AUV Path Planning Under Ocean Current Disturbances Based on Deep Reinforcement Learning

To address three-dimensional path planning for autonomous underwater vehicles in marine environments with seabed, obstacles, and ocean-current disturbances, this paper proposes an improved deep reinforcement learning method termed NPD-D3QN. A three-dimensional marine environment is constructed using a bathymetric digital elevation model, spherical obstacles, and real ocean-current data with a first-order GaussMarkov disturbance term. The path-planning problem is formulated as a Markov decision process with a designed state space, discrete action space, and reward function. Based on D3QN, NoisyNet, prioritized experience replay, and a distillation replay pool are introduced to enhance exploration, sample efficiency, and training stability. Simulation results show that the proposed method achieves faster convergence, higher success rates, and lower overall cost than the comparison methods. Dynamicobstacle experiments further demonstrate its adaptability and response capability in time-varying environments. Therefore, the proposed method provides an effective solution for threedimensional AUV path planning in complex marine environments.