Neurorehabilitation with robot has a potential to improve the motor function of post stroke patients. We are developing a rehabilitation robot to provide hand open and close training for patients. The robot is triggered by upper limb forearm surface electromyography(EMG). However, owing to complex arrangement of the muscle layers and muscle cross talk during the both open and close motions of the hand, traditional classification techniques of motion intention does not perform successfully in its implementation. Further, presence of motion artifacts such as natural supination/pronation, wrist flexion/extension also has become a challenge to overcome in classifying hand open/close and relax states successfully. Thus in this study, we propose a real-time classification method that uses OneClass SVM to extract hand open/close and relax motions and enables to cater for the individual differences of the different users. In the evaluation experiments with 5 different subjects, the proposed method could successfully classify motion intention of hand open/close and relax in the presence of different motion artifacts.