Dynamic bipedal walking by controlling only the equilibrium of intrinsic elasticities

This paper presents a methodology for controlling dynamic bipedal walking in a compliantly actuated humanoid robotic system. The approach is such that it exploits the natural leg dynamics of the single and double support phase of the gait. The present approach avoids to close a torque control loop at joint level. While simulation implementations of torque based walking for series elastic actuator (SEA) humanoids display very promising results, several robustness issues very often appear in the experiments. Therefore we introduce here a minimalistic controller, which is based on feedback of control input collocated variables, with the only exception of zero joint torque control. Reshaping of the intrinsic elasticities by control is completely avoided. In order to achieve a coordinated movement of swing and stance leg during single support phase, an appropriate one-dimensional manifold of the motor positions is designed. This constrained behavior is experimentally shown to be compatible with the intrinsic mechanical oscillation mode of the double support phase. The feasibility of this methodology is experimentally validated on a human-scale, anthropomorphic bipedal robotic system with SEA actuation.