An Information-Interdependence Deep Reinforcement Learning Path Planning Scheme for AUV With Ocean Currents Utilization

Autonomous underwater vehicle (AUV) possesses great potential in underwater applications, and path planning, as the fundamental technology, is a guarantee and prerequisite for stable and reliable performance. Despite extensive research in this area, certain limitations persist. For instance, accurately modeling the ocean environment remains challenging, and the reliability of simulations based on numerical models is often questioned. Furthermore, existing algorithms may lack a comprehensive understanding and effective utilization of the environment, leaving room for further improvement. To break through the current limitations, we propose an Information-Interdependent Soft actor-critic path planning method (<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$I^{2}S$</tex-math></inline-formula>). Firstly, the ocean model generated by the double Gaussian function is combined with the real terrain, which ensures accuracy while reducing the computational cost and comprehensively improving the credibility of the simulation environment. Secondly, the information-interdependent module is designed, which integrates multi-source data integration and feature encoding to extract key ocean states. This enhances the AUV's ability to perceive and adapt to dynamic environmental changes. Finally, the elaborate dynamic composite reward function integrates task completion, energy consumption level, and motion efficiency, which can guide the AUV to intelligently utilize the ocean currents as well as effectively shorten the training period.