Motion planning in semistructured environments with teaching roadmaps

Abstract

Motion planning is a hot topic in robotics, and the sampling-based algorithms have gained their popularities in research areas. However, these methods are still not suitable for real-world motion planning problems, because it is computationally expensive to completely explore the high-dimensional configuration space (C-space) of robots. Inspired by the related works on learning from demonstration, we propose a novel motion planning method named teaching roadmaps, which can take advantage of the optimal teaching data and quickly find a new path in the similar scenarios. The theoretical analysis and our experiments indicated that our approach is probabilistically complete, and it can find a feasible path faster than other sampling-based methods in similar environments.