Path Planning Algorithms for Mobile Robots Based on Deep Reinforcement Learning

摘要

This paper gives an introduction of path planning algorithms for mobile robots based on deep reinforcement learning (DRL). Firstly, the traditional path planning algorithms are compared with the deep reinforcement learning path planning algorithms. One key advantage of DRL-based algorithms is their ability to meet real-time interactions due to lower calculation requirements. Then, some basic path planning algorithms based on deep learning used in unknown and dynamic environment are introduced. After that, the efforts on the improvement of the basic DRL-based path planning algorithms are discussed. Their training results are compared with the traditional path planning algorithms and the DRL-based path planning algorithms without improvements. Finally, the advantages and disadvantages of DRL-based path planning algorithms are analyzed. Although these algorithms have already a relative fast convergence speed and high success rate, they still do not meet the accuracy requirement for sophisticated work. Besides, the train period is also expected to be shorten further. The potential improvements in the future are also pointed.