Escaping Depressions in LRTS with Wall Following Method
Yue Hu, Qi Zhang, Long Qin, Quanjun Yin · 2017
In fields like robotics, methods of real-time search like LRTS (Learning Real-Time Search) are extensively used to undertake motion planning. Nevertheless, due to limited local information, LRTS suffers much from the irrational scrubbing behaviors because of slow learning and state re-visitation in heuristic depressions. To make agents behave more reasonable, we in this paper firstly propose a hybrid algorithm of LRTS and wall following approach, which has ever been used for assisting agents controlled by APF (Artificial Potential Field) method to escape local minima. We apply proper switching conditions between the behavior patterns and ease agents from visually odd thrashing behavior. Experiments show that our approach involves much less planning amount and substantially reduce state re-visitation in comparison with pure LRTS.