Verification of Automated Parking of Semi-trailer using Reinforcement Learning

Abstract

Since the semi-trailer has a hitch point, the tractor pushes the trailer from the rear when backing up in the coupled state. And when the semi-trailer needs to turn, it is necessary to turn the steering wheel in the opposite direction of the desired turning direction. Therefore, the operation is extremely difficult and requires more complicated processing than that of an ordinary heavy-duty vehicle. Semitrailers, which require complex operations, have a high affinity with driver assistance systems such as automated parking systems. In this study, the reinforcement learning is applied to automated reverse parking of a semi-trailer. Reinforcement learning does not require training data and has the advantage of automatically finding the optimal action. A model of a semi-trailer is created in a simulation environment, and investigated whether the semi-trailer can acquire the ability to park backward using reinforcement learning. Sensors are placed at the center and four corners of the trailer and tractor, and rewards are set depending on whether the sensors are within the parking space or not. As a result, an automated reverse parking model of a semi-trailer was successfully built in the simulation.