Scholarlink

<p>This study introduces a disturbance observer-based control framework that is enhanced through sequential gain optimization using reinforcement learning, specifically for servo-driven robotic manipulators. The proposed approach combines robust output feedback control with a disturbance observer, allowing for accurate estimation and compensation of external disturbances and unmodeled dynamics. To automate and optimize the tuning of controller and observer gains, the Twin-Delayed Deep Deterministic Policy Gradient (TD3) algorithm is utilized, enabling the system to adaptively refine its performance based on varying operating conditions. The disturbance observer plays a crucial role in improving regulation tasks by reconstructing disturbance signals in real-time, which enhances tracking accuracy and closed-loop stability. Unlike traditional manual tuning methods, this reinforcement learning-based strategy allows for dynamic and data-driven gain selection, resulting in more cohesive and responsive control behavior. Experimental validation was conducted on a DC motor system acting as an actuator within a mobile robotic platform. The results show significant improvements in trajectory tracking, response time, energy efficiency, and robustness against disturbances. These findings highlight the potential of combining disturbance observer-based control with reinforcement learning for advanced manipulation tasks that require high precision and adaptability.</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

Disturbance observer-based control with sequential gain optimization via reinforcement learning applied to servo-driven manipulators

  • 期刊论文

  • 发表日期：

    2026年5月11日

查看全文
  • Luis Luna,
  • Alejandro Guarneros Sandoval,
  • Isaac Chairez

摘要

This study introduces a disturbance observer-based control framework that is enhanced through sequential gain optimization using reinforcement learning, specifically for servo-driven robotic manipulators. The proposed approach combines robust output feedback control with a disturbance observer, allowing for accurate estimation and compensation of external disturbances and unmodeled dynamics. To automate and optimize the tuning of controller and observer gains, the Twin-Delayed Deep Deterministic Policy Gradient (TD3) algorithm is utilized, enabling the system to adaptively refine its performance based on varying operating conditions. The disturbance observer plays a crucial role in improving regulation tasks by reconstructing disturbance signals in real-time, which enhances tracking accuracy and closed-loop stability. Unlike traditional manual tuning methods, this reinforcement learning-based strategy allows for dynamic and data-driven gain selection, resulting in more cohesive and responsive control behavior. Experimental validation was conducted on a DC motor system acting as an actuator within a mobile robotic platform. The results show significant improvements in trajectory tracking, response time, energy efficiency, and robustness against disturbances. These findings highlight the potential of combining disturbance observer-based control with reinforcement learning for advanced manipulation tasks that require high precision and adaptability.