Neural Correlates of Control of a Kinematically Redundant Brain-Machine Interface

Abstract

Brain-machine interfaces (BMIs) use signals from the brain to control cursors or robotic arms, with potential applications for restoring the ability for users to interact with the physical world around them. BMIs that are kinematically redundant allow for many viable solutions for the same task. While natural motor control involves the coordinated movements of kinematically redundant limbs, it is unclear how the brain might control the redundant degrees of freedom (DOF) in a BMI. In this study, we analyze a previously collected dataset where a macaque controlled a 4 DOF virtual arm in 2D space. A Kalman filter was used to decode neural signals from motor cortices into the four joint angle velocities. The monkey was instructed to move the virtual arm from a center target to eight peripheral targets, distributed evenly around a circle in a self-initiated center-out task. The monkey was able to achieve high accuracy in the task in the first day, but reach times continued to decrease over learning and endpoint trajectories became more stereotyped. We found that the neural activity fired in more correlated patterns over days with increased firing rates, suggesting a consolidation of neural activity into a high-level representation of the joint angles, optimizing endpoint control.