Optimal design and control of a hand exoskeleton

This paper deals with the optimal design and control of an exoskeletal robot. First, the motion data from the fingers of a normal subject was captured by a vision system. As the human finger joints cannot be modeled by single revolute joints due to changing instantaneous centre of rotation, we have used 4-bar mechanisms to model each joint. Optimal 4-bars have been designed using genetic algorithms, by minimizing the error between a coupler point and points traced by the finger links. It is shown that the designed 4-bars can accurately track the motion of the human fingers. The exoskeleton is controlled by using the EMG signals obtained from the subject´s muscles. The relation between the EMG and finger motion is first learned, using a neural net. Based on the learned parameters, the subjects EMG signal is used to control a simulation of the exoskeleton joint motion. A comparison between Recurrent Neural Network and Multi Layer Perceptron for classifying and mapping the EMG to finger position was also carried out.