A Hierarchical Path Planning Approach Based on Reinforcement Learning for Mobile Robots

摘要

In this paper, we propose a novel hierarchical path planning algorithm for mobile robots based on A~* and reinforcement learning (RL) with the structure of two layers. In the first layer, we adopt the A~* search algorithm to plan a geometric path and select several points as sub-target points for the planning of the next stage. In the second layer, a local path planning algorithm based on an approximate RL method called Least Square Policy Iteration (LSPI) is used to find a kinematically feasible path with these sub-targets. After learning, the local path planner in the second layer has good generalization performance. The path obtained by the proposed algorithm is smooth and safe for executing. Simulations have been carried out and the results demonstrate the validity of the proposed scheme.