The invention belongs to the technical field of UUV control, and specifically relates to a UUV real-time collision avoidance planning method based on deep double-Q network reinforcement learning. Theinvention provides an autonomous collision avoidance planning method suitable for the working environment and the perception characteristics of a UUV (unmanned Underwater vehicle). A strategy of the UUV is continuously improved by performing continuous trial and error interaction between the UUV and the environment and generating a reward or punishment signal by using successful or failed experience, so that the UUV has the self-learning capability. Through the method disclosed by the invention, a network system has self-learning capability in the local collision avoidance planning of the complex environment to realize an end-to-end model, and the deep learning and the reinforcement learning are combined to be applied to the solution of the collision avoidance planning problem by directlylearning a mapping relation of a state and the action from an original data set. By using the deep reinforcement learning, the condition that the strategy cannot be executed due to complex path can beavoided, the development period of the project is shortened in the actual application, the implementation is more concise, efficient, and high in robustness.