Feedback error learning at the muscle level: A unified model of human motor adaptation to stable and unstable dynamics

Abstract

This thesis presents a computational model of human motor adaptation to novel dynamics. It is shown that feedback error learning at the muscle level, in addition to a deactivation strategy to minimize agonist-antagonist cocontraction, can acquire an inverse dynamics model compensating for reproducible dynamics and anisotropic impedance to counteract irreproducible dynamics. This ensures the learning of appropriate motor commands to compensate for stable and unstable novel dynamics. The motion trajectories, evolution of muscle activity, and endpoint impedance resulting from full dynamics simulation are consistent with all available experimental results (Burdet et al 2001A, Franklin et al 2003A, Franklin et al 2003B, Osu et al 2003), indicating that the model is plausible. The sensitivity to model parameters and the predicted results in various novel dynamics are also examined.