Evaluating Active Assessment and Learning Control for Robotic Rehabilitation

Abstract

Robotic exoskeletons show great promise in supplementing rehabilitation interventions. These devices provide clinicians with instantaneous feedback for impairment and movement quality, and unique interventions, such as decoupling strength and dexterity via gravity compensation in functional exercise movements. Accounting for shortcomings in robot-only therapeutic interventions, combining functional electrical stimulation and exoskeleton control allows for combined effort from hybrid muscular and motor interventions. This combined modality allows for torque generation from stimulated muscle recruitment with the precision, accuracy, and repetition capabilities from wearable exoskeletons. This dissertation investigates active-based assessment methodologies and validates individualized, robust hybrid control techniques with an upper-extremity exoskeleton. First, a pilot study establishes movement quality in the context of Fugl-Meyer Upper-Extremity (FM-UE) in unconstrained movement space accompanied by a novel reaching and movement compensation framework. Focusing on a chronic stroke population, these implementations lay the groundwork for understanding the role of gravity compensation for unconstrained, upper-extremity motion tasks. Second, this dissertation applies and evaluates deep neural network~(DNN) based control algorithms to adapt to unmodeled system dynamics. By approximating exoskeleton dynamics, DNNs are able to act as highly effective function approximators, especially in highly nonlinear systems such as neuromuscular movement. This work presents first of its kind hardware implementation of multiple learning-based control methods for both motor and hybrid control for single and multi joint movements on an series-elastic driven upper extremity exoskeleton, validated in healthy and motor function impaired participants. Results reveal improved performance in learning based control methods and promote the efficacy of robotics in rehabilitation. Specifically, this work highlights the importance of control architecture and the role that activation functions play in the context of movement quality for upper-extremity tasks. In summary, this dissertation presents new insights into the role of active assessment and hybrid control schemes which have potential for application in rehabilitation interventions and assessments.