The next generation machines are expected to have more close physical interactions with human, so it is important that the machine should sense human’s motion and intention for natural and safe interactions. While conventional sensors such as force transducer, vision-, and position-sensors have been widely used, these sensors have inherent delay. The advantage of using bio-signals is that using the signals which precede the movement can compensate the delay, but the research using bio-signals instead of these sensors has been in progress.
This thesis focuses on the possibility evaluation of a real-time motion prediction method using surface electromyography (sEMG) signals for a physical human-machine interaction (pHMI). A real-time motion prediction method was implemented by using sEMG signals from 5 channels and an artificial neural network (ANN) algorithm. Interaction experiments were performed with two elements: the existence of physical contacts and the post-processing of predicted motion data. The results showed that the machine can sense human movements from sEMG signals, and the reaction of the machine can be almost synchronous with the human. The results also represent that the performance of using sEMG were not better than the cases using position sensors but reasonable for pHMI. The post-processing did not affect significantly the performances of the interaction, but the manipulator movements became stable. The interaction using sEMG will be more natural with further studies on the theoretical model of the combined interface of a human and a machine.