Robust control of hand prosthesis using electromyography, accelerometers and gyros

Sammanfattning: The goals of this project are to investigate the viability of using the MyoTM armband developed by Thalmic labs as a tool for myoelectric control of a hand prosthesis as well as investigate whether the armbands ability to record its spatial position and orientation can be used to increase the performance of the prosthesis control program. Using the surface electrodes, gyroscopes and accelerometers of the MyoTM armband, a control system for myoelectric control of a hand prosthesis was created and tested on nine different users. The system recognizes eight different hand gestures and uses an artificial neural network as classification tool. The system was evaluated by two main criteria; gesture classification accuracy and response time. Offline testing showed an average gesture classification accuracy of over 90%, however, the accuracy was greatly reduced in online testing and a great variance in accuracy between different gestures was observed. Likely in large part caused by user inexperience with the system. The average system response time for all gestures was in the range of 100-200ms, again, with variations between gestures. Using positional data in addition to the myoelectric signals recorded by the armband showed no significant improvement in accuracy in best case scenarios and a greatly reduced accuracy in worst case scenarios, while the use of positional data still has the potential to improve the performance of a prosthesis control system the wrong approach to do so was likely chosen in this project and positional data was excluded in the final version of the program, using instead only myoelectric data for classification.