Iterative Learning for Gravity Compensation in Impedance Control

• Published in: IEEE/ASME transactions on mechatronics pp. 1 - 12
• Main Authors: Li, Teng, Zakerimanesh, Amir, Ou, Yafei, Badre, Armin, Tavakoli, Mahdi
• Format: Journal Article
• Language: English
• Published: IEEE 2024
• Subjects: Aerospace electronics; Gravity; Gravity compensation; Impedance; impedance control; iterative learning; physical human–robot interaction; robot-assisted arthroscopy; Robots; Surgery; Task analysis; Uncertainty
• ISSN: 1083-4435; 1941-014X
• DOI: 10.1109/TMECH.2024.3386407

Robot-assisted arthroscopic surgery has been increasingly receiving attention in orthopedic surgery. To build a robot-assisted system, dynamic uncertainties can be a critical issue that could bring robot performance inaccuracy or even system instability if cannot be appropriately compensated. Disturbance observer is a common tool to be used for disturbance estimation and compensation by taking all uncertainties as disturbances, but this will refuse human-robot interaction since the human-applied force will also be regarded as a disturbance by the observer. Iterative learning for gravity compensation can be another promising way to solve this problem when gravity compensation is the main concern. In this article, a gravity iterative learning (Git) scheme in Cartesian space for gravity compensation, integrating with an impedance controller, is presented. A steady-state scaling strategy is then proposed, which released the updating requirements of the learning scheme and also extended its validity to trajectory-tracking scenarios from set-point regulations. The deriving process and convergence properties of the Git scheme are presented and theoretically analyzed, respectively. A series of simulations and physical experiments are conducted to evaluate the validity of the scaling strategy, the learning accuracy of the Git scheme, and the effectiveness of the learning-based impedance controller. Both simulation and experimental results demonstrate good performance and properties of the Git scheme and the learning-based impedance controller.