اجتناب از موانع درک عمق تقویت عمیق برای کشتی های دریایی بدون سرنشین

This research is concerned with the problem of obstacle avoidance for the underactuated unmanned marine vessel under unknown environmental disturbance. A concise deep reinforcement learning obstacle avoidance (CDRLOA) algorithm is proposed with the powerful deep Q-networks architecture to overcome the usability issue caused by the complicated control law in the traditional analytic approach. Furthermore, a comprehensive reward function is specifically designed for obstacle avoidance, target approaching, speed modification, and attitude correction. Compared to the analytic methods, the proposed algorithm based on reinforcement learning shows notable advantages in utility and extendibility. With the same CDRLOA system, the targets and the constraints are highly customizable for various of special requirements. Extensive experiments conducted have demonstrated the effectiveness and conciseness of the proposed algorithm.