Myoelectric Sensing for Intent Detection and Assessment in Upper-Limb Robotic Rehabilitation

Abstract

This thesis explores how surface electromyography (EMG) -- the measurement of muscle force through voltage changes at the skin surface – can be of use to the field of upper-limb robotic rehabilitation. We focus on two main aspects: detecting human intention from measured muscle activity and assessing human motor coordination through synchronous muscle activations known as muscle synergies – each examples of the bidirectional communication found in tightly integrated human-robot interaction. EMG-based intent detection presents an opportunity to examine and promote human engagement at the neuromuscular level, enabling new protocols for intervention that could be combined with robotic rehabilitation, particularly for the most impaired of users. Meanwhile, the latest research in motor control proposes that natural, healthy human movement can be characterized by the presence of certain muscle synergies, and that the alteration of these synergies indicates a disruption, from neurological impairment or some other physical constraints, in natural movement. Wearable robotic devices are capable of altering muscle synergies, and though the mechanisms are not yet understood, a focus on altering muscle synergies is a promising new approach to neurorehabilitation. This thesis employs a robotic exoskeleton for the elbow and wrist joints designed for research in robotic rehabilitation of individuals with neurological impairments and now integrated with a myoelectric control interface. We first demonstrate the ability of a myoelectric interface to discern the user’s intended direction of motion in single-degree-of-freedom (DoF) and multi-DoF control modes with 10 able-bodied participants and 4 participants with incomplete cervical spinal cord injury (SCI). Predictive accuracy was high for able-bodied participants (averages over 99% for single-DoF and near 90% for multi-DoF), and performance in the SCI group was promising (averages ranging from 85% to 95% for single-DoF, and variable multi-DoF performance averaging around 60%), which is encouraging for the future use of myoelectric interfaces in robotic rehabilitation for SCI. Second, we explore the identification of synchronous muscle synergies in the muscles controlling the elbow and wrist, and the possible effects of robot-imposed task constraints on the neural constrains represented by synergy patterns. Our results indicate that constraining the unused degrees of freedom during a single-DoF movement inside the exoskeleton does not have a significant effect on the underlying muscle synergies in the task, and that methodological choices in muscle synergy analysis also do not have a large effect on the outcome. With all of these findings, we have achieved a deeper understanding of the value myoelectric sensing can bring to upper-limb robotic rehabilitation, and how much potential it has to advance the field toward greater accessibility to individuals of all levels of impairment.