2D LiDAR based Reinforcement Learning for Multi-target Path Planning in Unknown Environment

Global path planning techniques have been widely employed in solving problems, however they found to be unsuitable for unknown environments. Contrarily, the traditional Q-learning method, which is a common reinforcement learning approach local planning, unable complete task multiple targets. To address these limitations, this paper proposes modified called Vector Field Histogram based (VFH-QL) utilized VFH information state space representation and reward function, on 2D LiDAR sensor. We compared performance of our proposed method with classical (CQL) through training experiments that were conducted simulated environment size 400 square pixels, representing 20-meter map. The contained static obstacles single mobile robot. Two conducted: experiment A involved target, while B results showed VFH-QL had 87.06% less time 99.98% better obstacle avoidance CQL. In B, was an average 95.69% than CQL 83.99% quality. then evaluated using benchmark dataset. indicated exhibited superior quality, efficiency 94.89% improvements 96.91% 96.69% over SARSA targets