Deep Reinforcement Learning Based Path Planning for Mobile Robots Using Time-Sensitive Reward

摘要

In the mobile robots’ field, the global path planning task in known map scenarios is an urgent problem to be solved. Deep Reinforcement Learning (DRL), an efficient decision-making method, has been widely used to solve path-planning problems. Nonetheless, as the map size increases, the existing DRL algorithms are prone to the problem of sparse rewards. The above drawback makes the mobile robot converge slowly on the map. Even in extreme cases such as trap maps, the robot obtains the optimal convergent solution differently. For this purpose, this paper encodes the various node information in the map separately as a state description of the environment. To efficiently perform path-planning tasks for mobile robots in various complex scenarios, a time-sensitive reward function based on DRL is presented. The simulation experiments on a variety of complex environmental maps are conducted. The experimental results demonstrate the effectiveness of the proposed method. Our method ensures that the DRL algorithm is able to converges to a feasible solution quickly.