The Role of Multiple Learning Systems in Sensorimotor Adaptation of Human Reaching
Humans are very good at learning to make new movements, whether this is to practice a skill that many other people can perform or to overcome a new situation that they have never encountered. For instance, astronauts learn to maneuver in zero gravity and skydivers learn to precisely control falling with the poise of an acrobat. The same learning is evident in everyday life, as people regularly adjust for the small changes to their movements caused by articles of clothing, such as the additional weight of a watch on the forearm or the differences in gait necessary for many types of shoes. In motor learning research, it has been an open question whether learning a new skill, such as the controlled fall from skydiving, is the same as altering an existing motor skill, such as reaching, to compensate for the weight of a watch. In my dissertation work I have focused on the question of how and why people alter their existing motor skills, a type of learning called adaptation.
Adaptation is a specific subset of motor learning that occurs when the sensory outcome of motor commands is systematically altered. In order to induce this adaptation in the laboratory, we manipulate the visual feedback that human participants see when they are performing reaching tasks. It is thought that this type of learning, visuomotor adaptation, is driven by the difference between the feedback that was predicted to occur and the actual feedback. This discrepancy in feedback is known as a sensory prediction error. If present, these errors indicate that the sensorimotor system is not properly calibrated, and future motor commands (and their predicted sensory outcomes) are adjusted to bring the system back into alignment. Adjustments made to the motor commands by this process are historically believed to be independent of other factors that commonly affect learning, such as reward and punishment. It is becoming increasingly accepted, however, that the behavior observed in sensorimotor adaptation tasks may not only be the output of error-based adaptation. In the work that forms my dissertation, we attempted to characterize the effect of three different systems on behavior in visuomotor adaptation tasks.
In Chapter 1, we examined savings-upon-relearning in visuomotor adaptation tasks. Savings is the phenomenon of faster relearning after something has been forgotten. Visuomotor adaptation seems to be an ideal form of learning to study savings-upon-relearning, as participants can rapidly learn to compensate for altered visual feedback while also decaying fully to baseline behavior within a single experimental session. Following this “forgetting” of the motor memory, participants can then be re-exposed to the same visual perturbation; savings would be evident if they compensate for the perturbation faster during re-exposure compared to the first learning event. This has been a conundrum for models of sensorimotor adaptation that function solely on sensory prediction errors, as the error size is the same for both learning episodes. If learning was only driven by these errors, it should proceed at the same rate both times. Here we examine the idea that this faster relearning comes from outside of the motor system and is not driven by sensory prediction errors, but rather an impetus to restore good task performance. Specifically, the results indicate that savings comes about because participants learn to implement a cognitive aiming strategy that helps them hit the target again. The difference in the rate of behavioral change arises because participants require time to develop the strategy when first encountering the altered visual feedback, but can then immediately implement it upon re-encountering the altered feedback.
In Chapter 2, we attempted to isolate the effects of error-based adaptation with a novel experimental manipulation. Participants were exposed to altered visual feedback and, unlike traditional adaptation studies, were fully informed of the nature of this alteration and explicitly told to ignore it. The specific visual feedback manipulation employed is known as a “visual error clamp,” where the visual cursor is set to a fixed heading angle. This means that no matter where the participant moves in the workspace, the feedback will always move in this direction instead of the direction of movement. We carefully manipulated the offset of the heading angle for this feedback relative to the direction participants were reaching in order to induce task-irrelevant sensory prediction errors. The only reason participants should adjust for these error clamps is if error-based learning is taking place given that they were told to ignore the feedback. We observed very robust adaptation in response to this manipulation. Surprisingly, the adaptation was consistent with that observed in typical adaptation studies in every way but one: the size of the change in behavior was not related to the size of the error. This is potentially a substantial challenge for theories of error-based adaptation, as they predict that there is either a linear or curvilinear relationship between error size and the magnitude of the adaptive response.
In Chapter 3, we explore the consequences that intrinsic biases have on visuomotor adaptation studies. When participants move without visual feedback, they often exhibit individual biases in the direction of their reaches. Here we show that there is a systematic bias for all participants, varying with the reach direction, and that it cannot be fully eliminated through visuomotor adaptation. This is because learning at any given reach direction is not fully independent of learning in other directions given that learning generalizes locally in the workspace. Furthermore if feedback is removed (a common manipulation in adaptation tasks), participants will drift back to this bias over time. If unaccounted for, this systematic bias (or its re-emergence) can be misinterpreted as a learning effect in adaptation tasks. We outline a few experimental and analytical techniques that can help account for this bias in these tasks so that future researchers can study adaptation without this contaminant.
Taken together, these studies show that many different processes contribute to the behavior of participants in sensorimotor adaptation tasks. These processes function with considerable independence and affect behavior in response to distinct stimuli. We have made an attempt to dissociate these processes primarily at a psychological level, a critical step for the investigation of the neural underpinnings of such processes.