Recurrent Proximal Policy Optimization Based Tractor-Trailer Wheeled Robot Automatic Parking Algorithm

摘要

Truck-trailer reverse parking poses significant challenges due to the system's inherent instability, complex road geometry, and collision avoidance requirement. Traditional approaches for trailer control rely on manually designed control policies, which may have limited applicability and scalability. This paper presents a novel approach for autonomous reverse parking of a tractor-trailer wheeled robot (TTWR) system in tight and complex environments. Utilizing the Proximal Policy Optimization (PPO) algorithm, the controller is trained, with a long short-term memory (LSTM) network enhancing the handling of sequential observation data. To improve safety and reliability, ultrasonic sensors are installed on the trailer to detect nearby obstacles and ensure safety. Furthermore, a novel reward function is introduced that encourages the TTWR to maintain a safe distance from surrounding obstacles while minimizing the parking trajectory distance and the steering maneuver. The effectiveness of the proposed LSTM-PPO algorithm is compared with the original PPO and other trending reinforcement learning algorithms. The results demonstrate the improved convergence stability and speed of the proposed approach, as well as its successful execution of end-to-end reverse parking maneuvers in unknown environments.