We present an implemented sensor-based planner for motion planning and exploration for eye-in-hand systems. A model-based motion planner is used to plan paths within the known part of the environment to further sense the unknown part of the environment. Each sensing action is viewed as gaining information about the status of configuration space. We introduce the notion of C-space entropy as a measure of ignorance or lack of information of C-space. The next view is planned so as to maximize expected entropy reduction (MER), or equivalently, expected information increase. Experimental results demonstrate that MER criterion results in efficient exploration of unknown environments and that the planner can make a robot arm move around safely (without collisions) while carrying out exploratory and purposive tasks in unknown environments.
展开▼
