This research has been performed by Liana Lorigo, advised by Professors Rodney A. Brooks and W. Eric L. Grimson. Funding is provided by Jet Propulsion Laboratories and an NSF graduate fellowship.
The current objective is two-fold. The direct application is a system suitable for a Mars Rover robot, which would move autonomously upon the rocky, crater-full Martian surface, gathering soil and atmospheric samples and images. At its mission's completion, the Rover would return to its lander/spacecraft for its journey back to Earth. The general computational goal is to push the limits of existing vision technology, as applied to this problem, thus determining the modifications, assumptions, and simplifications necessary for a complete visually-guided robotic system in this real-time, real-world environment.
A
Modular Obstacle Avoidance System (click for algorithm
description)
This work addresses the problem of designing a mobile robot to avoid
obstacles while traveling in unstructured environments, that is,
environments for which no knowledge of the appearance of the ground or
the locations or appearance of the obstacles is available. An
autonomous obstacle avoidance system has been designed and
implemented. With this system, Pebbles can travel around various
cluttered environments safely. The only sensor used is a single
uncalibrated camera at the front of the robot.
Environments
Image of Mars surface from Viking 2 lander.
One goal of this research is the development of a robotic
system to explore the Martian surface by using visual cues to avoid
rocks and craters too large to traverse. Some of the settings in
which the system was tested -- rough outdoor areas and an indoor
room of rocks and gravel -- were motivated by this surface. The
system was also tested in several other indoor settings, such as
lounges, corridors, and offices.
Platform
The system is housed in the Pebbles III Robot, designed and built
by IS Robotics, Inc.
Pebbles is equipped with a Motorola 68332 processor and a Chinon 3mm
camera positioned at the front of the robot 10.5 inches off the
ground. Furthermore, the system uses a visual-processing
hardware system, the CVM (Cheap Vision
Machine), designed by Ian Horswill and Chris Barnhart. The processor is a Texas
Instruments C30 DSP. The vision software is written in C and runs
on the CVM. The system runs a smaller control program, written in L
(a subset of LISP written by Professor Brooks) on the 68332.