A RL-Based MPC Algorithm for AUV Trajectory Tracking

In underwater environments, Autonomous Underwater Vehicles (AUVs) face numerous challenges in executing trajectory tracking tasks due to complex water currents, uncertain obstacle interference, and irregular seabed topography. Existing control methods often rely on manually set weight parameters, which lack sufficient flexibility and robustness, making it difficult to maintain high performance in complex underwater environments. This paper proposes an algorithm that combines Model Predictive Control (MPC) with Control Barrier Functions (CBF), using Reinforcement Learning (RL) to dynamically adjust the weight parameters, thereby achieving trajectory tracking and autonomous obstacle avoidance for AUVs in complex underwater environments. Firstly, CBF is integrated into the objective function of MPC to ensure effective obstacle avoidance while performing trajectory tracking tasks. Secondly, RL is introduced to gradually optimize the MPC weight parameters, improving the system's adaptability and tracing performance in complex underwater environments. Simulation results show that the RL-based MPC-CBF algorithm demonstrates significant performance advantages in AUV trajectory tracking and obstacle avoidance tasks.