Dynamic environmental factors such as ocean currents affect the path-planning capabilities of AUVs to some extent, leading to the traditional path-planning technology is difficulty in enabling the AUVs to complete the navigation task. To design the optimal path for AUVs that can guarantee navigation safety, shorten the navigation time, and reduce the energy consumption of AUVs under a complex underwater environment, reinforcement learning is proposed to improve the dynamic path planning method of AUVs. The distributed value output to enhance the Double Deep Q-Network algorithm is used to obtain more information and increase the accuracy and stability of the algorithm. The improved algorithm is obtained to utilize the direction of the currents to help the glide smoothen AUV and reduce energy consumption, thus better accomplishing dynamic path planning. Finally, a complex marine environment is introduced to conduct simulation experiments on the improved dynamic path planning algorithm to verify the feasibility of the dynamic path planning system.