Toward Optimal Configuration Space Sampling

Efficient motion planning is obtained by focusing computation on relevant regions of configuration space. In t he following we propose a new approach to multi-query samplingbased motion planning, which exploits information obtained from earlier exploration and its current state to guide exploration. This approach attempts to minimize the selection of samples to th ose required to completely capture configuration space connect ivity. Our planner constructs an approximate model of configuration space that is used in conjunction with a utility function to select configurations with maximal expected importance giv en the planner’s current state. The resulting utility-guided planner is online and adaptive. Its behavior adjusts to the developi ng state of the motion planner and its understanding of the confi guration space. Experimental comparisons with existing planners show that this utility-guided approach significantly decreases the runtime required for motion planning.