Dynamic Motion Planning for Efficient Visual Navigation under Uncertainty

This paper deals with a planning problem for a mobile robot to reach a goal quickly and safely under uncertainty of vision and motion. If a robot moves to the goal using only dead reckoning, the positional error may increase by the motion uncertainty. The error can be reduced by visual information. But the vision also includes uncertainty caused by the pixel quantization. It also takes considerable computational time(cost). The robot can reach the goal quickly without stopping using the continuous motion during visual processing, and with minimum observation count. By the speed control, the robot moves to each viewpoint. The viewpoint is adaptively determined according to both the predicted positional uncertainty and the con guration of obstacles. The planner works online to cope with the actual error. An experimental result using a mobile robot with stereo vision shows the validity of the proposed method.