Virtual Model Control for Dynamic Lateral Balance

Abstract

Motivated by an interest in human-like controllers for humanoids to increase their social acceptance, we investigate lateral balancing for artistic performances on challenging surfaces. Control design for lateral balancing in humanoids has primarily focused on optimal control techniques. While these techniques generate balancing controllers, it remains unclear whether humans use similar strategies. Here we propose that humans prefer intuitive task-space control for lateral balancing on simple as well as challenging surfaces. We develop a virtual model controller for this task and compare with simulations of a planar model, the resulting balancing behavior against human lateral balancing on flat ground and on a seesaw as an example of a challenging surface. We find that the proposed controller can be tuned to respond to balance disturbances on flat ground in a human-like way, and that it mimics human behavior on a seesaw including the failure to stabilize the board, even though an optimal LQR controller is capable of stabilizing it. The results support the hypothesis that humans prefer intuitive control in lateral balancing and suggest that state-of-the-art control approaches in robotics may go beyond what humans can accomplish. These limitations should be taken into account when designing human-like controllers for humanoids.