Data-Driven Neuromechanical Modeling, Estimation, and Control for Hand-Assistive Robotic Interfaces

Abstract

Mechanical interactions at the hand affect its entire neuromuscular circuit - from hand kinematics and peripheral nervous activation, to motor and somatosensory cortex activation, to cognition and decision making. Neuromechanics is the approach of understanding mechanical interactions as the biproduct of these subsystems. An important domain of neuromechanical analysis is the development of hand-assistive robotics, a diverse range of technologies that facilitate neuromuscular rehabilitation for individuals with motor impairments, sensory feedback to accelerate skill acquisition, and assistance to support functional movement and strength. Hand assistive devices are typically designed for locally stationary neuromechanical states such as kinematics and muscle activation, providing maximally effective interactions only under fixed experimental conditions. In practice, neuromechanical states like fatigue and motor learning impose nonstationary transition dynamics that can impede interaction objectives and reduce performance quality. This thesis introduces a suite of computational neuromechanics tools comprising the NeuroSiGHT, NeuroMERGE, and NeuroGAIN algorithms. NeuroSiGHT integrates neuromuscular theory and topological data analysis to construct a robust partially observable Markov decision process model of stationary and nonstationary neuromechanical states from surface electromyography. NeuroMERGE refines these insights by introducing an active exploration strategy for assistive robots to estimate neuromechanical state transition dynamics unsupervised through controlled perturbations. NeuroGAIN extends this by employing generative architectures to forecast state transition dynamics and optimally evolve estimation and control strategies. To explicate its utility, the suite is validated in simulation and human participant experiments aiming at improving motor performance. The significant take-away of this thesis is reduction of neuromuscular fatigue and effort, and enhanced hand motor-cognitive efficiency in hand motor skill acquisition and performance. The neuromechanical suite enables practical neuromuscular interaction control at the hand.