Scholarlink

Active suspension system (ASS) plays a crucial role in enhancing both vehicle ride comfort and handling stability. Modern control algorithms for active suspension typically require a mathematical model, but the inherent nonlinearity of actual systems introduces modeling errors. This paper presents a DA-DDPG network architecture that incorporates the DAgger algorithm for dataset augmentation to improve model convergence. By using deep reinforcement learning (DRL), the system interacts directly with the environment, allowing deep neural networks to capture the system's nonlinear characteristics. Simulation results demonstrate that the DA-DDPG algorithm offers superior control performance across all frequencies compared to heuristic methods like skyhook control and outperforms traditional DRL strategies in terms of convergence. Bench tests confirm that, under various conditions such as speed bumps and C-grade random roads, the DA-DDPG algorithm consistently exceeds optimal control-based approaches in minimizing vertical vehicle vibrations. Its real-time performance and robustness are also validated.

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

Integrated Active Suspension Deep Reinforcement Learning Control Strategy with Dagger Algorithm

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Wang Xiaoyu,
  • Chen Weihao,
  • Chen Xiaokai,
  • Liu Xiang

摘要

Active suspension system (ASS) plays a crucial role in enhancing both vehicle ride comfort and handling stability. Modern control algorithms for active suspension typically require a mathematical model, but the inherent nonlinearity of actual systems introduces modeling errors. This paper presents a DA-DDPG network architecture that incorporates the DAgger algorithm for dataset augmentation to improve model convergence. By using deep reinforcement learning (DRL), the system interacts directly with the environment, allowing deep neural networks to capture the system's nonlinear characteristics. Simulation results demonstrate that the DA-DDPG algorithm offers superior control performance across all frequencies compared to heuristic methods like skyhook control and outperforms traditional DRL strategies in terms of convergence. Bench tests confirm that, under various conditions such as speed bumps and C-grade random roads, the DA-DDPG algorithm consistently exceeds optimal control-based approaches in minimizing vertical vehicle vibrations. Its real-time performance and robustness are also validated.