A controller for dynamic walking in bipedal robots

Abstract

This paper presents an approach for the closed-loop control of actuated biped that allows natural looking and energy efficient walking. Rather than prescribe kinematic trajectories or kinematic constraints, the approach is based on the prescription of state dependent torques that ¿encourage¿ patterned movement. Some of the prescribed torques are referenced to the inertial reference frame, which largely decouples the angular dynamics of the robot, and as such greatly simplifies the selection of control parameters. Implementation of torques from the inertial coordinate frames is enabled by a joint torque computation which is motivated by Gauss's principle of least constraint. The proposed approach is implemented in simulation on an anthropomorphic biped, and is shown to quickly converge to a natural looking gait limit cycle. Simulations are conducted with various control parameters and different initial conditions. The authors also show that walking speed can be altered in a simple manner by varying two intuitive controller parameters. The mechanical cost of transport computed on a representative dynamic walk is used to validate energy efficiency of the proposed control approach.