Real-Time Path Planning for Multiple Autonomous Underwater Vehicles in Dense Obstacle Fields: The DA-IFDS Approach

Abstract

Multiple autonomous underwater vehicles (AUVs) face dense static and dynamic obstacle threats during path planning in complex and dynamic subsea environments. Therefore, each AUV must achieve safe, reliable, and real-time path planning under limited local observations. This article proposes a 3-D path planning method for multiple AUVs, named the dynamically adaptive-interfered fluid dynamical system (DA-IFDS) algorithm. It employs a two-layer real-time decision-making framework: the high-level decision-making module outputs step-by-step parameters based on proximal policy optimization (PPO), while the low-level module determines the velocity direction of each step according to these parameters. Such a hierarchical design improves the learning efficiency and adaptability of multiple AUVs. This algorithm models obstacles as disturbances in the flow field and incorporates these parameters into the action space. It employs bidirectional gated recurrent units to process variable-length observation sequences, effectively addressing the observation inconsistency faced by traditional methods in dynamic obstacle scenarios. In addition, it designs a diversified reward function. Simulation results demonstrate that the proposed DA-IFDS algorithm outperforms soft actor-critic-IFDS, deep deterministic policy gradient-IFDS, and conventional PPO algorithms in terms of success rate, obstacle-avoidance performance, and planning efficiency, validating its effectiveness and superiority in real-time path planning for multiple AUVs in dense static and dynamic obstacle scenarios.