Scholarlink

<p>Accurate dynamics parameter identification is pivotal for achieving high-precision control in industrial robotics. However, existing methods often face challenges in handling outliers, enforcing semi-definite constraints, and compensating for unmodeled dynamics. To address these issues, this paper proposes a two-stage framework that integrates robust physical parameter identification with residual compensation. First, a reverse-order identification strategy is employed to mitigate the adverse effects of inter-joint coupling. Building on this, a Least Absolute Deviation (LAD) estimator is introduced to calculate robust weights for outlier suppression. Simultaneously, Cholesky parameterization is utilized to transform the semi-definite constrained optimization into an unconstrained problem, intrinsically ensuring physical consistency through the mathematical structure. Subsequently, Physics-Informed Neural Networks (PINNs) are deployed to capture and compensate for unmodeled dynamic characteristics. Experimental results demonstrate that the proposed method effectively attenuates measurement noise and outliers, yielding a significant reduction in torque prediction errors.</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

Physically consistent and robust dynamics identification of industrial robots via PINN-based residual compensation

  • 期刊论文

  • 发表日期：

    2026年7月29日

查看全文
  • Wenbin Luo,
  • Pingjiang Wang,
  • Chaoxi Lin,
  • Deyu Su

摘要

Accurate dynamics parameter identification is pivotal for achieving high-precision control in industrial robotics. However, existing methods often face challenges in handling outliers, enforcing semi-definite constraints, and compensating for unmodeled dynamics. To address these issues, this paper proposes a two-stage framework that integrates robust physical parameter identification with residual compensation. First, a reverse-order identification strategy is employed to mitigate the adverse effects of inter-joint coupling. Building on this, a Least Absolute Deviation (LAD) estimator is introduced to calculate robust weights for outlier suppression. Simultaneously, Cholesky parameterization is utilized to transform the semi-definite constrained optimization into an unconstrained problem, intrinsically ensuring physical consistency through the mathematical structure. Subsequently, Physics-Informed Neural Networks (PINNs) are deployed to capture and compensate for unmodeled dynamic characteristics. Experimental results demonstrate that the proposed method effectively attenuates measurement noise and outliers, yielding a significant reduction in torque prediction errors.