Predicting reach goal in a continuous workspace for command of a brain-controlled upper-limb neuroprosthesis

A controller for an upper-limb functional electrical stimulation system could use intended reach goal to generate a set of stimulation patterns that would move the hand to the desired location via a reasonably naturalistic velocity profile. Although discrete classifiers have been successfully used to predict movement goal from a fixed number of possible reach locations using neural activity recorded during movement planning, practical implementation of this paradigm for use in upper-limb neuroprostheses requires the ability to predict a reach goal anywhere within a person´s workspace. Using neural data collected from monkeys during brain-controlled movements of a virtual cursor and robotic arm, we evaluated how well the direction versus magnitude of the final movement goal could be predicted from varying lengths of neural data collected after the target appeared. Although a majority of the channels were significantly modulated with intended movement direction, only 10-20% showed any significant modulation related to the magnitude of the movement goal. We propose a method of trajectory generation that could use the more reliably encoded directional information in the neural activity to control both magnitude and direction of a goal oriented reaching movement