Trajectory tracking control for Ackerman vehicle based on improved reward function

This article focuses on an online reinforcement learning algorithm (ORRL) for trajectory tracking control of a class of nonlinear systems. Based on the Q-learning framework, the trajectory tracking problem can be solved under the complex paths by designing an optimal reward function integrating mult...

Full description

Saved in:
Bibliographic Details
Published in2024 43rd Chinese Control Conference (CCC) pp. 519 - 524
Main Authors Xie, Hui, Ma, Xiaoyu, Qin, Qiuyue, Sun, Xingjian
Format Conference Proceeding
LanguageEnglish
Published Technical Committee on Control Theory, Chinese Association of Automation 28.07.2024
Subjects
Accuracy
Decision making
Prediction algorithms
Process control
Q-learning
Reinforcement learning
reward functions
Trajectory
Trajectory tracking
trajectory tracking control
Online AccessGet full text

Cover

More Information
Summary:This article focuses on an online reinforcement learning algorithm (ORRL) for trajectory tracking control of a class of nonlinear systems. Based on the Q-learning framework, the trajectory tracking problem can be solved under the complex paths by designing an optimal reward function integrating multiple sub objectives. The number of iterations and the complexity of the algorithm are reduced by the multi-objective reward mechanism. At the same time, the information of sub-objectives is fully considered to provide richer decision-making information, such that the control process is smoother. Moreover, the short-term action decision and long-term control effect can be balanced by the ORRL. The real vehicle results are given to compared the MPC, Stanley, LQR, standard reinforcement learning algorithm and the ORRL method, and show the higher trajectory tracking accuracy, smoothness in complex trajectory scenes and better adaptive ability.
ISSN:1934-1768
DOI:10.23919/CCC63176.2024.10662434