Affiliations: [a] Department of Electrical Measurements, Lund Institute of Technology, P.O. Box 118, SE-221 00 Lund, Sweden | [b] Department of Physiological Sciences, Lund University, Lund, Sweden
Abstract: A myoelectric control system for prostheses was developed and evaluated on six healthy subjects. These were able to control a computer-animated hand in real-time with a 20 Hz update rate. A data glove, equipped with joint angle sensors, was used to train the system and to evaluate the continuous predictions of joint angles. A linear envelop filter was used for EMG signal pre-processing and the recognition of muscle patterns was carried out with local approximation using the “lazy learning” algorithm. Furthermore, an on-line learning was used to provide feedback to the subjects. The results show that the subjects increased their performance during the experiment and all subjects performed eight or more movements with 100% accuracy in their last recording session. The final median delay for the predicted hand joint positions, compared with the recorded, was in the range of 50 to 100 ms. Off-line evaluation has earlier been done on amputees while using a data glove on the contralateral hand. The real-time control system outlined in this paper offers an effective myoelectric prosthesis control that is suitable for a miniaturized low cost implementation. This paper is a part of a ongoing development and refinement of hand prosthesis carried out within the Artificial Hand Project.
Keywords: EMG, myoelectric, hand prosthesis, recognition, real-time control, local approximation
