Mechanisms of motor learning: by humans, for robots

Whenever we perform a movement and interact with objects in our environment, our central nervous system (CNS) adapts and controls the redundant system of muscles actuating our limbs to produce suitable forces and impedance for the interaction. As modern robots are increasingly used to interact with objects, humans and other robots, they too require to continuously adapt the interaction forces and impedance to the situation. This thesis investigated the motor mechanisms in humans through a series of technical developments and experiments, and utilized the result to implement biomimetic motor behaviours on a robot. Original tools were first developed, which enabled two novel motor imaging experiments using functional magnetic resonance imaging (fMRI). The first experiment investigated the neural correlates of force and impedance control to understand the control structure employed by the human brain. The second experiment developed a regressor free technique to detect dynamic changes in brain activations during learning, and applied this technique to investigate changes in neural activity during adaptation to force fields and visuomotor rotations. In parallel, a psychophysical experiment investigated motor optimization in humans in a task characterized by multiple error-effort optima. Finally a computational model derived from some of these results was implemented to exhibit human like control and adaptation of force, impedance and movement trajectory in a robot