基于PBRS-SAC算法的无人车路径规划研究

针对复杂环境下无人车路径规划问题，在软演员评论家(SAC)算法的框架下进行改进。通过在奖励函数的设计上融合基于势能的回报塑形（PBRS）思想，并加入双连帧等训练技巧,设计了PBRS-SAC算法。之后在Ubuntu操作系统上搭建基于Gazebo的仿真环境，分别模拟静态与动态实验环境进行训练。最后，通过消融实验、敏感性测试实验与鲁棒性分析实验验证该算法的有效性。

Study on Path Planning for Unmanned Vehicles Using PBRS-SAC Algorithm

Aiming at the path planning problem of unmanned vehicles in complex environments, an improvement was made within the framework of the soft actor-critic algorithm. The PBRS-SAC algorithm was designed by incorporating the idea of potential-based reward shaping into the reward function design, and by integrating training techniques such as double concatenation frames. Subsequently, a simulation environment based on Gazebo was constructed on the Ubuntu operating system to simulate static and dynamic experimental environments for training. Finally, the effectiveness of the algorithm was assessed through ablation experiments, sensitivity testing experiments, and conducted robustness analysis experiments.