require a human

interface as well as sufficient cable to search the waters around the base platform (Ruiz, 2001)

B.

MOTIVATION Advancements have been made in the area of robotics

for underwater environments over the past several years. AUV development began as far back as 1960 with experimental prototypes available in the 1980s. development, see (Blidberg, 2001). For a history on AUV AUVs possess the unique

ability to safely operate in littoral areas for search, detection, and classification of mines and for hydrographic reconnaissance and intelligence. 2 To broaden the

capabilities industrial vehicle robotic

of and

underwater environmental it is

vehicles

for in

military, multiple a robust of

applications to

operations, system

essential the

design

that

exhibits

maximum

degree

autonomy, both through navigation and sensory processing. One of the greatest technological challenges facing AUVs and the robot community today is that of navigation around obstacles. problem of While most underwater vehicles can solve the localization and maneuvering, many do not

possess the capability to move around obstacles that arise in their programmed path, specifically in unmapped areas near the littorals where mine-like objects or other

potential hazards are prevalent.

Land robots and crawling

vehicles are capable of obstacle and collision avoidance using a stop-back-turn principle that swimming vehicles cannot (Healey, Kim, 1999). This thesis will present a

solution to the obstacle avoidance problem for the Remote Environmental Measuring Unit System (REMUS) AUV.

C.

OBSTACLE AVOIDANCE FOR AUTONOMOUS UNDERWATER VEHICLES The obstacle avoidance problem has been under research

since the advent of underwater vehicle technology. approaches have been used to solve this

Several for

problem

underwater robots.

One approach is that of wall-following This method a close a

or obstacle contour following (Kamon, 1997). utilizes proximity the obstacle around boundaries the to

determine until

path

obstacle

reaching

position on the obstacle boundary where it can break away and return to its course. until the obstacle no The boundary following continues longer 3 blocks the desired path.

Experimental results using Kamons wall following algorithm show that this technique produces minimal path distances around obstacles. The approach proposed by Moitie and Suebe [Moite & Suebe, 2000] uses an obstacle avoidance system consisting of four subsystems: a digital terrain manager used to

estimate the sea floor altitude, a global planner used to generate waypoints to guide the AUV to a given target, a reflex planner to check the trajectories of the global

planner, and an obstacle avoidance sonar for environmental mapping. All of these subsystems are used to determine a

viable area of the state space from which a viable (or escape) trajectory can be used. The Vector Field Histogram (VHF) technique (Borenstien and Koren, 1991) consists of a two-stage data reduction process that uses a two-dimensional Cartesian histogram

grid as a world model.

The first stage is data reduction

to a onedimensional local polar histogram with each sector representing an obstacle density. The second stage

involves a selection of the sector with the lowest obstacle density. an The steering model is then reduced to calculating vector aligned with the selected

avoidance-heading

sector.

D.

PATH PLANNING Path planning is a tool used for devising collision

free trajectories for robot vehicles in a structured world where mission specifications and environmental models are known. execution Path for planning the commonly occurs prior to mission

existing 4

environmental

constraints.

Environmental

data

allows

path

planners

to

design

paths

around known physical obstacles such as trees and pillars or hazardous environments such as rough terrain or high turbulence areas. Path planning differs from obstacle

avoidance in that obstacle avoidance is performed in a nonstructured world that is initially assumed to be free of obstructions. of an underwater to However, due to the unpredictable nature environment, allow for path safe planning vehicle alone is

insufficient

navigation.

Obstacle avoidance is a necessary tool for in situ response to unknown environmental conditions and hazards. Several path planning techniques have been developed for both land based and subsurface robots. received the most attention in recent One that has years is the

potential field approach in which an artificial potential field is defined to reflect the structure of the space around the vehicle (Thrope, 1985, Krogh, 1986). A

repulsive field pushes the vehicle away form an indicated obstacle while an attractive field pulls a vehicle toward a goal. The path to the goal is minimized through the space.

It is configured to have a global minimum at the desire terminal state of the vehicle. The main drawback to this

approach lies in the fact that local minima may entrap the robot trajectory. A second approach considered by Latombe (1991) is that of cell decomposition in which the workspace is divided into non-overlapping cells represented by nodes. The space

is then searched from starting point to the end node using a graph search algorithm to determine the path of free cells. 5

Further

progress

has

been

made

to

incorporate

path

planning and obstacle avoidance in a more dynamic program. Stentz (1994) develops a path planning algorithm known as D* for partially known environments in which a sensor is also available to supplement a map of the environment. It

combines what is known of the global environment prior to mission missions. which a with acquired local environmental data during

The D* technique uses a cost based approach in directed graph arc of arcs an is generated prior to The

mission

with

each

having

associated

cost.

robots sensor can then measure arc costs in its local vicinity and generate known and estimated arc values that compromise a map. Lane (2001) uses an approach known as dynamic

programming.

This method considers a modular system that

handles different needs of the environment while the robot is in motion. module that These modules consist of a segmentation identifies a regions feature a of the sonar image a the

containing tracking

obstacles, module that

extraction dynamic

module, of

provides

model

obstacle, a workspace representation that builds a symbolic representation of the vehicles surroundings, and finally a path planning module that represents each obstacle as a constraint. The maneuvering solution is then based on

minimizing the path length to the goal. While described several above are of the path for planning land techniques robots vice

designed

underwater robots and involve much simpler dynamic motions, the challenge of underwater robot technology is in the

difficulty of ceasing or changing a forward motion given a short notice sonar return.

E.

SCOPE OF THIS THESIS THE REMUS VEHICLE The REMUS vehicle was developed at Woods Hole

Oceanographic Institute (WHOI) in the Oceanographic Systems Laboratory. It is designed to perform hydrographic

reconnaissance in the Very Shallow Water (VSW) zone from 40 to 100 feet. As seen in Figure 1, it is 62 inches long and It weighs 80 pounds in air and can

7.5 inches in diameter.

operate in depths up to 328 feet, but typically operates between 10 and 66 feet. The aft end propeller enables Its four

REMUS to reach a maximum speed is 5.6 knots.

fins, two horizontal and two vertical on either side and just forward of the propeller, allow pitch and yaw motions for maneuvering. Table 1 includes the remaining functional

and physical characteristics.

Figure 1. REMUS VEHICLE Currently, REMUS is equipped with a number of sensors that can generate water hydrographic clarity, maps, maps and of water

currents,

temperature,

salinity

profiles, as well as some acoustic profiles.

While REMUS

is fitted with two side-scan sonars that are used to detect objects on or near the sea floor, a forward-looking sonar

would give it the ability to detect objects in front of the vehicle.

Table 1.

REMUS Functional and Physical Characteristics

PHYSICAL/FUNCTIONAL AREA Vehicle Diameter Vehicle Length Weight in Air External Ballast Weight Operating Depth Range Transit Depth Limits Typical Search Area Typical Transponder Range Operational Temperature Range Speed Range Maximum Operating Water Current Maximum Operating Sea State Battery Endurance

CHARACTERISTIC 7.5 in 62 in 80 lbs 2.2 lbs 10 ft to 66 ft 328 ft 875 yds X 1093 yds 1640 yds +32F to +100F 0.5 knots to 5.6 knots 2 knots Sea State 2 1 kW-hr internally rechargeable Lithium-ion 20 hours at 3 knots; 9 hours at 5 knots

F.

THESIS STRUCTURE The intent of this research is to develop a forward-

looking

sonar

model

that

supports

obstacle

avoidance

behaviors on the REMUS vehicle. accomplished through the

This is a two step process firstly, develop a

following:

robust steering model for the REMUS vehicle as a necessary building block for obstacle avoidance behaviors; secondly, build obstacle avoidance control into the steering model to enable safe navigation of the very shallow water

environment while gathering or verifying environmental and minefield data. for the REMUS To this end, the steering model designed vehicle is 8 based on known hydrodynamic

coefficients heading

and

will for

incorporate both single

an

obstacle

avoidance threat

command

and

multiple

environments. Chapter II will focus on the development of the

equations of motion for the REMUS AUV.

Chapter III will

describe the steering control laws associated with the EOM for REMUS. algorithm simulation Chapter IV will discuss the obstacle avoidance developed analysis for for REMUS. the Chapter V will present behavior

obstacle

avoidance

discussed in the previous chapter and Chapter VI will offer conclusions and recommendations for future study.

10

II. STEERING MODEL

A. GENERAL Modeling vehicles terms of of rigid from body dynamics of to other produce for underwater only in The

differs the

modeling applied

robots

forces

motion.

approach taken with and underwater vehicle is that of a moving body in free space without constraint. The

propulsion and maneuvering forces on the moving body are hydrodynamic and hydrostatic in origin and are caused by interactions with the ocean water particles local to the body, rather than interactions with the ground as those of land based robots. These forces are often controllable and

can thus be studied from a perspective of stabilization. (Healey class notes)

B.

EQUATIONS OF MOTION IN THE HORIZONTAL PLANE The following paragraphs describe the development of

the steering model used to control the REMUS vehicle.

This

model was adapted from that of the ARIES AUV (Healey & Marco, 2001) and is based on the following assumptions: the vehicle behaves as a rigid body the earths rotation is purposed of acceleration vehicle center of mass negligible components for of the the

the primary forces that act on the vehicle are inertial and gravitational in origin and are derived from hydrostatic, propulsion, thruster, and hydrodynamic lift and drag forces

The equations of motion (EOM) for steering are derived using a Newton-Euler approach that relates the vehicles 11

position and motions in the local plane to those in the global plane. The geometry of the global and local

coordinate systems can be seen in Figure 3 below. X x

Ro y Z z

Figure 2. Local and Global Coordinate System (From: Marco and Healey, 2001)

Healey, (1995) shows that the local velocity vector

[ u , v , w]

where u is forward speed (surge), v is side slip and w is any component velocity in the local Z

(sway)

direction (heave), can be easily transformed to the global velocity vector & & & 1 through the Euler angles , , X ,Y , Z

and as follows:

" X u " v = T ( , , ) Y " w Z

Where T is the transformation matrix:

(1)

cos cos , cos sin sin sin cos , cos sin cos + sin sin (2) T ( , , ) = sin cos , sin sin sin + cos cos , sin sin cos cos sin cos sin cos cos sin ,
1

12

However, the connection between angular attitude and angular velocity is not as simple. Rate gyros in use today

measure the components of inertial angular velocity of a vehicle that lie along the vehicles body axes. Healey derives the inertial angular rates in Thus, of

terms

components that have angular velocities about the global axes and then transforms them as above to the final

reference frame.

The final transformation takes the form:

" 0 0 p q = T ( ) T ( ) T ( ) 0 + T ( ) T ( ) " + T ( ) 0 " 0 r 0

(3)

in which the rate components from each Euler angle are viewed as follows:

the change of rotation as a vector quantity lying along the original Z axis the rate of change of as a vector quantity lying along the Y axis of the first intermediate frame and the rate of change of as a vector lying along the X axis of the final body frame

with the result:

" -sin p 1 0 q = 0 cos sin cos " " r 0 -sin cos cos
For small angular rotations, it is evident that:

(4)

= p; = q; = r.

"

"

"

13

The final EOM are developed in the body fixed frame coordinates position, using velocity, these and inertial frame of quantities the of

acceleration

vehicles

center of mass. vector equation

The translational equation of motion is a relating the global acceleration of the

center of mass to the net sum of all of the forces acting on the vehicle in three degrees of freedom (X,Y,Z) as:

& + & g + g + v} F = m{v

(5)

The rotational equation of motion is derived from equating the sum of the applied moments about the vehicles center of mass to the rate of change of angular momentum of the vehicle about its center of mass. The mass moment of

inertia of the vehicle, I, about its center of gravity changes with loading. Thus, the mass moment of inertia is

evaluated about the body-fixed frame that lies along the vehicles axis of symmetry. The rotational equation of

motion in vector form thus becomes:

& + ( ) + m{ g v & + g v} M o = I o

(6)

With the addition of weight and buoyancy terms that act at the centers G and B, Healey, (1995) derives the equations of motion for a six degree of freedom model as: SURGE EQUATION OF MOTION
& r v r r + wr q x G (q 2 + r 2 )+ y G ( pq r & ) + z G ( pr + q & ) ] +(W B ) sin = X f m[ u

(7)

SWAY EQUATION OF MOTION

& r + u r r wr p + x G ( pq + r & ) y G ( p 2 + r 2 )+ z G (qr p & ) ] (W B ) cos sin = Y f m[ v

(8)

HEAVE EQUATION OF MOTION

& r u r q + v r p + x G ( pr q & ) + y G (qr + p & ) z G ( p 2 + q 2 ) ] +(W B ) cos cos = Z f (9) m[ w

14

ROLL EQUATION OF MOTION

& + I z I y qr + I xy ( pr q & ) I yz q 2 r 2 I xz ( pq + r & ) + m[ y G (w & u r q + vr p) Ix p & r + u r r wr p )] ( y G W y B B ) cos cos + (z G W z B B ) cos sin = K f z G (v

(10)

PITCH EQUATION OF MOTION

& + (I z I z ) pr I xy (qr + p & ) + I yz ( pq r & ) + I xz p 2 r 2 m[x G (w & u r q + vr p) I yq & r v r r + wr q )] + (x G W x B B ) cos cos + (z G W z B B ) sin = M f z G (u

(11)

YAW EQUATION OF MOTION

& + I y I x pq I xy p 2 q 2 I yz ( pr + q & ) + I xz (qr p & ) + m[x G (v & r + u r r wr p ) Izr & r v r r + wr q )] (x G W x B B ) cos sin ( y G W y B B ) sin = N f y G (u

(12)

Where: W = weight B = buoyancy I = mass moment of inertia terms ur, vr, wr = component velocities for a body fixed system with respect to the water p, q, r = component angular velocities for a body fixed system xB, yB, zB = position difference between geometric center and center of buoyancy xG, yG, zG = position difference between geometric center and center of gravity Xf, Yf, Zf, KF, Mf, Nf = sums of all external forces acting in the particular body fixed direction Healey (1995) further simplifies Equations 7 thru 12 with the following assumptions:

The center of mass of the vehicle lies below the origin (zG is positive) 15

xG and yG are zero The vehicle properties is symmetric in its inertial

The motions in the vertical are negligible (i.e. [wr, p, q, r, Z, , ] = 0) ur equals the forward speed, Uo.

The simplified equations of motion are thus:

ur = U o

(13) (14) (15) (16) (17) (18)

& r = mU o r + Y f (t ) mv
& = N f (t ) I zz r

& =r
& = U cos v sin + U X o r cx & = U sin v cos + U Y o r cy

C.

HYDRODYNAMIC COEFFICIENTS The modeling motions Under at of submerged vehicles in a assumes straight there small line is a

forward transit.

nominal

speeds

steady

motion

conditions,

balance between the hydrodynamic drag and propulsion forces as well as the weight and buoyancy forces. The predominant

forces from lift that arise in directions other than the longitudinal direction are caused from small angles of

attack and side slip. relative velocities

Hydrodynamic forces are related to accelerations of the fluid and

and

vehicle that result from any motions that deviate from the straight line path assumed above. Due to the symmetry of

vehicles about their longitudinal axis, the components of fluid motion in the transverse 16 direction are often

independent

of

motions

in

the

longitudinal

direction.

Healey proposes that due to the symmetry of the vehicle, one can heuristically would affect class determine the that only in uses a subset of

motions direction

loading and

any the

particular following

(Healey

notes)

expressions to describe the hydrodynamic forces of sway and yaw respectively:

Y f = f (vr , dvr / dt , r , dr / dt , p, dp / dt , t ) f = f ( p, dp / dt , vr , dvr / dt , r , dr / dt , t )

(19) (20)

It is evident that the sway and yaw motions are coupled in horizontal plane steering. Roll motion coupling is common The fluid series These

but is often one way and is thus not considered. forces above are in often linearized motion using

Tayor

expansion

terms

individual

components.

expansion terms are termed hydrodynamic coefficients and depend on the shape characteristics of the vehicle. Errors in these coefficients will have a significant affect on the natural stability of the vehicle as they are the building blocks of the dynamics matrix. Through the assumption of

small motions, the expression for the transverse (sway) force is:
& r + Yv r v r + Y r & + Yr r Y f = Yv &r v &r

(21)

and for the expression for the rotational (yaw) force is:
& r + N vr v r + N r & + Nrr N f = Nv &r v &r

(22)

This leads to:

Yv &r = Y f &r v

; Yv r =

Yf v r

; Yr& =

Yf Y ; Yr = f ; & r r

(23-26)

and 17

Nv &r =

N N f N ; N v r = f ; N r& = f ; & &r r v v r

Nr =

N f ; r

(27-30)

Where:

Yv &r = coefficient for added mass in sway

Yr& = coefficient for added mass in yaw Yvr = coefficient of sway force induced by side slip Yr = coefficient of sway force induced by yaw N v&r = coefficient for added mass moment of inertia in sway N r& = coefficient for added mass moment of inertia in yaw N vr = coefficient of sway moment from side slip N r = coefficient of sway moment from yaw

The

hydrodynamic

coefficients

for

steering

for

the

REMUS vehicle were adapted from thesis work performed by MIT (Prestero, 2001) establishing estimates of all vehicle coefficients. Force contributions from lift, drag and

added mass are summed to provide a set of combined force coefficients for both locally linearized and large angle motions. With modification, Table 2 below includes the of interest was to the by discussion adding above. The

coefficients value for

Yvr

determined

the

linearized

combined coefficients for crossflow drag, Z wc , body lift, Z wl , and fin lift, Z wf . by adding the The value for Yr was similarly determined linearized combined coefficients for The

crossflow drag, Z qc , added mass, Z qa , and fin lift, Z qf . value for

N vr

was determined from first principles using

Hoerners (1965) equation for body lift moment

1 M uwl = N uwl = d 2 c yd xcp 2

(31)

18

where the center of pressure, xcp , is centered at a point between 0.6 and 0.7 of the total body length from the nose. The moment from coefficients those in for the D rudder, N d by 3.5 to and

Yd ,

were for

scaled

Appendix

account

variation in experimental data.

Figure 7-7 of Prestero

shows a turn rate of 10 deg/sec with 4 degrees of rudder. This is approximately 3.5 times what the REMUS model

predicts (33.69 deg/sec).

Table 2.

REMUS Hydrodynamic Coefficients for Steering

Yv &r
Yr&

-3.55e01 kg 1.93 kg m/rad -6.66e01 kg/s (Same as Zw) 2.2 kg m/s (Same as Zq) 1.93 kg m -4.88 kg m2/rad -4.47 kg m/s -6.87 kg m2/s (Same as Mq) -3.46e01/3.5 kg m/s2 5.06e01/3.5 kg m/s2

Yvr
Yr

N v&r
N r&

N vr
Nr

Nd Yd

Finally, Johnson (2001) determined that rudder action produces forces that when linearized are: Y r (t ) and N r (t ) . The dynamics of the vehicle are thus defined as: 19

& r = mU o r + Yv & r + Yv & + Yr r + Y r (t ) mv &r v &r v r + Y r &r & = Nv & r + N vr v r + N r & + N r r + N r (t ) I zz r &r v &r

(32) (33) (34)

& =r

D.

VEHICLE KINEMATICS The kinematics of the vehicle are described by

Equations (32) and (33) where Ucx and Ucy are the current velocities in the associated direction. These two

equations, as well as the simple relation of heading to its derivative, compose the steering dynamics of REMUS in
& = Ax + Bu, and can be expressed as follows: matrix form, M x

m Yv&r N &r v 0

I zz N r& 0

Yr&

0 0 1

& r Yvr v r & = N vr & 0

Yr mU o Nr 1

0 v r Y 0 r + N r (t ) 0 0

(35)

where

r (t )

is

generalized

command

that

represents

the

control input to both rudders.

E.

VEHICLE DYNAMICS The final assumption made for vehicle dynamics

(Johnson, 2001) is that the cross coupling terms in the mass matrix are zero. This is based on the vehicles

symmetry and the rudders being very close to equidistant from the body center. Thus, in matrix form, the final

vehicle dynamics are defined as:

m Yv&r 0 0 0 I zz N r& 0 0 0 1 & r Yvr v r & = N vr & 0 Yr mU o Nr 1 0 v r Y 0 r + N r (t ) (36) 0 0

20

III. CONTROL METHODS AND ARCHITECTURE

A. GENERAL CONTROL THEORY Obstacle avoidance maneuvers for robots are complex in that they must be performed as a reaction to a stimulus from a sensor. and They become for an issue of even greater must

interest

concern

underwater

robots

that

execute local reflexive maneuvers, or maneuvers in which the vehicle must process a sonar return, determine if that return is a threat along its proposed path, and further navigate around the threat before regaining its original path. Through sensor measurements, nonlinear path

deviations can be developed to avoid these threats, while still scanning the underwater environment for possible

mines and other environmental data. Due to their autonomy, control of AUVs is relatively difficult. However, in spite of the uncertainty of

hydrodynamic forces, feedback control has been a suitable solution used to provide commands to actuators that control and stabilize the motion of underwater vehicles (Healey and Marco, 2001). important fact Riedel (1999) asserts that the single most contributing to the difficulty in the

control of underwater vehicles is the desire to control them along or about two or more axes. This and leads to

stronger

coupling,

larger

nonlinearities

more

state

equations in the equations of motion.

Additional factors

that contribute to the control problem are as follows:

21

A small AUV may be controllable in all six DOF Actuator dynamics are much smaller on underwater vehicles

Power and control for the vehicle is limited by the onboard capacity of the vehicle

Human

intervention

for

fault

processes

is

not

possible These same factors contribute to the obstacle avoidance

problem due to the fact that nonlinear control is necessary during avoidance maneuvers. rate and is a very REMUS has a very high turn vehicle. Thus, REMUS

responsive

requires more robust control.

This type of control can be

achieved with both sliding mode theory and through a simple dead reckoning or follow the rabbit track guidance

technique. The REMUS steering model uses autopilot controls for maneuvering based on the NPS ARIES state variable time

domain model (Marco, Healey, 2001). associated with the control

Autopilot is the name that stabilize the

systems

motion of vehicles.

As described by Marco, there are four These

different autopilots for flight maneuvering control.

consist of independent diving, steering/heading, altitude above bottom, and cross-track error controllers. All four

modes are de-coupled for ease of design and are based on sliding mode control (SMC) theory. Sliding mode control is

a robust technique, or one that provides high performance through widely varied operating conditions, used for

compensation of nonlinear systems as well as for systems whose parameters vary in a 22 predictable way with speed

(Healey, 1992). they effectively

Sliding mode controls are ideal in that replace an nth order system with an

equivalent 1st order system.

They are simple to use and

easy to implement with minimal tuning, making them ideal for use in control design. this model to include Two tuning factors are used in Eta_FlightHeading,

and

Phi_FlightHeading, , as seen in Appendix A.

B.

REMUS CONTROL ARCHITECTURE The key for to a robust control model is as the use of by

feedback

specific

motion

variables

measured

sensors to drive the vehicles actuators (control planes, rudders or thrusters). The steering controller is the only

autopilot controller necessary for modeling addressed in this thesis. for It is a second order model that uses r and modeling side-slip velocity, v, as a

feedback,

disturbance that can be overcome by the robust SMC model. Additionally, simple line-of-sight guidance is used to

maintain track path by looking ahead to planned waypoints.

C.

SLIDING MODE CONTROL Using multivariable sliding mode control methods, an

accurate

steering

controller

can

be

developed.

These

methods are used with predominantly linear system models as opposed to the SMC methods used for nonlinear systems
"

(Healey, 1992).

Revising the EOM for a state variable r (t )

to the general form

x = Ax + Bu
23

"

(37)

where x R n*1 ; A R n*n ; B R n*r ; u R r *1 , and u is the rudder angle, a SMC can be designed to drive this state to stable solution, or one in which the sliding surface = 0, R 1 . sliding surface defined as: %; x % = xx = s'x com where space. of the s is a vector of directions in the state (38) error With the

The elements of are the lengths of the projection state error vector,

xcom

is

dynamic

exogenous

% is the variable created as a command signal to track, and x

state error which is required to be driven to zero so that the command state equals actual state. The values of s are found by the requirement that when = 0 , the system dynamics must exhibit stable sliding on the surface. Thus,

the closed loop dynamics are given by the poles of the closed loop matrix as,

( A bk2 ) = Ac , with k2 = [ s ' B]1 s ' A

(39)

where k2 is chosen by pole placement and Acs=0 to achieve the condition

= 0.

The

eigenvectors

of

the

Ac

matrix

determine the linear state feedback gains for each state used to define the sliding surface as follows: " (t ) (t ) = s2 "(r_ com r (t )) + s3 " LOS (40)

The poles selected for the REMUS model SMC solution were moved farther from zero than those in the original ARIES model in order to stabilize the system dynamics. As seen

in Appendix A, these poles were placed at [1.4 -1.45 0.0]. The pole at the origin is necessary to allow for the single sliding constraint for the single input system implied by 24

 = 0 . The remaining poles both exhibit stable dynamics as

they are in the left half plane. The gains obtained from this pole placement were [k1 k2 k3] = [0.769 0.6 0.0] for [v, r, ] respectively. in equation (39) and With the sliding surface defined the gains determined in the LOS from pole

placement, becomes:

the

commanded

rudder

controller

dr (t ) = k 2"r (t )" "tanh( (t ) / )

(41)

where and are tuning factors equal to 0.5 and 0.1 respectively.

D.

LINE OF SIGHT GUIDANCE This purpose of the Line of Sight (LOS) controller is

to

reduce the

the

heading

error

to

zero. for

The ARIES

REMUS

model and

adapts

original

LOS

guidance

[Marco

Healey (2001)] with a follow-the-rabbit technique similar in nature to the transducer based dead-reckoning approach with which REMUS operates. The LOS controller forces the

vehicle to head in the direction of the current waypoint by

" , as the difference defining the error in the heading, LOS

between the commanded line of sight and the actual heading, or:

% (t ) LOS = (t )track (t )
% (t ) % where (t )track = arctan(Y wpt ( i ) , X (t ) wpt ( i ) )

(42) (43)

The commanded heading is based on the angle between the current position and the next waypoint. simply adds an additional on the 25 look-ahead toward The REMUS model point the next or dead-

reckoning

point

track

waypoint

forward of the vehicle position as seen in Figure 3 below. The distance to this point is incorporated into the heading error as follows:

% (t ) LOS = (t )track (t ) arctan(cte(t ) / rabbit )

(44)

where rabbit is the look-ahead point and cte is the cross track error between the actual vehicle position and the desired track.

Figure 3. Track Geometry and Velocity Vector Diagram

While LOS guidance controls REMUS along the track from waypoint to waypoint, a different method is used to

determine when to turn as the waypoints are approached. The following command is used to ensure that REMUS will begin tracking the next waypoint when approaching the

present waypoint:

26

sqrt ( X _ Way _ Error (t ) 2 + Y _ Way _ Error (t ) 2 ) <= W _ R s(t) < 0.0 ss (t ) < rabbit (45)

where W_R is the watch radius around the waypoint, s is the distance remaining on track, and ss is the radial distance to go to the next waypoint. Thus, REMUS will begin to

track off the next waypoint if it has entered the watch radius around its present waypoint, if is has passed its present waypoint, or if the rabbit distance is greater than the radial distance to go to the waypoint.

27

THIS PAGE INTENTIONALLY LEFT BLANK

28

IV. OBSTACLE AVOIDANCE MODEL

A. THE REMUS SEARCH PATH Path planning for the REMUS vehicle is based on the information to be gathered during a mission. in minefields to search and classify REMUS is used objects

mine-like

whose location is frequently known. widely used to map the very shallow

However, it is also water zone of the

littoral region where an accurate map may not exist to provide hydrographic maps with for use by fleet units. The

search path used for this vehicle is commonly referred to as the lawnmower technique and is used to cover a square grid area. Depending on search area and target detection

analysis performed prior to a mission, this path may vary. This thesis models a REUMUS path that uses rows

approximately 200 meters in length with 15-40 meters of separation as seen in Figure 4 below.
Typical REMUS Search Path 50

40

30

20

10

-10

-20

-30

-40

-50

50

100

150

200

250

Figure 4. Typical REMUS Search Path 29

B.

SONAR MODEL This model uses a two-dimensional forward-looking

sonar with a 120 horizontal scan and a 110-meter radial range as seen in Figure 5. based on a viable 400KHz This is an estimated range sonar for frequency. underwater As Lane

contends,

obstacle

avoidance

vehicles

necessitates high resolution, reliable, multi-beam sonars of this type (Lane, 2001). The probability of detection is

based on a cookie-cutter approach in which the probability of detection is unity within the scan area and zero

anywhere else.

Bearing is measured to the nearest degree

and range is measured every meter.

110 m

REMUS

120

Figure 5. Forward-look Sonar Model

The advantage of using a forward-looking sonar over side-scan sonars in object avoidance is twofold. One, it

allows for scanning ahead of the vehicle which facilitates reaction to detected obstacles, and two, it allows for

possible overlap of acoustic imagery ahead of the vehicle providing more accurate detection information. currently configured with two side-scan sonars. REMUS is Based on

the swath width of the sonar, REMUS must make narrow passes 30

over a given area at 15-40 meter increments for adequate coverage of the sea floor. more difficult to mount, A forward look sonar, while would prove more capable in

preventing collision and would allow for mapping a more efficient path in cluttered environments.

C.

HEURISTICS There are several methods used for obstacle avoidance

in robot vehicles today. 1.)

(Several are outlined in Chapter

The obstacle avoidance model developed in this thesis

is based on the product of bearing and range weighting functions that form the gain factor for a dynamic obstacle avoidance behavior. The basis for the weighting functions The weighting functions from the fuzzy logic

lies in a fuzzy logic methodology. are MATLAB membership functions

toolbox with the parameters selected to maximize obstacle avoidance behavior. The membership function for bearing is

a Gaussian curve function of the form:

-( x - c )2

w1=1 0

(2

(46)

where the parameters x, c, and are position (or angular position in degrees for the purpose of this model), center, and shape respectively. Gaussian curve. Shape defines the steepness of the

The values selected for these parameters

to provided sufficient tuning in this membership function were -90:90, 0, 20 respectively. The bearing weighting

function can be seen in Figure 6 below.

31

Bearing Weighting Function 1

0.9

0.8

0.7

0.6 Weight

0.5

0.4

0.3

0.2

0.1

0 -100

-80

-60

-40

-20

0 20 Bearing (degrees)

40

60

80

100

Figure 6. Bearing Weighting Function

It is evident that the weight given to an object dead ahead of the vehicle is closer to unity than one that is over 30 to port or starboard. The membership function for range is an asymmetrical polynomial spline-based curve called zmf for its z shape and is of the form

w2= zmf (x,[a b])

(47)

where a and b are parameters that locate the extremes of the sloped portions of the curve. called breakpoints In and define to These parameters are the curve changes avoidance

where

concavity.

order

maximize

obstacle

behavior, these values were tuned to be (sonrange-99) and (sonrange-90). With this selection, the range weight is

approximately unity for anything closer than 20 meters and zero for anything farther than 40 meters from REMUS. range weighting function can be seen in Figure 7 below. 32 The

Range Weighting F i

0.9

0.8

0.7

0.6 Weight

0.5

0.4

0.3

0.2

0.1

20

40

60 Range (m)

80

100

120

Figure 7. Range Weighting Function

A final weight based on both bearing and range is calculated from the product of w1 and w2. This weight

becomes the gain coefficient that is applied to a maximum avoidance heading for each individual object. The maximum heading is / 4 as seen below:

oa (t , c) = w1w2( / 4)
where t is the time step and c is the obstacle

(48) being

evaluated.

The avoidance heading for all obstacles over a

single time step (or one look) is then

oalook (t ) = oa (t , c)
1

(49) at every time

Following

an

evaluation

of

each

obstacle

step, a final obstacle avoidance heading term is determined from the sum of the obstacle avoidance heading of each individual object within a specified bearing and range from the vehicle or 33

 oatot (t ) =
where cc is the counter

oalook (t )
cc
used to determine how

(50) many

obstacles fall into this window. normalize this overall obstacle

The counter is used to avoidance term to an

average for all of the obstacles within the range above. This bearing and range of the window is determined through a rough evaluation of the weighting functions. In order to

fall into the window, the gain factor must be equal to or exceed a value of w1w2=0.15. The obstacle into avoidance vehicle term

oatot(t)
error

is

then in

incorporated

heading

(discussed

Chapter 3, equation (43)) as:

% LOS (t ) = track (t ) cont (t ) arctan(cte(t ) / rabbit ) + oatot (t )

(51)

This heading error drives the rudder commands to maneuver around detected objects in the track path. The overall

object avoidance system dynamics can be seen in the diagram below:

Figure 8. Block Diagram System Dynamics

34

V.
A.

VEHICLE SIMULATION

BASIC SINGLE POINT OBSTACLE AVOIDANCE The initial test performed on the two-dimensional

sonar model was navigation around a single point obstacle. This is the simplest obstacle avoidance test for the 2-D model. Three variations of this test were run for the The first was for a

basic single point obstacle avoidance. single point on the path.

The second was for a single Finally, a run was

point to the right or left of the path.

performed to test the accuracy of the steering and obstacle avoidance model for each of the four quadrants. This was

achieved by running the REMUS through a figure-eight path that had a single point obstacle at the midpoint of each leg. Results for single point obstacle runs can be seen in the figures below. The first two tests were repeated for a

cluster of points designed to mimic an obstacle with length and width both on the path and just off the path and will be addressed in the next section.
Obstacle 40

30

20

10

0 Y (m) -10 -20 -30 -40 -50 0

50

100 X (m)

150

200

250

Figure 9. Single Point Obstacle Run (On Path) 35

Rudder-r (deg), Heading-b(deg), Psioatot-g(deg) 150

Rudder-r (deg), Heading-b(deg), Psioatot-g(deg) 100

100

80

60
50

g e D 0

g e D

40

20

0
-50

-20
-100

50

100

150

200 Time

250

300

350

400

450

40

50

60 Time

70

80

90

Figure 10. Single Point Obstacle Run: Rudder/Heading/oa

40

Obstacle Run

30

20

10

0 Y (m) -10 -20 -30 -40 -50 0

50

100 X (m)

150

200

250

Figure 11. Single Point Obstacle Run (Off Path)

36

Rudder-r (deg), Heading-b(deg), Psioatot-g(deg) 150

100

50

Deg

-50

-100 0 50 100 150 200 250 Time (sec) 300 350 400 450

Figure 12. Rudder/Heading/oa (Off Path) Figures 10, 12, and 14 show the rudder dynamics, vehicle heading, and obstacle avoidance heading term for the

duration of each vehicle run.

The rudder action has a

direct correlation with the obstacle avoidance heading and overall vehicle heading. heading are the The large angle motions of the turns made to track the

ninety-degree

ordered vehicle path. with each of these

There is an associated rudder action turns 9. a as seen by the corresponding

rudder curve. programmable whether

These rudder curves show that the maximum rudder is with For turn all or dynamic obstacle behaviors, avoidance

associated

maneuver, the rudder initiates the turn with this maximum value. vary. In order to regain track, the rudder action may The major heading changes to track the path require

a full rudder for a longer period of time than do the obstacle avoidance heading changes. constant horizontal value 37 on the This is evident in the rudder curve. The

difference in Figure 9 and Figure 11 is in the direction of turn to maneuver around the obstacle. When the obstacle is When it is

on the path, the vehicle maneuvers to the left.

off the path, the vehicle maneuvers to the opposite side of the obstacle.
Obstacle Run 50

40

30

20

10 Y (m)

-10

-20

-30

-40

-50 -30

-20

-10

10 X (m)

20

30

40

50

Figure 13. Figure-Eight Obstacle Run

Rudder-r (deg), Heading-b(deg), Psioatot-g(deg)

Rudder-r (deg), Heading-b(deg), Psioatot-g(deg) 150

100

100

80

50

60

0 Deg
Deg 40

-50
20

-100
0

-150
-20

-200

20

40

60

80 Time (sec)

100

120

140

160

35

40

45

50 Time (sec)

55

60

65

Figure 14. Rudder/Heading/oa Figure-Eight 38

Figures 13 and 14 show the results for the vehicle run through the figure-eight path. not maintain the track as Although the vehicle does as it does the

accurately

previous runs, it completes the run with proper dynamics for each quadrant. The obstacle avoidance heading is not

equal for each of the obstacle avoidance behaviors due to the fact that the vehicle is not weighing the same number of obstacles along each leg of the figure-eight. It only

weights the obstacles that fall within the scan with the proper proximity as described in Chapter IV. B. MULTIPLE POINT OBSTACLE AVOIDANCE A single point obstacle avoidance model is far simpler than a multiple point obstacle avoidance model not only in the maneuvering of the vehicle, but also in maintaining the obstacle picture. For multiple point obstacle

avoidance, it is necessary to have a model that reacts to obstacles in a certain proximity to its path rather than all possible obstacles seen by the sonar scan. functions obstacle allow picture. for an The accurate REMUS model compilation builds an Weighting of this

obstacle

counter for obstacles having a weighting function product greater than 0.15 as discussed in the previous chapter. This value allows for a maximum rudder and bearing weight of approximately 0.386, the square root of 0.15. Referring

to the membership functions in Figure 6 and Figure 7, a value of 0.386 correlates to a bearing and range of approximately +/-30 and 30 meters respectively. As seen in the following figures, REMUS successfully avoids multiple points and multiple point clusters in the same fashion it avoided a single 39 points. The rudder

dynamics are minimal during all avoidance maneuvers for an efficient model. point or multiple All of the obstacle runs, for single point obstacle avoidance, show REMUS

responding to obstacles in advance of the actual obstacle position. refined While this model has not been optimized with the early response time would allow

techniques,

sufficient processing time in an actual sonar return for real-world environments. The dynamics of REMUS are very

reactive such that REMUS regains the track path directly after the passing an obstacle. ideal due to the proximity Though this behavior is not at which REMUS passes the

obstacle, through optimization, it could be improved.

Obstacle Run

50

40

30 20

10 Y (m) 0

-10 -20

-30 -40

-50 0

50

100 X (m)

150

200

250

Figure 15. Multiple Single Point Obstacle Run

40

Rudder-r (deg), Heading-b(deg), Psioatot -(deg) 150

100

50
Deg

-50

-100

50

100

150

200

250 Time (sec)

300

350

400

450

Figure 16. Rudder/Heading/oa

Obstacle Run

50 40 30 20 10 Y (m) 0 -10 -20 -30 -40 -50 0

50

100 X (m)

150

200

250

Figure 17. Multiple Point Obstacle Run

41

Rudder-r (deg), Heading-b(deg), Psioatot-g(deg) 150 100

Rudder-r (deg), Heading-b(deg), Psioatot-g(deg)

100

80

60 50 Deg Deg 0 20 -50 -100 0 50 100 150 200 250 Time (sec) 300 350 400 450

40

-20 30 40 50 60 70 Time (sec) 80 90 100 110

Figure 18. Multiple Point Obstacle Run: Rudder/Heading/oa The vehicle heading in Figure 18 (bottom right) can be offset by 90 in or to compare the vehicle dynamics with the obstacle avoidance heading. As seen in Figure 19 below, an 90

obstacle appearing in the vehicle path causes the vehicle heading to deviate the from its track as path heading of

approximately heading. the total

same

amount

the

obstacle

avoidance

These two headings do not exactly match because heading incorporates additional factors as in

equation (51).
Rudder-r (deg), Heading-b(deg), Psioatot-g(deg) 15

10

Deg

-5

-10

-15

-20

-25 75 80 85 90 95 Time (sec) 100 105 110 115

Figure 19. Vehicle Heading Comparison with 90 Offset 42

The above figures present the obstacle avoidance for the weighting functions described in the previous chapter. A comparison can be made for different values of the

weighting functions to show the utility of the selected functions. breakpoints A range at weighting function and that uses

defined

(sonrange-95)

(sonrange-70) Figure

changes the vehicle dynamics around the obstacles.

20 shows the curve for this alternate weighting function.

Range Weighting Function 1

0.9

0.8

0.7

0.6 Weight

0.5

0.4

0.3

0.2

0.1

20

40

60 Range (m)

80

100

120

Figure 20. Alternate Range Weighting Function The vehicle response is too early with the alternate breakpoints, although the off track distance increases by approximately a half meter. For mine countermeasures

operations, a higher off track distance increases vehicle safety. side Thus, However, the sonar configuration of REMUS supports imaging as off well track as possible is forward more looking. for

scan

minimizing

distance

ideal

obtaining accurate side scan data.

Figure 21 shows the

dynamic behavior comparison of the two weighting functions.

43

Obstacle Run 50

40

30

20

10
Y (m)

-10

-20

-30

-40

-50

50

100 X (m)

150

200

250

Obstacle Run 46

Obstacle Run

16 44 14 42 12
Y (m)

40

Y (m)

10

38

6 36 4 34 70 80 90 X (m) 100 110 120 60 70 80 90 X (m) 100 110 120 130

Figure 21. Range Weighting Function Dynamics Comparison

44

VI. CONCLUSIONS AND RECOMMENDATIONS

A. CONCLUSIONS Obstacle avoidance for autonomous vehicles is widely studied for a variety of applications. This thesis focuses One of the

on a particular application for the REMUS AUV.

most critical factors in obstacle avoidance behavior is the ability to discern how a vehicle will react to its

environment.

It is necessary to model realistic sensors

that gather sufficient environmental data for safe vehicle navigation. The sensor modeled in this thesis will be used

by the Center for Autonomous Underwater Vehicle Research in future operations and requires an accurate model prior to implementation. The model shows that with appropriate

onboard processors, the REMUS vehicle could, if necessary, execute a local reflexive maneuver. to use range and bearing data REMUS has the ability a sonar return to

from

determine if that return constitutes a threat along its proposed path and further navigate around the threat before regaining its original path. Through weighting functions,

nonlinear path deviations can be achieved to avoid these threats, while still scanning the underwater environment for possible mines and other environmental data. There remains a need for a fast and effective means of interpreting the sonar data. Visual analyses of sonar

returns are made daily in naval applications.

This ability

has to be effectively implemented in an underwater vehicle for obstacle avoidance to be successful. One method would

be through analysis of shadow areas in sonar returns.

45

Often, a sonar scan does not pick up the same obstacle each time it passes over a given area. However, multiple

scans with positive detection over a decreasing range will allow for the processing system to correlate a positive detect on a specific bearing and range to an obstacle. Thus, the model developed in this thesis accurately

represents a sonar in that on each time step, the vehicle sees every object within the bearing and range of the scan. A last point to be made for this model is a concern for the overuse of actuators for dynamic movements. multiple point obstacle field with several In a

dynamic

movements, the vehicle has a significant number of rudder bangs or direction changes in a very short period of time. This dynamic rudder action will dissipate power and Thus, a more to non-

will quickly wear out the servomechanisms. robust design, or one that eliminates

response

hazardous obstacles, might be necessary in a high clutter environments.

B.

RECOMMENDATIONS There are many areas in which this thesis work can be

improved upon to build a more complete and robust obstacle avoidance model for the REMUS vehicle. The most obvious

but most complicated of these is the development of a three dimensional model. This would require adding a depth to

the sonar scan such that the scan would cover somewhere from ten degrees above the horizontal to thirty degrees below the horizontal. In order to implement such a model,

the vehicle EOM would have to be modified to include diving and climbing maneuvers for obstacle avoidance. 46 To produce

a more exact model, it would be necessary to conduct an open water test with the REMUS vehicle to determine

hydrodynamic coefficients for diving as well as steering. Additionally, the incorporation of a CTE controller to the steering model once experimental data is obtained would

make it more robust.

A CTE controller is not functional in

the model at present due to the lack of experimental values for coefficients in the CTE equations. A second addition to the proposed model that would increase its utility would be through speed control. A

model with acceleration and deceleration capability would allow for more dynamic obstacle avoidance. For example, if

REMUS turned to an area of increased obstacles due to an obstacle avoidance command from some other object in its path, a speed reduction could follow to permit data

processing prior to driving a new path. A speed controller would be useful for a model that incorporates moving obstacles as well as stationary. proposed model uses only stationary obstacles in The the

vehicle path.

By incorporating a range rate variable into

the avoidance control, the vehicle could compare its own speed with the relative speed at which it closes the

obstacle and thus determine if the detected obstacle is moving. Use of range rate data would allow REMUS to better

determine the safest path around obstacles. The Fuzzy Logic methodology used to develop weighting functions for obstacle avoidance behavior may not be the most accurate method available. However, simple additions

to this model could make it more accurate, such as using range rate as a weighting 47 factor. Additionally, an

optimization could be performed on the implementation of the weighting function gain factor so that REMUS clears each obstacle by a specified distance, does not begin

avoidance behavior too early, and does not return to track at such sharp angles. Finally, information errors must be in vehicle into position account and for sensory dynamic

taken

behaviors to be accurate.

Currently, REMUS navigates a

track through transponder cross-fix data that has about a 2-3 meter positional error associated with it. steering longer model runs under the assumption GPS that While the REMUS no may

uses

these

transponders,

position

errors

still be a factor.

If future REMUS vehicles can operate

using autopilots in the steering model, there will be only slight errors in sensory information as the position of the obstacles are in a local frame of reference with respect to the vehicle. model uses Additionally, the REMUS obstacle avoidance these relative positions to plan and reflexive

maneuvers.

Through

Concurrent

Mapping

Localization

(CML) techniques, REMUS could store obstacles it has passed in a database for use in planning a return path to its original position or for a possible rendezvous for data transfer. techniques. Ruiz (2001) gives a thorough overview of CML

48

APPENDIX A
% This mfile uses corrected hydrodynamic coeff from MIT to develop % a steering model. It models REMUS running through a field of % multiple obstacles, both single points and those with lenght % and width. clear clf clc % REMUS Characteristic Specifications: L = W = g = m = V = rho D = 1.33; 2.99e02; 9.81; W/g; 1.543; = 1.03e03; .191; % Length in m % Weigth in N % Acceleration of gravity in m/s^2 % Mass in kg % Max Speed in m/s % Density of Salt H20 in kg/m^3 % Max diameter in m

%State Model PArameters U = 1.543; % m/s Boy = 2.99e02; xg = 0; yg = 0; zg = 1.96e-02; % in m Iy = 3.45; %kg/m^3 (from MIT thesis) Iz=Iy; % MIT REMUS Coeff (Dimensionalized) disp('MIT REMUS Coefficients'); Nvdot = 1.93; Nrdot = -4.88; Yvdot = -3.55e01; Yrdot = 1.93; %Nv = -4.47; should be same as Mw which is stated as +30.7 % should be -9.3 but going by Hoerner eqn, we get about 4.47 Nv = -4.47; Nr = -6.87; %Same as Mq; Yv = -6.66e01; %Same as Zw; Note should be -6.66e1 from MIT thesis not 2.86e01 Yr = 2.2 ; %Same as Zq = 2.2; MIT has miscalculation Nd = -3.46e01/3.5; % Nd and Yd scaled by 3.5 to align w/exp data Yd = 5.06e01/3.5; % The Steering Equations for the REMUS are the following. % These equations assume the primarily horizontal motions ... MM=[(m-Yvdot) -Yrdot 0;-Nvdot (Iz-Nrdot) 0;0 0 1]; AA=[Yv (Yr-m*V) 0;Nv Nr 0; 0 1 0]; BB=[Yd;Nd;0];

49

A=inv(MM)*AA; B=inv(MM)*BB; C=[0,0,1]; D=0; A2=[A(1:2,1),A(1:2,2)];B2=[B(1);B(2)]; xss=inv(A2)*B2; poles = eig(A2); RadGy = sqrt(Iz/(W/g)); % in meters RadCurv = U/(xss(1)); % in meteres SideSlip = atan2(xss(1),U)*180/pi; % in deg/s [num,den]=ss2tf(A,B,C,D); z=roots(num); p=roots(den); % Desired closed loop poles for sliding: k=place(A,B,[-1.4,-1.45,0.0]); % Closed loop dynamics matrix Ac=A-B*k; [m,n]=eig(Ac'); S=m(:,3); % *************************************** TRUE = 1; FALSE = 0; DegRad = pi/180; RadDeg = 180/pi; % Define Obstacles:(put them in near track for trial runs) Xo(1) = 10; % First obstacle x-dist ref global origin in m owidth(1) = 1; % First obstacel width in m Yo(1) = 90; % First obstacle y-dist ref global orinin in m olgth(1) = 1; % First object length in m Xo(2) = 40; % Second obstacle x-dist ref global origin in m owidth(2) = 5; % Second obstacle width in m Yo(2) = 100; % Second obstacle y-dist ref global origin in m olgth(2) = 3; % Second obstacle length in m Xo(3) = 6; % Third obstacle x-dist ref global origin in m owidth(3) = 3; % Third obstacle width in m Yo(3) = 110; % Third obstacle y-dist ref global origin in m olgth(3) = 3; % Third obstacle length in m numobs = 3; numpts = 0; for p=1:numobs numpts=numpts + owidth(p)*olgth(p); end AreaObs = []; psioa=zeros(8000,numobs);

50

Define Sonar Grid Parameters:

sonrange = 110; % radial range in m based on 400 KHz frequency theta = 2*pi/3; % angular arc in rad % Builds obstacles in Xo and Yo matrices: Xobs=[]; Yobs=[]; for p = 1:numobs % model each point as an obstacle so the sonar can see them individually for pp=1:olgth(p) if owidth(p)>1 for q = 1:(owidth(p)) Xobs=[Xobs,(Xo(p)+(q-1))]; Yobs=[Yobs, (Yo(p)+(pp-1))]; end elseif owidth(p)==1 Xobs=[Xobs, Xo(p)]; Yobs=[Yobs, Yo(p)]; else Xobs = Xobs; Yobs = Yobs; end end end % Set time of run dt = 0.125/2; t = [0:dt:1800]'; size(t); % Set initial conditions start=10; v(1) = 0.0; r(1) = 0.0; rRM(1) = r(1); % This is the Initial Heading of the Vehicle psi(1) = 50.0*DegRad; % This is the Initial Position of the Vehicle X(1) = -50.0; % Meters Y(1) = 10; % This data from track.out file No_tracks=7; Track=[10.0 10.0 2.75 2.75 0 1.25 1.00 0 25.00 8.00 40.00 10.0 210.0 2.75 2.75 0 1.25 1.00 0 25.00 8.00 200.00 25.0 210.0 2.75 2.75 0 1.25 1.00 0 25.00 2.00 15.00 25.0 10.0 2.75 2.75 0 1.25 1.00 0 25.00 2.00 200.00 40.0 10.0 2.75 2.75 0 1.25 1.00 0 25.00 2.00 15.00 40.0 210.0 2.75 2.75 0 1.25 1.00 0 25.00 2.00 200.00 41.0 210.0 2.75 2.75 0 1.25 1.00 0 25.00 2.00 1.0]; track=Track(:,1:2); SurfaceTime = Track(:,9);

51

SurfPhase

= Track(:,8);

% Read in wayopoints from track data assumes track is loaded for j=1:No_tracks, X_Way_c(j) = track(j,1); Y_Way_c(j) = track(j,2); end; PrevX_Way_c(1) = X(1); PrevY_Way_c(1) = Y(1); r_com = 0.0; % Set Rudder angle saturation: sat = 9; % Degrees % Set Watch Radius: W_R = 2.0; % Set dead-reckoning/look-ahead distance: rabbit = 9;

x(:,1) = [v(1);r(1);psi(1)]; Eta_FlightHeading = 0.5; model Phi_FlightHeading = 0.1; model % Lowered this from 1.0 on AERIES % Lowered this from 0.5 on AERIES

% Below for tanh Eta_CTE = 0.05; % (NA given that no CTE controller is used) Eta_CTE_Min = 1.0; Phi_CTE = 0.2; % (NA given that no CTE controller is used) Uc = []; Vc = []; SegLen(1) = sqrt((X_Way_c(1)-PrevX_Way_c(1))^2+(Y_Way_c(1)PrevY_Way_c(1))^2); psi_track(1) = atan2(Y_Way_c(1)-PrevY_Way_c(1),X_Way_c(1)PrevX_Way_c(1)); for j=2:No_tracks, SegLen(j) = sqrt((X_Way_c(j)-X_Way_c(j-1))^2+(Y_Way_c(j)Y_Way_c(j-1))^2); psi_track(j) = atan2(Y_Way_c(j)-Y_Way_c(j-1),X_Way_c(j)X_Way_c(j-1)); end; j=1; Sigma = []; Depth_com = [];

52

dr=[]; drl = []; drl(1) = 0.0; Depth_com(1) = 5.0; WayPointVertDist_com = [5.0 5.0 5.0 5.0 5.0 5.0 5.0]; for i=1:length(t)-1 Depth_com(i) = WayPointVertDist_com(j); X_Way_Error(i) = X_Way_c(j) - X(i); Y_Way_Error(i) = Y_Way_c(j) - Y(i); % DeWrap psi to within +/- 2.0*pi; psi_cont(i) = psi(i); while(abs(psi_cont(i)) > 2.0*pi) psi_cont(i) = psi_cont(i) - sign(psi_cont(i))*2.0*pi; end; psi_errorCTE(i) = psi_cont(i) - psi_track(j); % DeWrap psi_error to within +/- pi; while(abs(psi_errorCTE(i)) > pi) psi_errorCTE(i) = psi_errorCTE(i) sign(psi_errorCTE(i))*2.0*pi; end;

% **

Always Calculate this % Beta = v(i)/U; Beta = 0.0; cpsi_e = cos(psi_errorCTE(i)+Beta); spsi_e = sin(psi_errorCTE(i)+Beta);

s(i) = [X_Way_Error(i),Y_Way_Error(i)]*... [(X_Way_c(j)-PrevX_Way_c(j)),(Y_Way_c(j)PrevY_Way_c(j))]'; % s is distance to go projected to track line(goes from 0100%L) s(i) = s(i)/SegLen(j); Ratio=(1.0-s(i)/SegLen(j))*100.0; % ss is the radial distance to go to next WP ss(i) = sqrt(X_Way_Error(i)^2 + Y_Way_Error(i)^2); % dp is the angle between line of sight and current track line dp(i) = atan2( (Y_Way_c(j)-PrevY_Way_c(j)),(X_Way_c(j)PrevX_Way_c(j)) )-atan2(Y_Way_Error(i),X_Way_Error(i) ); if(dp(i) > pi), dp(i) = dp(i) - 2.0*pi;

53

end; cte(i) = s(i)*sin(dp(i)); if( abs(psi_errorCTE(i)) >= 00.0*pi/180.0) %| s(i) < 0.0 ), %used to read 40.0*pi not 00.0*pi for CTE controller % Use LOS Control LOS(i) = 1; psi_comLOS(i) = atan2(Y_Way_Error(i),X_Way_Error(i)); %psi_comLOS = pi/2; % Test for heading controller stability % Construct Bearing/Range to each obstacle(point): cc=0; psioalook(i)=0; increaseweight=FALSE; for c=1:numpts Bearing(i,c) = atan2((Yobs(c)-Y(i)),(Xobs(c)-X(i)))psi(i); Range(i,c) = sqrt((Yobs(c)-Y(i))^2+(Xobs(c)-X(i))^2); if Range(i,c)<=sonrange & (theta/2<=Bearing(i,c)<=theta/2) % Use Fuzzy logic if c>1 sepang=abs(Bearing(i,c)-Bearing(i,(c-1))); if ((Range(i,c))^2 + (Range(i,(c-1)))^2 2*Range(i,c)*Range(i,(c-1))*... cos(sepang))<2*D increaseweight=TRUE; end end % Develop weighting factor based on Range (w1) w = 0:1:sonrange; w1(i)=zmf(Range(i,c), [(sonrange-99) (sonrange90)]); % [] above are breakpoints in the curve % Develop weighting factor based on Bearing (w2) Posit = (-90:1:90)'; % in degrees Center = 0; Shape = 20; w2(i)=gaussmf(Bearing(i,c)*RadDeg, [Shape, Center]); % A MATLAB membership function: EXP(-(Posit Center).^2/(2*Shape^2)) if increaseweight w1(i)=2*w1(i); w2(i)=2*w2(i); end % Only want to weight the obstacle c once in each time step if (w1(i)*w2(i))>0.15 cc=cc+1; % counter for obstacles in being

54

avoided at time t % Object Bears to Left if Bearing(i,c)>0 psioa(i,c)=-w1(i)*w2(i)*(pi/4); % Object Bears to Right elseif Bearing(i,c)<=0 psioa(i,c)=+w1(i)*w2(i)*(pi/4); end psioalook(i)=(psioalook(i)+psioa(i,c)); end end end if cc>0 psioatot(i)=psioalook(i)/cc; else psioatot(i)=psioalook(i); end psi_errorLOS(i) = psi_track(j) - psi_cont(i)atan2(cte(i),rabbit) + psioatot(i); if(abs(psi_errorLOS(i)) > pi), psi_errorLOS(i) = ... psi_errorLOS(i) 2.0*pi*psi_errorLOS(i)/abs(psi_errorLOS(i)); end; Sigma_FlightHeading(i) = (-S(1,1)*v(i))*0.0+S(2,1)*(r_com r(i)) + S(3,1)*psi_errorLOS(i); % Have taken out v influence in Sigma_FlightHeading above dr(i) = (-k(1,1)*v(i))*0.0-k(1,2)*r(i)Eta_FlightHeading*tanh(Sigma_FlightHeading(i)/Phi_FlightHea ding); %dr(i) = -k(1)*v(i)-k(2)*r(i)+k(3)*(psi_errorLOS(i)); %if ((i>1200)&(i<1400)); dr(i)=-4*DegRad; end; % turn test else % Use CTE Controller LOS(i) = 0; if(cpsi_e ~= 0.0), % Trap Div. by Zero ! % SMC Soln Sigma(i) = U*rRM(i)*cpsi_e + Lam1*U*spsi_e + Lam2*cte(i); dr(i) = (1.0/(U*a*cpsi_e))*(-U*b*rRM(i)*cpsi_e + U*rRM(i)^2*spsi_e-Lam1*U*rRM(i)*cpsi_e Lam2*U*spsi_e 2.0*Eta_CTE*(Sigma(i)/Phi_CTE)); else dr(i) = dr(i-1); end;

55

end; % End of CTE Controller if(abs(dr(i)) > sat*pi/180) % change from 0.4 radians on ARIES dr(i) = sat*pi/180*sign(dr(i)); end; % State Variable Formulation: x_dot(:,i+1) = [ A(1,1)*v(i) + A(1,2)*r(i) + B(1)*dr(i); A(2,1)*v(i) + A(2,2)*r(i) + B(2)*dr(i); r(i)]; x(:,i+1) = x(:,i)+dt*x_dot(:,i); v(i+1) = x(1,i+1); r(i+1) = x(2,i+1); psi(i+1) = x(3,i+1); rRM(i+1) = r(i+1); % Wave Motions: Uc = 0.0; Vc = 0.0; %Kinematics X(i+1) = X(i) + (Uc + (U)*cos(psi(i)) - v(i)*sin(psi(i)))*dt; Y(i+1) = Y(i) + (Vc + (U)*sin(psi(i)) + v(i)*cos(psi(i)) )*dt; % % Check to See if we are Within the Watch_Radius or if we passed the WP Change to next WP if radial distance to go is less than rabit distance or if we passed the WP or if we are within the WR if(sqrt(X_Way_Error(i)^2.0 + Y_Way_Error(i)^2.0) <= W_R | s(i) < 0.0 | ss(i)<rabbit), disp(sprintf('WayPoint %d Reached',j)); if(j==No_tracks), break; end; PrevX_Way_c(j+1) = X_Way_c(j); PrevY_Way_c(j+1) = Y_Way_c(j); j=j+1; end; end; dr(i+1) = dr(i); cte(i+1) = cte(i); s(i+1) = s(i); ss(i+1) = ss(i); % Plot this obstacle run: figure(1); grid; plot(t([1:i+1]),dr*180/pi,'r'); hold; plot(t([1:i+1]),psi*180/pi,'b');

56

plot(t([1:i]),psioatot*180/pi,'g'); ylabel('Deg'); xlabel('Time'); title('Rudder-r (deg), Heading-b(deg), Psioatot-g(deg)'); figure(2) axis equal; plot(Y,X); title('Obstacle Run'); xlabel('X (m)'), ylabel('Y (m)') grid; hold plot([Y_Way_c(1) PrevY_Way_c(1)],[X_Way_c(1) PrevX_Way_c(1)],'r'); for ii=2:No_tracks, plot([Y_Way_c(ii) Y_Way_c(ii-1)],[X_Way_c(ii) X_Way_c(ii1)],'r'); end; % Plot Obstacles plot(Yobs, Xobs,'gh'); hold; figure(3); grid; plot(t([1:i+1]),dr*180/pi,'r'); hold; plot(t([1:i+1]),(psi-pi/2)*180/pi,'b'); plot(t([1:i]),psioatot*180/pi,'g'); ylabel('Deg'); xlabel('Time'); title('Rudder-r (deg), Heading-b(deg), Psioatot-g(deg)');

57

THIS PAGE INTENTIONALLY LEFT BLANK

58

LIST OF REFERENCES

Blidberg, Richard D., The Development of Autonomous Underwater Vehicles (AUVs); A Brief Summary, Autonomous Undersea Systems Institute, ICRA, Seoul, Korea, May 2001. Borenstein, J. and Koren, Y., The Vector Field HistogramFast Obstacle Avoidance for Mobile Robots, IEEE Journal of Robotics and Automation Vol 7, No 3, Jun 1991, pp. 278-288. Concept of Operations, Organic Offboard Mine Reconnaissance, Director, Naval Expeditionary Warfare Division (N85) and Submarine Warfare Division (N87). Gray, Keith W., Obstacle Detection and Avoidance for an Autonomous Farm Tractor, M.S. Utah State University, Logan, UT, 2000. Gill, Mark A. C., Obstacle Avoidance in Multi-Robot Systems, World Scientific Publishing Company Pte. Ltd., 1998. Healey, A.J., Dynamics of Marine Vehicles (ME-4823), Class Notes, Naval Postgraduate School, Monterey, CA, 1995. Healey, A. J., Command and Control Demonstrations with Cooperating Vehicles, ONR Research Proposal in response to ONR BAA 01-012 Demonstration of Undersea Autonomous Operation Capabilities and Related Technology Development, August 2001. Healey, A. J., Kim, J., Control And Random Searching With Multiple Robots, Proceedings IEEE CDC Conference 2000, Sydney Australia, Nov. 2000 paper No INV-2303. Healey, A. J., Marco, D. B., Slow Speed Flight Control of Autonomous Underwater Vehicles: Experimental Results with NPS AUV II Proceedings of the 2nd International Offshore and Polar Engineering Conference, San Francisco, July 14-19 1992.

59

Healey, A.J. and David Lienard, Multivariable Sliding Mode Control for Autonomous Diving and Steering of Unmanned Underwater Vehicles, IEEE Journal of Oceanic Engineering, Vol. 18, pp 327-339, Jul 1993. Hoerner, Sighard. author, 1965. Fluid Dynamic Drag. Published by

Johnson, Jay, Parameter Identification of the ARIES AUV, M.S. Thesis Naval Postgraduate School, Monterey, CA, June 2001. Kamon, I. And Rivlin, E., Sensory-based motion planning with global proofs, IEEE Transaction on Robotics and Automation, Vol 13, no. 6, 1997. Krogh, B. H. and Thorpe, C. E., Integrated Path Planning and Dynamic Steering Control for Autonomous Vehicles, Proceedings of the 1986 IEEE International Conference on Robotics and Automation, Apr, 1986, pp. 1664-1669. Lane, David M., Petillot, Yvan, and Ruiz, Ioseba Tena, Underwater Vehicle Obstacle Avoidance and Path Planning Using a Multi-Beam Forward Looking Sonar, IEEE Journal of Oceanic Engineering, Vol. 26, Apr 2001. Latrobe, J. C., Robot Motion Planning, Kluwer Academic Publishers, Norwell, MA, 1991. Marco, D.B and A.J. Healey, Command, Control and Navigation Experimental Results With the NPS ARIES AUV, IEEE Journal of Oceanic Engineering Special Issue, 2001. Moitie, R. and Seube, N., Guidance Algorithms for UUVs Obstacle Avoidance Systems, OCEANS 2000, Brest, France. Prestero, Timothy, Verification of a Six-Degree of Freedom Simulation Model for the REMUS Autonomous Underwater Vehicle, M.S. Thesis Massachusetts Institute of Technology, Sep 2001. Ruiz, Ioseba Joaquin Tena, Enhanced Concurrent Mapping and Localization Using Forward-looking Sonar, PHD Thesis Heriot-Watt University, Sep 2001.

60

Stentz, Anthony, Optimal and Efficient Path Planning for Partially-Known Environments, Proceedings of the IEEE International Conference on Robotics and Automation (ICRA '94), Vol. 4, May, 1994, pp. 3310 - 3317. Stokey, Roger and Von Alt, Chris, REMUS Remote Environmental Measuring UnitS Operations and Maintenance Manual, Woods Hole Oceanographic Institution, Apr 2001. Thorpe, C. F., Path Relaxation: Path Planning for a Mobile Robot, Carnegie-Mellon University Robotics Institute, Mobile Robots Laboratory, Autonomous Mobile Robots, Annual Report, Sep 1986, pp. 12-19. Warren, C.W. A Technique for Autonomous Underwater Vehicle Route Planning, IEEE Journal Of Ocean Engineering, Vol. 15, no. 3, pp. 199-204, 1990.

61

THIS PAGE INTENTIONALLY LEFT BLANK

62

INITIAL DISTRIBUTION LIST

1. Defense Technical Information Center Ft. Belvoir, VA Dudley Knox Library Naval Postgraduate School Monterey, CA Mechanical Engineering Department Chairman, Code ME Naval Postgraduate School Monterey, CA Naval/Mechanical Engineering Curriculum Code 34 Naval Postgraduate School Monterey, CA Professor Anthony J. Healey, Code ME/HY Department of Mechanical Engineering Naval Postgraduate School Monterey, CA Dr. Donald Brutzman, Code UW/Br Undersea Warfare Group Naval Postgraduate School Monterey, CA Dr. T. Swean, Code 32OE Office of Naval Research Arlington, VA Doug Horner Naval Post Graduate School Monterey, CA L. C. Van Alt WhoodsHole Oceanographic Institute Boston, MA LT Lynn Fodrea Naval Surface Warfare Center, Dahlgren Division Dahlgren, VA

2.

3.

4.

5.

6.

7.

8.

9.

10.

63