Bayesian Estimation of Potential Performance Improvement Elicited by Robot-Guided Training

Improving human motor performance via human-robot collaboration is a major interest in society in many different contexts, such as rehabilitation, sports training, and so on. Despite the increasing demands, the potential and limitations have been controversially discussed. This study proposes a versatile method that can statistically elaborate on the relation between the performance improvements and the initial skill level. The procedure is explained by applying an experimental data of 20 healthy subjects interacting with a robot-assisted motor training system from our laboratory. The subjects are physically guided through an ideal motion by the haptic interface, which is a major approach in robotic rehabilitation to facilitate the motor functional recovery. Meanwhile, such haptic guidance training is conjectured to improve motor performance with lower skills. Although some studies show such a tendency, a method for defining the effective boundary level has not been proposed. Identifying the boundary promises positive training effects for target users of each task or type of robotic training. This study proposes an identification method to figure out the training effect's dependence on the initial skill level thorough modelling the skill level change. With the proposed statistical method, the initial skill's boundary level could be simultaneously derived as inferring the model parameters. The pre- and post-performance showed that the post-performance can be presumed depending on each subject's initial skill level.