基于近端策略优化的两栖无人平台路径规划算法研究

In order to solve the algorithm problem of the training speed and stability in local path planning of amphibious unmanned platform, a proximal policy optimization (PPO) algorithm was improved, establishing a foundation of multi-sensory information input for the amphibious platform. Actually, four perceptual information input schemes and speed-enhanced reward function were proposed to adapt to the dynamic and static environment. The experimental results show that the amphibious unmanned platform with BEV+V state-space input structure and discrete action space demonstrates high success rate and low timeout rate in path planning, which is superior to the traditional methods and other schemes. Simulation and comparative experiment results show that the state space data structure with the combination of aerial view and speed combined with the speed enhancement reward function algorithm can improve the algorithm performance, increasing convergence speed up to 25.58%, the success rate of path planning up to 25.54%, and descending the timeout rate by 13.73%.