Autonomous pathfinding for underactuated AUVs using FDHNN.

Autonomous Underwater Vehicles (AUVs) are increasingly pivotal in exploring and mapping environments, due to their autonomous pathfinding capabilities. Despite the development of various algorithms for this task, many do not adequately address the challenges posed by underactuation-a common characteristic of AUVs. Moreover, the constrained field of view inherent to sonar and visual sensors introduces additional directional limitations that complicate the detection process. To overcome these obstacles, we introduce an innovative trajectory planning method known as the Finite Direction Hopfield Neural Network (FDHNN). Our method employs eight directional neurons per grid cell with customized neural connection weights to create a path that caters to underactuated AUVs, ensuring adaptability in various aquatic environments and adherence to initial and terminal directional constraints. Complementing the global planning provided by FDHNN, we have refined the Dynamic Window Approach (DWA) for local maneuvering within the velocity domain. Simulation-based validation has underscored its robustness and superior performance, particularly when benchmarked against advanced algorithms such as dual deep Q-learning and Dubins paths. Our proposed trajectory planning method markedly improves the reliability of underactuated AUVs navigating with start and end directional limitations, offering substantial advancements in underwater detection and exploration.