Classification of upper arm EMG signals during object-specific grasp.

Electromyographic (EMG) signals can represent an interesting solution to control artificial hands because they are easy to record and can allow the user to control different robotic systems. However, after limb amputation the 'homologous' muscles are no more available to control the prosthetic device and for this reason complex pattern recognition approaches have to be developed to extract the voluntary commands by the user. This makes the control strategy less natural and acceptable and asks for alternative approaches. At the same time, it has been recently shown that (in monkeys) it is possible to discriminate grasping tasks just analyzing the activation onset/offset of upper limb muscles during the reaching phase. This kind of information can be very interesting because it can allow the development of a natural EMG-based control strategy based on the natural muscular activities selected by the central nervous system. In this paper, preliminary experiments have been carried out in order to verify whether these results can be confirmed also in human beings. In particular, a support vector machine (SVM) based pattern recognition algorithm has been developed and used for the prediction of grip types from the EMG recorded from proximal and distal muscles during reach to grasp movements of three able bodied subjects.