Utility-Guided Random Trees

Random trees planning has been used to solve connectivity problems in a variety of problem domains. In particular, these approaches are some of the most successful approaches to robotic motion planning. In this work we examine the broad array of random-tree that have been proposed for motion planning. From this survey we distill a general algorithmic framework for random-tree planning that abstracts the shared functionality used by all planners. Examination of this algorithm shows the parts of random-tree planning that have been explored in depth and those that have received little attention. Using this analysis, we show that two functions: exploration length and connecting growth have been largely unexplored. To address this, we propose utility-guided implementations of these functions. These functions are incorporated into a complete utility-guided random-tree motion planner. Experimental results with this planner in challenging artificial and real world planning problems indicate that this approach is significantly more efficient and reliable.