Robotic movement training as an optimization problem: designing a controller that assists only as needed

One of the prevailing paradigms of physical rehabilitation following neurologic injury is to "assist-as-needed"; that is, the rehabilitation therapist manually assists patients in performing movements, providing only as much assistance as needed to complete the movement. Several research groups are attempting to automate this principle with robotic movement training devices. This paper derives an "assist as needed" robotic training algorithm by framing the problem as an optimization problem. We assume that motor recovery can be modeled as a process of learning a novel sensory motor transformation. The optimized robotic movement trainer is then an error-based controller with a forgetting factor. It bounds kinematic errors while systematically reducing its assistance. The same controller also works well if the dominant dynamics of recovery are akin to a strengthening process. We experimentally validate the controller with an unimpaired subject by demonstrating how the controller can help the subject to learn a novel sensory motor transformation (i.e. an internal model) with smaller kinematic errors than typical. The task studied here is walking on a treadmill in the presence of a novel dynamic environment. The assist-as-needed controller proposed here may be useful for limiting error during the learning of tasks in which large errors are dangerous or discouraging.