Water Surface Autonomus Navigation Simulation via Multi-agent Proximal Policy Optimization

To address the challenges of autonomous navigation and intelligent obstacle avoidance for clusters of homogeneous unmanned surface vehicles (USVs) in intricate environments, a self-driving algorithm utilizing multi-agent reinforcement learning (MARL) is introduced. A multi-obstacle scenario is created, where the interactions among unmanned underwater vehicles are used to assess the game situation. The USV fleet undergoes training through the multi-agent proximal policy optimization algorithm (MAPPO) to facilitate autonomous driving. This approach allows for efficient route planning and effective obstacle evasion, ensuring that the vehicles reach their destination with the desired orientation and relative positioning. The simulation environment is constructed with the Unity game engine, providing visualization during training to enhance algorithm robustness. Experimental results demonstrate that both the proposed algorithm and simulation setup exhibit strong feasibility and good performance.