Abstract: In order to improve the safety of unmanned surface vehicles navigation, an intelligent obstacle avoidance algorithm D_TD3 based on the improved twin delayed deep deterministic policy gradient (TD3) is proposed. A hybrid risk assessment model was proposed, and a dynamic ship domain was constructed using unmanned ship motion parameters and maritime collision avoidance rules, and the collision risk degree was calculated hierarchically. According to the collision risk degree, collision avoidance rules, etc., a guiding reward function was designed. A priority sampling method with mixed experience pool was proposed to improve the efficiency of the algorithm. Experimental results show that the D_TD3 algorithm can effectively realize the navigation mission, the success rate reaches more than 85%, the average navigation time is shortened, and the robustness of navigation is significantly improved.