AUV path planning based on improved IFDS and deep reinforcement learning

Abstract

Existing autonomous underwater vehicle (AUV) path planning algorithms are rapidly developing and perform well in solving optimal paths. However, the performance of these algorithms in real environments is significantly worse than that in simulated environments due to the influence of currents in real marine environments. To this end, this paper proposes an algorithm that improves the fusion of perturbed flow field and deep reinforcement learning and adds the influence of random currents to the environment, which further improves the overall accuracy of AUV obstacle avoidance in dynamic environments and enhances the AUV's adaptability to the real environment. This study also compares the results obtained using four fused deep reinforcement learning algorithms simulated in different scenarios, and the results show that the proposed algorithm can enable AUV to realize dynamic path planning in unknown environments.