A gain-scheduling approach to model human simultaneous visual tracking and balancing

In this study, we endeavor to better understand the human motor control system in order to help transposing some of its features onto humanoid robots. The postural coordination task investigated is related to an experimental paradigm that consists in visual target tracking task while balancing. We want to test whether the human biomechanical responses, namely phase / antiphase coordination mode transition, as exhibited during the actual experiments can be modeled by a linearized double inverted pendulum and parallel independent PD feedback control loops. Remarkably, these loops implement joint space control using cartesian task space variables. Furthermore, we want to see how the feedback control gains given by an optimization procedure scale w.r.t frequency or target motion magnitude. A closed-loop synthesis is developed that consists in minimizing a minimum torque criterion under both balance and task constraints. We show that the optimal feedback control gains obtained yield model responses consistent with the literature. In a second part, we implement a gain-scheduling approach where control gains values are predicted via interpolation. Finally, our approach implements a controller capable of achieving the task even when the frequency of the target motion varies over time.