Pattern generation and compliant feedback control for quadrupedal dynamic trot-walking locomotion: experiments on RoboCat-1 and HyQ

Abstract

In this paper, we introduce a method that synergistically combines an analytical pattern generator and a feedback controller frame, which are developed for the purpose of synthesizing dynamic quadrupedal trot-walking locomotion on flat and uneven surfaces. To begin with, the pattern generator analytically produces feasible and dynamically balanced joint motions in accordance with the desired trot-walking characteristics, with no empirical parameter tuning requirements. In concurrence with the pattern generation, a two-phased controller frame is constructed for closed-loop sensory feedback: (i) virtual admittance controller via force sensing, (ii) upper torso angular momentum regulation via gyro sensing. The former controller evaluates joint force errors and generates the corresponding joint displacement for a given set of virtual spring-damper couples. Together with the position constraints, these displacements are additionally fed-back to local servos for achieving compliant quadrupedal locomotion with which the position/force trade-off is addressed. The second controller, that is simultaneously used, evaluates the upper torso angular momentum rate change error using measured and reference orientation information. It then regulates the torso orientation in a dynamically consistent way as the rotational inertia is characterized. In order to validate the proposed methodology several experiments are conducted on both flat and uneven surfaces, using two robots with distinct properties; a $$\sim $$~7 kg cat-sized electrically actuated quadruped (RoboCat-1), and a $$\sim $$~80 kg Alpine Ibex-sized hydraulically actuated quadruped (HyQ). As a result we demonstrate continuous, repetitive, compliant and dynamically balanced trot-walking cycles in real-robot experiments, adequately confirming the effectiveness of the proposed approach.