Scholarlink

<p>Redundantly actuated constrained mechanisms (RACMs) are ubiquitous in various engineering applications. However, their non-Euclidean geometric structures and complex dynamics present significant challenges for modeling and control. This article propose a novel data-driven method for dynamic modeling and model predictive control (MPC) of redundantly actuated constrained mechanisms. For dynamic modeling, we propose neural network-based bilinear models on the active coordinates submanifolds, integrating both physics and data. Specifically, a new local parameterization technique allows us to consider dynamics on the active coordinates submanifold. It reduces the dimension of the ambient space and rigorously maintains position and velocity constraints. Moreover, deep bilinear networks on the active coordinates submanifold (DBNACS) effectively learn the lifting function and construct bilinear surrogate models. For MPC, we derive a real-time computable scheme for trajectory tracking based on the surrogate model. Numerical experiments including scenarios with friction and data corruption confirm the superiority of the proposed method over existing methods in both prediction accuracy and control performance.</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 bilinear models on manifolds for redundantly actuated constrained mechanisms: dynamics modeling and model predictive control

  • 期刊论文

  • 发表日期：

    2025年12月4日

查看全文
  • Haocheng Ying,
  • Ye Ding

摘要

Redundantly actuated constrained mechanisms (RACMs) are ubiquitous in various engineering applications. However, their non-Euclidean geometric structures and complex dynamics present significant challenges for modeling and control. This article propose a novel data-driven method for dynamic modeling and model predictive control (MPC) of redundantly actuated constrained mechanisms. For dynamic modeling, we propose neural network-based bilinear models on the active coordinates submanifolds, integrating both physics and data. Specifically, a new local parameterization technique allows us to consider dynamics on the active coordinates submanifold. It reduces the dimension of the ambient space and rigorously maintains position and velocity constraints. Moreover, deep bilinear networks on the active coordinates submanifold (DBNACS) effectively learn the lifting function and construct bilinear surrogate models. For MPC, we derive a real-time computable scheme for trajectory tracking based on the surrogate model. Numerical experiments including scenarios with friction and data corruption confirm the superiority of the proposed method over existing methods in both prediction accuracy and control performance.