Angular momentum primitives for human walking: biomechanics and control

Abstract

Towards the goal of developing stable humanoid robots and leg prostheses, we present a biologically motivated control strategy for walking where system angular momentum is explicitly controlled. Using human kinematic gait data, we calculate the distribution of spin angular momentum throughout the human body at slow and self-selected walking speeds. Principal component analysis reveals three angular momentum primitives that explain 99% of the walking data for sagittal plane body rotations. In addition, our analysis shows that the angular momentum primitives are invariant with walking speed. Using these biomechanical results, we simulate human walking during the single support phase using a morphologically realistic humanoid model walking in the sagittal plane. There is minimal predefined specification of the desired gait motion. With only the model's walking speed and stride length as an input, our control system searches for joint reference trajectories that minimize the error between the model's angular momentum distribution and the biologically determined distribution. Resulting model joint kinematics are in qualitative agreement with human gait data, suggesting that exploiting invariant angular momentum primitives In humanoid control may prove critical to achieving biological realism in legged robots and prostheses. The angular momentum primitives framework can substantially simplify the process of gait synthesis and enable the operator of a humanoid robot or powered leg prosthesis to easily change stride length and/or walking speed.