Learning stability-guaranteed skill and adaptive control strategies from demonstrations for heterogeneous component robotic machining

Hongmin Wu, Xueqian Zhai, Haochen Zheng, Zhaoyang Liao, Zhihao Xu, Xuefeng Zhou,
Learning stability-guaranteed skill and adaptive control strategies from demonstrations for heterogeneous component robotic machining,
Journal of Manufacturing Processes,
Volume 151,
2025,
Pages 506-520,
ISSN 1526-6125,
https://doi.org/10.1016/j.jmapro.2025.06.095.
(https://www.sciencedirect.com/science/article/pii/S1526612525007534)
Abstract: Heterogeneous Material Components (HMCs), including structures such as lamination and transverse splicing, are widely used in modern manufacturing, but often face problems with structure, size, and surface quality due to process uncertainties. Traditional robot control methods with fixed impedance parameters struggle to adapt to the diverse materials in robotized machining tasks, leading to instability and inefficiency. This paper introduces a human demonstration-based robotic machining framework for HMCs. We develop a stable, nonlinear dynamical system that handles rhythmic movements and adapts to external perturbations. Our force-related velocity adjustment strategy ensures smooth transitions between contact phases, and we design a variable impedance controller to handle varying material dynamics. The proposed framework, validated through two experiments, enhances machining accuracy and reduces impact force compared to fixed-parameter controllers. Specifically, the average diameter error in the ”paper-plastic foam-wood” laminated component drilling task is 0.676mm, representing a 54% improvement over traditional controllers. In the polishing experiment with ”wood-iron” transverse splicing components, the maximum impact force during the transition from contact to non-contact is 3.384N, with a cross-material maximum impact force of 3.547N. Results indicate that our stable skill learning and adaptive variable impedance control framework effectively improves the processing of HMCs, including metrics such as maximum impact force, drilling diameter, and surface roughness.
Keywords: Robot skill learning; Variable impedance control; Learning from demonstration; Heterogeneous materials machining