Randomised Rough-Terrain Robot Motion Planning

Abstract

A fundamental property of rough-terrain motion planning is that each configuration of the robot on the terrain can be assigned a characteristic navigational difficulty based on the terrain topography and physical properties of the underground. This measure is computed in an application-dependent manner based on the properties of the robot and terrain model employed. In this work we propose a motion planner based on rapidly exploring random trees (RRTs) which takes into consideration the characteristics of the underground. In the suggested solution, the randomised expansion of the RRTs is biased towards regions of low navigational difficulty. The motion planner generates trajectories which follow areas of easy navigation and only deviates to harder regions where inevitable. In particular, the maximal difficulty on the path is approximately minimised. For single-query motion planning tasks, bidirectional RRTs have proven to be effective in rapidly computing a path between the initial and goal configurations. To deal with complex distributions of the terrain characteristics, this concept has been extended. Rooted at randomly chosen configurations, a number of additional RRTs is grown. While the algorithm discussed is intended for rough-terrain motion planning and demonstrated in this context, it can easily be adapted to other domains where a characteristic measure for the desirability of attaining individual configurations can be defined