Safe Navigation of Autonomous Underwater Vehicles Using Physics-informed Neural Networks

This research emphasizes the critical role of effective collision avoidance techniques for ensuring the safe operation of Autonomous Underwater Vehicles (AUVs), specifically focusing on integrating obstacle avoidance features into motion planning under different underwater scenarios. The paper focuses on the integration of obstacle avoidance features into motion planning for Autonomous Underwater Vehicles (AUVs) using Control Barrier Function (CBF). The Quadratic Programming (QP) problem with CBF as a constraint is converted to the Hamilton-Jacobi-Bellman (HJB) equation and solved by a Physics-informed Neural Network (PINN), a popular PDE approximator. The novelty of the work lies in proposing an innovative approach combining CBF's safety aspects with PINN for avoiding multiple obstacles, particularly in the depth control maneuvers of AUVs, reducing computational intensity. Depth control trajectory tracking is achieved through switching controls between Proportional-Integral-Derivative (PID) control and underwater sense-and-avoidance PINN (USAAPINN), ensuring safe navigation in the presence of obstacles. Numerical simulations validate the efficacy of the proposed PINN framework.