文献类型：会议论文

英文题名：Lane following method based on Q-PID algorithm

作者：Li, Jiahong[1,2]; Yao, Yongqiang[1,3]; Ma, Nan[4]

第一作者：Li, Jiahong;李佳洪

机构：[1] College of Robotics, Beijing Union University, Beijing, 100101, China; [2] Vrije Universiteit Brussel, Artificial Intelligence Lab, Brussels, 1050, Belgium; [3] Beijing Union University, Beijing Key Laboratory of Information Service Engineering, Beijing, 100101, China; [4] Beijing University of Technology, Faculty of Information Technology, Beijing, 100124, China

第一机构：北京联合大学机器人学院

通讯机构：[4]Beijing University of Technology, Faculty of Information Technology, Beijing, 100124, China

会议论文集：Proceedings - 2022 18th International Conference on Computational Intelligence and Security, CIS 2022

会议日期：December 16, 2022 - December 18, 2022

会议地点：Chengdu, China

语种：英文

外文关键词：Learning algorithms - Proportional control systems - Reinforcement learning - Two term control systems

摘要：The PID (Proportional-Integral-Derivative) controller has been broadly applied in many control engineering tasks due to its simplicity and fast computation as the model-free low-level control strategy. However, it still suffers from instability due to its feedback mechanism, especially in the complex self-driving driving task, e.g., the lane following task. Traditional approaches to this problem include classical PID tuning and expert system-based PID tuning methods but are not suitable due to the low sample efficiency in the uncertain environment which is not fully known. In this paper, we proposed Q learning-based PID (Q-PID) algorithm to solve the problem. In the algorithm, the policy of the optimal parameters of PID are learned via incremental exploration-exploitation procedure, i.e., learn the approximated Q-value function with an experience replay mechanism and calculate the optimal policy by maximizing the Q-function. The simulation results in the lane following task demonstrate the feasibility of the proposed algorithm. ? 2022 IEEE.