Computational and psychophysical studies of motor learning

A remarkable characteristic of our motor system is its enormous capacity for change, manifest in our ability to acquire new skills (motor learning) and to adapt existing behaviours to better suit the environment (motor adaptation). Common to both is the need to store information about the environment and the body, which can be thought of as building or updating internal models. Two learning rules used in the literature to learn internal models for control (Feed back Error Learning and Distal Supervised Learning) are analysed, examining possible experimental predictions. Adaptation to changing environments is studied using paradigms based on reaching movements under switching sequences of two conflicting force-fields. There is good evidence that the motor system adapts to an external force-field by acquiring an internal model that predicts future corrections from past errors. However, the switching force-field paradigm reveals that in a situation where the environment is non-stationary, i.e. changing over time, such predictive control can break down and efficient feedback control becomes more important. This paradigm is used to study the role of online feedback, dual adaptation, modularity of internal models and the effect of gradually increasing task difficulty. The role of feedback versus feedforward control is qualitatively explained by a simple probabilistic model of adaptation. The model makes predictions about behaviour in environments of varying degrees of uncertainty, which are tested experimentally using a grip-force paradigm. Many studies of motor learning have focused on simple tasks. The final chapter examines motor learning of a complex skill: playing Kendama or cup and ball. Kendama is simulated using virtual haptic and visual feedback, allowing all aspects of the task to be experimentally controlled. The role of visual and haptic feedback, eye-hand coordination and error propagation during learning are studied using this set-up