Obstacle Based Fast Marching Tree for Global Motion Planning

In this paper we introduce a novel method to improve the general performance of fast marching tree for collision-free path planning called obstacle-based fast marching tree. Our proposed OB-FMT* samples the map in discussion initially to obtain the vertexes of all obstacles and connect those vertexes to build a collision-free roadmap, an A*algorithm is adopted to find an optimal path in the built roadmap. Then multiple prolate hyper spheroids with focuses on vertexes of optimal path are established to limit the sampling range. Within the limited sampling point set, we utilize FMT* to find optimal solution with high efficiency. Theoretical analyses are given to prove the probabilistic completeness and computational complexity of our proposed method. Experiments in multiple scenarios show that our algorithm has a better performance comparing with other existing algorithms including RRT*, FMT*, Informed RRT*.