A Framework for Autonomous Impedance Regulation of Robots Based on Imitation Learning and Optimal Control

In this work, we propose a framework to address the autonomous impedance regulation problem of robots in class constrained manipulation tasks. framework, human arm endpoint stiffness model is used extract task geometry along trajectory, which then encoded offline and reproduced online by Gaussian Mixture Model (GMM) Regression (GMR), respectively. Furthermore, full Cartesian robot formulated through an optimal control problem, i.e., Linear-Quadratic Regulator (LQR), (extracted from demonstrations) considered as time-varying weighting matrix Q. The eventually realised consistent controller. A tank-based passivity observer implemented give evidence on stability system during variations. To evaluate performance comparative experiment with three different settings (i.e., proposed without LQR GMM/GMR) for Franka Emika Panda perform door opening was conducted. results reveal that our outperforms other two, terms tracking error interaction forces.