A push recovery strategy for a passively compliant humanoid robot using decentralized LQR controllers

Abstract

This paper presents a control scheme that is directed towards the performance of push recovery on the compliant humanoid robot, COMAN. The novelty offered by this work is related to the use of a decentralized controller based on an initial Limited Quadratic Regulator (LQR) design on a humanoid robot in addition to the regulation of the actual joint positions instead of the motor positions. Moreover, the ankle-knee strategy is examined through the use of a compliant double inverted pendulum model. A key feature of the propounded approach lies in the controller's ability to regulate the system's inherently compliant dynamics through considering not only the motor-related variables but also those of the link-side, appearing after the passive compliant element. Consequently, this leads to a control method that is capable of stabilizing the robot by means of increasing the damping on the link, which is essential given the system's oscillatory behaviour once it has been perturbed.