Experimental evaluation of deadbeat running on the ATRIAS biped

Abstract

Theoretical spring mass models with deadbeat foot placement policies reveal very robust running in the presence of large and frequent, unexpected gait disturbances. Although this performance goes beyond what has been demonstrated on run- ning machines, a transfer of this theory has only been investigated for simplified monopod systems. Here we investigate how well the control strategies developed for the spring mass model transfer to more complex and human-like robots. We use a model-based control framework to implement the spring mass behavior on a bipedal robot of human scale and weight with articulated legs and an actively stabilized trunk. Evaluating the tracking performance in robot experiments on undisturbed and disturbed running, we find that our controller achieves tracking consistent with the underlying model for velocity changes of ±0.2 m · s−1. For larger velocity changes and ground height disturbances up to ±15 cm, the controller performance degrades but the robot maintains running. Based on perturbed simulations of the simplified model, we conclude that the degradation is largely related to force disturbances not considered in the underlying spring mass control theory. The results highlight both limitations of the existing spring mass theory for control of more complex machines and an SMM-based control that generates robust and versatile behavior in running robots.