The autonomous navigation and obstacle avoidance for USVs with ANOA deep reinforcement learning method

摘要

The unmanned surface vehicle (USV) has been widely used to accomplish missions in the sea or dangerous marine areas for ships with sailors, which greatly expands protective capability and detection range. When USVs perform various missions in sophisticated marine environment, autonomous navigation and obstacle avoidance will be necessary and essential. However, there are few effective navigation methods with real-time path planning and obstacle avoidance in dynamic environment. With tailored design of state and action spaces and a dueling deep Q-network, a deep reinforcement learning method ANOA (Autonomous Navigation and Obstacle Avoidance) is proposed for the autonomous navigation and obstacle avoidance of USVs. Experimental results demonstrate that ANOA outperforms deep Q-network (DQN) and Deep Sarsa in the efficiency of exploration and the speed of convergence not only in static environment but also in dynamic environment. Furthermore, the ANOA is integrated with the real control model of a USV moving in surge, sway and yaw and it achieves a higher success rate than Recast navigation method in dynamic environment.