Electronic Thesis and Dissertation Repository

Thesis Format

Integrated Article


Doctor of Philosophy




Gribble, Paul L


The motor system is continuously monitoring our performance, ensuring that our actions are occurring as planned. Sensory prediction errors, which arise from a discrepancy between the expected and actual sensory consequence of a motor command (i.e., a planned action), are assumed to drive sensorimotor adaptation. Sensorimotor adaptation is thought to involve changes in motor output that allow the motor system to regain its former level of performance in perturbed circumstances. We employed experimental paradigms that involved both mechanical and visual perturbations to evoke sensory prediction errors while participants performed planar reaching movements. Movement error activates learning processes in the brain, which alter our behaviour in the future. A prominent model of short-term adaptation is built upon the theory that there appear to be at least two processes of varying timescales operating together as humans learn to counteract sensorimotor disturbances: a fast process that learns to reduce errors quickly but also quickly forgets, and a slow process that learns to reduce errors slowly but slowly forgets. The purpose of this dissertation was to track the mechanisms of short-term motor adaptation within the framework of a two-state model. Collectively, our three studies reinforce the hypothesis that short-term sensorimotor adaptation, occurring over short time scales (e.g., over a period of minutes), is supported by at least two underlying processes. Substantiated by our first and third study, we have shown that both the fast and slow adaptation processes are responsive to a history of error and both contribute to savings. The motor system receives sensory feedback about both the environment and the body on a continual basis, in addition to predictive feedforward commands. How feedback gains are changed can vary greatly, based on the state of the body and environment, as well as the behavioural context of learning. It has been routinely suggested that adaptation in response to a perturbation, results in a gradual shift over the course of error-reduction from a feedback-driven mode of control to more predictive, feedforward control. Based on the results of our second study, we demonstrate that the fast process of feedforward adaptation parallels the modulation in gain of the feedback response over the course of learning to counter a force field perturbation. We propose that the fast process, estimated from overall learning, may alternatively be an identification of the feedback controller, while the slow process is the recalibrated forward model. And lastly, while unpacking the result of our third study we further suggest that it is the slow process which stores a memory component from prior training which is then later accessed by both processes during subsequent learning.

Summary for Lay Audience

Reaching toward an object requires that you plan your actions in advance, estimate how those actions will unfold, and use available feedback to make any corrections, if needed. When we learn new actions we initially rely heavily on feedback from our vision and from our sense of where our body is relative to a goal object. Initially our movements vary from attempt to attempt, but with practice we improve our performance. The main theory as to why we can improve suggests that we actually become better at predicting the outcome of an action plan; this reduces the need to depend on feedback, which has a delay before it reaches the brain. Overall, we want the movements we actually make to match the movements we intend to make, and any difference is considered an error. When we experience an error, interestingly, we do not just uniformly adjust our movements. Our work is based on a computational approach to understanding how we change our behaviour in the face of movement errors and suggests that the experience of an error engages two processes of varying timescales operating together: a fast process that learns to reduce errors quickly but also quickly forgets, and a slow process that learns to reduce errors slowly but slowly forgets. If we look at these two processes, we can see which process is most dominant to the overall change in behaviour at a given point in time, and estimate the reliability of each input source. We propose that the fast process may indicate when we prioritize feedback, while the slow process indicates when we prioritize the intended action. Additionally, we considered the phenomenon of savings, which is a classic feature of human motor behaviour and is defined as the ability of prior learning to speed up subsequent relearning. We suggest that the slow process stores a form of memory that allows the motor system to evaluate if we have encountered this error previously. This memory is available to both processes and when we are relearning a task, we are willing to learn more from that same error.