SUMMARY

The goal of this research is to develop theories, methods, and tools to understand the mechanisms of neuromotor adaptation in human-robot physical interaction, in order to improve the stability and performance of the interaction. Human power-assisting systems (e.g., powered lifting devices that aid human operators in manipulating heavy or bulky loads) require physical contact between the operator and machine, creating a coupled dynamic system. This dynamic coupling has been shown to introduce inherent instabilities and performance degradation due to a change in human stiffness; when instability is encountered, a human operator often attempts to control the oscillation by stiffening their arm, which leads to a stiffer system with more instability. Robot co-worker controllers must account for this issue. In this work we set out to 1) understand the association between neuromuscular adaptations and system performance limits, 2) develop probabilistic methods to classify and predict the transition of operator’s cognitive and physical states from physiological measures, and 3) integrate this knowledge into a structure of shared human-robot control. We developed a model of the human operator endpoint stiffness, characterized at the musculoskeletal level, that can account for deliberate stiffness increase at the endpoint through the incorporation of muscle coactivation. We also developed a switching admittance control approach which can account for changes in the operator's muscle coactivation and is able to generate cognitive states in an unsupervised manner, given a relevant training dataset. Finally, a novel variable admittance control approach, which significantly reduces grasp contact instability commonly encountered in fixed admittance control settings, was developed, analytically derived, and provides solutions for both constant mass and variable mass parameter cases.