Trajectory planning for 4WIS autonomous ground vehicles in RORO terminals using CBF-enhanced deep reinforcement learning

In the operational environment of Roll-On/Roll-Off (RORO) terminals, intelligent autonomous ground vehicle (AGV) systems, which integrate perception and motion planning modules, are widely used for the transportation of vehicles and cargo. The motion planning module plays a crucial role in improving operational efficiency and reducing energy consumption. To meet the stringent requirements for trajectory smoothness, kinematic feasibility, and safety in transfer tasks, we propose a novel real-time trajectory planning framework. This framework integrates control barrier functions with proximal policy optimization, enabling direct control command outputs without the need for separate tracking modules typically required in traditional planning methods. We design a composite multi-objective reward function to optimize key performance indicators, including time efficiency, trajectory smoothness, and collision safety, while adhering to the AGV's kinematic constraints. Simulation experiments and physical tests validate the effectiveness of the proposed framework, with results demonstrating significant improvements in motion performance and operational efficiency. The method also exhibits real-time applicability, with an average computation time of less than 3 ms per control step, making it suitable for practical implementation in RORO terminals.