Abstract
Recently, robot arm for humanoid has been widely developed for medical, welfare, education use. Mainly, surface electromyogram (SEMG) data has been used for hand motions classification. Motion classification of hand using SEMG shows high classification rate. Though big motion of arms can be classified in high agreement rate, detailed motions including fingers and forearm are difficult to classify. Therefore, we thought that addition of a new parameter is effective for improvement of classification rate. So, we adopted forearm shape changes added to SEMG data. We got SEMG signals from six electrodes and shape changes from six strain gages. We classified nine motions; all fingers flexion, all fingers extension, four fingers flexion, pronation, supination, palmar flexion, dorsal flexion, radial flexion, ulnar flexion. And we applied SVM to classify the motions and evaluated agreement rate. As a result, we showed approximately 54% of agreement rate by only using SEMG data and approximately 81% of agreement rate by only using strain gauges data. We could show agreement rate of more than 98% by using both of SEMG data and strain gages data.
