Adaptation for Brain-Computer Interfaces

Abstract

Brain-computer interface (BCI) technology is evolving rapidly, and shows promise for restoring and even augmenting dextrous control of movement. Recent studies have demonstrated control of robotic arms and computer cursors for basic positioning and manipulation tasks, using signals recorded from cortex. It is clear, however, that substantial improvements are required to achieve the level of reliability and dexterity to enable clinical translation. BCIs present an instance of the dual control problem, wherein a controller is challenged with two distinct but inseparable problems: identifying a novel system, and concurrently driving that system to a desired state with minimum effort. This thesis addresses basic questions about how the brain adapts to novel movement control tasks. Based on models validated in human-computer interface (HCI) studies, we first develop a performance metric for comparing manual and brain control of positioning tasks. Next, we explore existing HCI study results on control gain selection for novel interfaces, yielding insights about gain selection consequences which are conserved across interface modalities. Finally, we identify a simple control policy structure which is evident during both brain and manual control. This yields the finding that the brain acts upon both delayed and predicted task state information in its control policies. Together, these findings enable more standardized investigation of BCI performance, and open new possibilities for performance improvement.