Kang, Qi (Continental Automotive Singapore), Hartmannsgruber, Andreas (Continental Automotive Singapore), Tan, Sze Hui (Continental Automotive Singapore), Zhang, Xian (Continental Autonomous Mobility US, LLC), Chew, Chee-Meng (National University of Singapore)

Deep Reinforcement Learning Based Tractor-Trailer Tracking Control

Scheduled for presentation during the Regular Session "Control of heavy vehicles" (FrAT4), Friday, September 27, 2024, 11:50−12:10, Salon 7

2024 IEEE 27th International Conference on Intelligent Transportation Systems (ITSC), September 24- 27, 2024, Edmonton, Canada

This information is tentative and subject to change. Compiled on December 26, 2024

Abstract

Abstract—This paper introduces a tracking control method for an autonomous tractor-trailer system using policy-based deep reinforcement learning. Approaches for planning and controlling traditional passenger vehicles cannot be easily transferred to this special system because its physical properties, like size and kinematics, are much more complex. To manage this complexity, we make use of a new and trending artificial intelligence tool: deep reinforcement learning to control the tracking of autonomous tractor-trailer systems by providing guidance trajectory information. The proximal policy optimization is adopted for training the agent with a reward function defined to penalize deviation and reward following a given track. We chose a B-spline-based lane representation to improve the observation of the agent’s state in the environment and the diversity of the training scenarios. The policy is trained and validated in the Pybullet simulator and our trained agent achieves good capability in navigating random and complex tracks.