An MPC-based two-dimensional push recovery of a quadruped robot in trotting gait using its reduced virtual model

Abstract

This paper addresses the push recovery of quadruped robots trotting on even terrain. Push recovery is the ability of a legged robot to maintain its balance in the presence of sudden external forces applied to the robot. Due to the nature of this dynamical gait, the quadruped robot can be modeled as a biped robot where each two cross legs of the quadruped are modeled as a virtual leg. Then, the virtual biped model is further reduced to a two-dimensional linear inverted pendulum plus flywheel model (LIPFM). Moreover, using the concept of capture points for the biped model, the desired locations for the center of pressure (COP) of the legs for recovering the robot's balance are calculated by designing a two-dimensional dynamic capture point estimator. A model-based predictive controller (MPC) is then proposed to adjust the footstep locations and generate a new walking-pattern. The resulting redesigned joint angles for the virtual biped model are then transformed back to those of the actual quadruped model. Simulation results show the effectiveness of the proposed method.