A Control Framework With Tactile Diffusion Policy and Variable Impedance for Unknown Surface Tracking
Zhiyi Li, Shenyuan Deng, Chao Zeng, Chenguang Yang
Abstract
Precise position tracking and compliant interaction between robots and unstructured environments have always been a research hotspot, particularly in unknown surface tracking tasks. Traditional approaches typically rely on force sensor data to estimate surface normals, but suffer from certain estimation errors due to inaccurate friction measurement. We innovatively regard unknown surface tracking as an imitation learning process, proposing a tactile diffusion policy (TacDP)-based orientation compensation method that enables autonomous surface normal tracking during scanning operations. Concurrently, to achieve force tracking and enhance robustness, a passivity-based variable impedance controller is introduced, which adaptively adjusts the stiffness law through quadratic programming (QP) optimization and guarantees the system passivity by incorporating energy tank-based constraints to ensure stable robot-surface contact. Simulation experiments support our selection of this controller. Experimental validation on complex curved surfaces demonstrates the effectiveness of the proposed control framework.
BibTeX
@inproceedings{ral2026_acontrolframewor,
title = {A Control Framework With Tactile Diffusion Policy and Variable Impedance for Unknown Surface Tracking},
author = {Zhiyi Li and Shenyuan Deng and Chao Zeng and Chenguang Yang},
booktitle = {RA-L 2026},
year = {2026}
}