Scholarlink

<p>The control of variable stiffness actuators (VSAs) is challenging because they have highly nonlinear characteristics and are difficult to model accurately. Classical control approaches using high control gains can make VSAs stiff, which alters the inherent compliance of VSAs. Iterative learning control can achieve high tracking accuracy for VSAs but generally lacks sufficient generalization. This study applies Gaussian process (GP) regression to design a stable tracking controller combining feedforward and low-gain feedback control actions for agonistic-antagonistic (AA) VSAs subjected to unknown dynamics. The GP model learns the inverse dynamics of AA-VSAs and provides model fidelity by the predicted variance. The stability analysis of the closed-loop system demonstrates that the tracking error is uniformly ultimately bounded and exponentially converges to a small ball under a given probability. Experiments on an AA-VSA named qbmove Advanced have validated the superiority of the proposed method with respect to tracking accuracy and generalization.</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

Stable Gaussian process tracking control of antagonistic variable stiffness actuators

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年6月5日

查看全文
  • Zhigang Zou,
  • Wei Li,
  • Chenguang Yang,
  • Yongping Pan

摘要

The control of variable stiffness actuators (VSAs) is challenging because they have highly nonlinear characteristics and are difficult to model accurately. Classical control approaches using high control gains can make VSAs stiff, which alters the inherent compliance of VSAs. Iterative learning control can achieve high tracking accuracy for VSAs but generally lacks sufficient generalization. This study applies Gaussian process (GP) regression to design a stable tracking controller combining feedforward and low-gain feedback control actions for agonistic-antagonistic (AA) VSAs subjected to unknown dynamics. The GP model learns the inverse dynamics of AA-VSAs and provides model fidelity by the predicted variance. The stability analysis of the closed-loop system demonstrates that the tracking error is uniformly ultimately bounded and exponentially converges to a small ball under a given probability. Experiments on an AA-VSA named qbmove Advanced have validated the superiority of the proposed method with respect to tracking accuracy and generalization.