A reinforcement learning-assisted ensemble differential evolution approach for 3 D AUV path planning
Shubham Gupta, Shitu Singh, Vinay Kumar · Ships and Offshore Structures · 2025
Autonomous underwater vehicle (AUV) has become an important technology in the area of the Internet of Underwater Things (IoUT). However, the unpredictable nature and complexity of the underwater environment present significant challenges for their autonomous operation. To enable reliable and effective performance, AUVs require efficient and stable path-planning algorithms. To tackle these issues, this paper presents a new framework of the well-known optimization algorithm called differential evolution. The proposed algorithm is referred to as RL-MuDE, which is especially designed for 3D path planning. This algorithm incorporates several search strategies, including the multi-mutation learning, combined guidance from elite individuals, archive population, reinforcement learning, and historical success and statistical distribution guided parameter tuning. All these strategies are combined with the DE to balance the diversity and convergence speed features during the optimization procedure and to avoid the search being prone to sub-optimal solutions. The developed RL-MuDE is validated over nine different scenarios of the AUV path planning, where threat information is varied and the dimension of the associated optimization problem is increased. Experimental validation and comparison with other metaheuristics have verified the competitive search ability of the proposed RL-MuDE algorithm in efficiently providing the optimal AUV path planning without any collisions from threats.