Next -->

Obstacle Avoidance and Navigation in the Real World by a Seeing Robot Rover

Hans Moravec
March 1980
Computer Science Department
Stanford University
(Ph.D. thesis)

Preface

The Stanford AI lab cart is a card-table sized mobile robot controlled remotely through a radio link, and equipped with a TV camera and transmitter. A computer has been programmed to drive the cart through cluttered indoor and outdoor spaces, gaining its knowledge about the world entirely from images broadcast by the onboard TV system.

The cart deduces the three dimensional location of objects around it, and its own motion among them, by noting their apparent relative shifts in successive images obtained from the moving TV camera. It maintains a model of the location of the ground, and registers objects it has seen as potential obstacles if they are sufficiently above the surface, but not too high. It plans a path to a user-specified destination which avoids these obstructions. This plan is changed as the moving cart perceives new obstacles on its journey.

The system is moderately reliable, but very slow. The cart moves about one meter every ten to fifteen minutes, in lurches. After rolling a meter, it stops, takes some pictures and thinks about them for a long time. Then it plans a new path, and executes a little of it, and pauses again.

The program has successfully driven the cart through several 20 meter indoor courses (each taking about five hours) complex enough to necessitate three or four avoiding swerves. A less sucessful outdoor run, in which the cart swerved around two obstacles but collided with a third, was also done. Harsh lighting (very bright surfaces next to very dark shadows) resulting in poor pictures, and movement of shadows during the cart's creeping progress, were major reasons for the poorer outdoor performance. These obstacle runs have been filmed (minus the very dull pauses).

Hans Moravec
March 2, 1980


Table of Contents

Chapter 1:   Introduction
Chapter 2:   History
Chapter 3:   Overview
Chapter 4:   Calibration
Chapter 5:   Interest Operator
Chapter 6:   Correlation
Chapter 7:   Stereo
Chapter 8:   Path Planning
Chapter 9:   Evaluation
Chapter 10:   Spinoffs
Chapter 11:   Future Carts
Chapter 12:   Connections

Appendix 1:   Introduction
Appendix 2:   History
Appendix 3:   Overview
Appendix 6:   Correlation
Appendix 7:   Stereo
Appendix 8:   Path Planning
Appendix 10:   Spinoffs
Appendix 12:   Connections


Acknowledgements

My nine year stay at the Stanford AI lab has been pleasant, but long enough to tax my memory. I hope not too many people have been forgotten.

Rod Brooks helped with most aspects of this work during the last two years and especially during the grueling final weeks before the lab move in 1979. Without his help my PhD-hood might have taken ten years.

Vic Scheinman has been a patron saint of the cart project since well before my involvement. Over the years he has provided untold many motor and sensor assemblies, and general mechanical expertise whenever requested. His latest contribution was the camera slider assembly which is the backbone of the cart's vision.

Don Gennery provided essential statistical geometry routines, and many useful discussions.

Mike Farmwald wrote several key routines in the display and vision software packages used by the obstacle avoider, and helped construct some of the physical environment which made cart operations pleasant.

Jeff Rubin pleasantly helped with the electronic design of the radio control link and other major components.

Marvin Horton provided support and an array of camera equiment, including an impressive home built ten meter hydraulic movie crane for the filming of the final cart runs.

Others who have helped recently are Harlyn Baker, Peter Blicher, Dick Gabriel, Bill Gosper, Elaine Kant, Mark LeBrun, Robert Maas, Allan Miller, Lynne Toribara and Polle Zellweger.

My debts in the farther past are many, and my recollection is sporadic. I remember particularly the difficult time reconstructing the cart's TV transmitter. Bruce Bullock, Tom Gafford, Ed McGuire and Lynn Quam made it somewhat less traumatic.

Delving even farther, I wish to thank Bruce Baumgart for radiating a pleasantly (and constructively) wild eyed attitude about this line of work, and Rod Schmidt, whom I have never met, for building the hardware that made my first five years of cart work possible.

In addition I owe very much to the unrestrictive atmosphere created at the lab mainly by John McCarthy and Les Earnest, and maintained by Tom Binford, and also to the excellent system support provided to me (over the years) by Marty Frost, Ralph Gorin, Ted Panofsky and Robert Poor.

Hans Moravec, 1980

Next -->