Scholarlink

<p>In order to simplify the trajectory tracking controller for distributed drive electric vehicles and enhance its overall performance, this study introduces a machine learning approach rooted in the deep deterministic policy gradient algorithm. The proposed control strategy concurrently addresses three key performance objectives: lateral position, longitudinal velocity, and lateral stability. The control framework adopts a centralized structure to directly govern the steering wheel angle and torque for all four wheels, thereby reducing the number of controllers required. A mathematical model for trajectory tracking is established based on the Markov decision model, and the multi-objective reward function is refined using an exponential formulation. Additionally, an adaptive episode termination condition based on fuzzy control is developed. Simulation results demonstrate that the proposed exponential reward function effectively accelerates model convergence, the adaptive episode termination condition significantly boosts learning efficiency, and the multi-objective reward function distinctly enhances the overall performance of trajectory tracking. Specifically, the root mean square values for lateral deviation and heading angle deviation were reduced by 85.24 and 37.24%, respectively, and the steering wheel angle decreased by 33.49%, demonstrating the superior stability and accuracy of the proposed control strategy.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep reinforcement learning-based centralized control for trajectory tracking of distributed drive electric vehicles

  • 期刊论文

  • 发表日期：

    2025年8月1日

查看全文
  • Zhiyong Zhang,
  • Jiarui Zhang,
  • Caixia Huang,
  • You Wu,
  • Jie Zhang

摘要

In order to simplify the trajectory tracking controller for distributed drive electric vehicles and enhance its overall performance, this study introduces a machine learning approach rooted in the deep deterministic policy gradient algorithm. The proposed control strategy concurrently addresses three key performance objectives: lateral position, longitudinal velocity, and lateral stability. The control framework adopts a centralized structure to directly govern the steering wheel angle and torque for all four wheels, thereby reducing the number of controllers required. A mathematical model for trajectory tracking is established based on the Markov decision model, and the multi-objective reward function is refined using an exponential formulation. Additionally, an adaptive episode termination condition based on fuzzy control is developed. Simulation results demonstrate that the proposed exponential reward function effectively accelerates model convergence, the adaptive episode termination condition significantly boosts learning efficiency, and the multi-objective reward function distinctly enhances the overall performance of trajectory tracking. Specifically, the root mean square values for lateral deviation and heading angle deviation were reduced by 85.24 and 37.24%, respectively, and the steering wheel angle decreased by 33.49%, demonstrating the superior stability and accuracy of the proposed control strategy.