Compensating for delays in brain-machine interfaces by decoding intended future movement

Typically, brain-machine interfaces that enable the control of a prosthetic arm work by decoding a subjects' intended hand position or velocity and using a controller to move the arm accordingly. Researchers taking this approach often choose to decode the subjects' desired arm state in the present moment, which causes the prosthetic arm to lag behind the state desired by the user, as the dynamics of the arm (and other control delays) constrain how quickly the controller can change the arm's state. We tested the hypothesis that decoding the subjects' intended future movements would mitigate this lag and improve BMI performance. Offline results show that predictions of future movement (≤ 200ms) can be made with essentially the same accuracy as predictions of present movement. Online results from one monkey show that performance increases as a function of the future prediction time lead, reaching optimum performance at a time lead equal to the delay inherent in the controlled system.