Space-filling forest for multi-goal path planning

Abstract

In multi-goal path planning, the task is to find a sequence to visit a set of target locations in an environment. The combinatorial part of the problem (finding the sequence) can be solved as an instance of Traveling Salesman Problem, which requires knowledge about collision-free paths (and distances) between the individual targets. Finding the collision-free paths between the targets is essential in this task. Sampling-based planners like Probabilistic Roadmaps (PRM) and Rapidly-exploring Random Tree (RRT) can be used to find these paths. However, PRM can be computationally demanding, as it attempts to connect each node in the roadmap to its neighbors, regardless of their later usage in the solution. Contrary, RRT is a tree-based planner, and one run can only provide paths starting in the root of the tree (a single target). In this paper, we propose a novel planner for multi-goal path planning. Multiple trees (forest) are constructed simultaneously from the targets and expanded by collision-free configurations until they touch each other or obstacles. Each tree, therefore, does not explore the whole configuration space (as in the case of RRT), and its construction is faster than PRM, as it uses lower number of edges. The efficiency of this new planning approach is demonstrated in the multi-goal path planning in 2D environment with tens of targets and with narrow passages.