Velocity-based robotic assistance for refining motor skill training in a complex target-hitting task using a bio-mimetic trajectory generation model: A pilot study

•A velocity-based robotic assistance (VRA) in a target-hitting task is proposed.•A reference motion requiring VRA is generated using a minimum-jerk model.•The VRA system can teach a trainee the customized reference velocity profile.•The results demonstrate that VRA significantly affected the motor skill training.

Several methods have been proposed for robotic assistance in motor learning/training. However, a few major concerns such as the design of the natural motion of the hand of a trainee/patient by a robotic device considering the motor ability of the trainee/patient and a safe and effective method for teaching a trainee/patient, especially in a complex task requiring motor timing, still need to be addressed. This paper proposes velocity-based robotic assistance (VRA) using a bio-mimetic trajectory generation model for motor skill training in a target-hitting task considering the concerns mentioned above. In the designed motor task, a trainee has to contact an approaching ball with a racket at the desired time to hit the target on the wall while predicting the behavior of the ball before and after the contact. A set of the racket angle and hand velocity at the contact time is defined as a task-related motor skill. In the motor training with VRA, the time scale of a primitive reference velocity profile is automatically adapted to individual levels of task-related motor skills recorded in the past trials with no robotic assistance. The robotic device then teaches the trainee the customized reference velocity profile to facilitate motor skill training. The effect of VRA on motor skill training in a target-hitting task was investigated with sixteen healthy volunteers (male university students aged 22-24 years) to verify the concept in this pilot study. The skilled hand movement for the designed motor task was first determined using a set of results measured from four skilled subjects, and the primitive reference velocity profile with multiple peaks was successfully regenerated in a minimum-jerk model by utilizing the task-related constraints. Next, a set of training experiments with and without VRA was conducted with twelve subjects who have no experience in the target task. The results for the healthy subjects of this pilot study demonstrated that the proposed VRA was efficacious in facilitating the acquisition of task-related motor skills (with almost half trials) and in reducing temporal errors of the desired velocity (by approximately 40%).