Reshaping Movements Through Avoidance of Negative Events: A General Adaptive Controller

Precision can be challenging in human movement. Nikolai Bernstein noted that humans exhibit trial-to-trial movement variability in their trajectories when executing repetitive tasks, while displaying a great deal of precision in the endpoint. Thus, he coined the phrase “Repetition without repetition” to describe this phenomenon. Motion patterns that are highly variable can be troublesome if they exceed boundaries of safety or stability. Thus, we assessed the ability for individuals to reshape movements along particular directions through both haptic and visual means, through a virtual reality intervention we call limit-push. For the haptic study, we divided 18 healthy and 10 stroke survivors in two groups that complete a repetitive interception task in a 3d virtual reality environment: a treatment group that receives our virtual intervention and a control group that does not. The virtual intervention consisted of perturbing subject movements by exerting forces on their arm directed away from an invisible box that was 3.75 cm along the anterior-posterior axis, coincident with the interception axis, whenever they were outside during the middle 200 trials of the task. These forces were deactivated in the last 200 trials. We found that healthy subjects redistributed their motions within the box region with moderate success, however stroke survivors as a group could not. However, some stroke survivors could redistribute their motions within the box, while a few were unable to adapt. Both populations showed negligible residual effects of the intervention when the forces were turned off, however. Subsequently, we tested if 9 healthy subjects could adapt to a visual intervention instead. Instead of robotic forces, we displaced the cursor from the hand proportional to the distance that subjects were away from either edge of the box region when they were outside during the middle 200 trials. We observed similar effects to the healthy population that received robotic distortions. In this case, however, we did observe a small residual effect of the intervention once the visual intervention was deactivated, in the initial trials. However, this residual effect disappeared by the end of the experiment. Finally, we developed a new model of motor control that can explain adaptation and the associated variability without resorting to optimization methods through the principle of explore-and-avoid. We demonstrated the ability to recreate distributions observed in the limit-push­ studies described above. Furthermore, we demonstrated the ability for this model to adapt and perform classic motor behavioral tasks, ranging from targeted reaching to visuomotor rotation and obstacle avoidance. Ideally this could be used as a novel robotic controller that can imbue a robot with autonomy. It can also potentially be used as a predictor of individual movement variability and performance in novel tasks, as well as a training simulator and a new tool to help design environments that challenge patients in more precise ways.