Rover visual obstacle avoidance

摘要: The Stanford AI Lab cart is a remotely controlled TV equipped mobile robot. A computer program has driven the through cluttered spaces, gaining its knowledge of world entirely from images broadcast by onboard system. The uses several kinds stereo to locate objects around it in 3D and deduce own motion. It plans an obstacle avoiding path desired destination on basis model built with this information. plan changes as perceives new obstacles journey. The system reliable for short runs, but slow. moves one meter every ten fifteen minutes, lurches. After rolling stops, takes some pictures thinks about them long time. Then path, executes little it. pauses again. It successfully 20 courses (each taking five hours) complex enough necessitate three or four swerves. Some weaknesses possible improvements were suggested these other, less successful, runs.