PRISM: Parameter Recovery Identification from Synthetic Modeling

Abstract

Identifying parameters of human motor learning dynamics remains an open-ended question despite significant research in the field. While robotics and system identification offer tools for modeling dynamical systems with precision, few studies have investigated persistent excitation of human dynamics to produce accurate models. Here, we tested a forward-inverse method to estimate joint stiffness and damping as well as feedback gains in reaching when disturbed. Planar arm movements were first simulated with a synthetic model, then regression was used to identify gains, and the most crucial trial conditions (task and disturbance) were identified as those that brought the most accurate information about the gain parameters. Our results indicate that a set of 40 movements performed under specific perturbation conditions yielded the best parameter estimation ($R^{2}=0.97$). Such technique employs simulated or digital twin methods to best estimate the arm's control properties, offering a framework to optimize the experimental design and paving the way for more individualized approaches.