An Improved Fuzzy Actor-critic Learning Approach for Path Planning with Position Constraints

In the last decades, path planning with position constraints attracts many attentions. In this paper, we propose an innovative approach named improved fuzzy actor-critic learning (IFACL) to solve this problem without modelling the map containing obstacles and complex calculation. Specifically, only the initial position, target position and obstacle position are needed as inputs for the algorithm to learn a desired path. Based on FACL, a penalty factor is added to enable agents obtaining the ability to avoid obstacles through punishing agents when exceeding position constraints. Then, to optimize the path planned, an excessive coordinate point which can be updated iteratively during the training process is utilized to calculate the reward with penalty factor. The simulation results prove the superiority and effectiveness of this algorithm in different scenarios with regular obstacles and hypothetical irregular obstacles. Due to the flexibility of FACL, this approach may be easily extended to path planning with velocity constraints and dynamic constraints.