Controlling a Dexterous Hand From Neural Signals

A BMI controls motion of an avatar hand to 8 different
targets along 4 multi-joint basis functions.

We currently are exploring dexterous control of an avatar hand through a brain-computer interface. To address the challenge of controlling the complexity of the hand, we emphasize a variety of approaches beyond fixed, linear decoding of neural activity. Instead, we used a hybrid controller with active dimension selection (ADS) that combined selection of the appropriate dimension with velocity control along that dimension. The four dimensions of control corresponded to four common grasps: precision pinch, three-finger tripod, power grasp, and thumb/little finger opposition.

Download a sample movie [zip].

The brain-machine interface controlled the hand
on the left to match the desired target hand
posture on the right.

« back to all projects