An EMG controlled robotic manipulator using neural networks

This paper proposes an adaptive human-robot interface using a statistical neural network which consists of a forearm controller and an upper arm controller. The forearm controller selects an active joint out of three joint degrees of freedom, and controls its driving speed or grip force according to EMG signals measured from a human operator. The upper arm controller controls the joint angle of the upper arm according to the position of the operator's wrist joint as measured by a 3D position sensor. Experiments have shown that the EMG patterns during forearm and hand movements can be classified with high accuracy using our network to be of use as an assistive device for a handicapped person.