An electromyography signals-based human-robot collaboration method for human skill learning and imitation

Abstract

Human beings have strong perception and decision-making ability. In human-human collaboration, the follower can adjust the motion speed by perceiving the change of external force information, and cooperate with the leader to complete the collaboration task. Transferring human collaboration skills to robots will help improve the flexibility and compliance of human-robot collaboration. For this purpose, an Electromyography (EMG) signals-based human-robot collaboration method is proposed. In the collaboration method, the human arm force, obtained by EMG signals and joint angles, is taken as the interface of human-robot interaction, and the robot is controlled to complete the collaboration task by learning the speed adjustment skills of human tutors. To reduce the inaccuracy and instability of human-robot interaction caused by the fluctuation of EMG signals, the adaptive data correction unit based on tremor information and the input-output control unit based on Naive Bayes are added to the Long Short-Term Memory (LSTM) neural network, and the parallel network structure is used to estimate the three-dimensional arm force. Aiming at the problem that the traditional collaborative control model cannot balance the accuracy and rapidity, a multi-model Gaussian process regression algorithm is used to capture the collaboration skills from inaccurate human-human demonstrations in the way of probability estimation. Finally, taking peg-in-hole assembly as an example, the proposed human-robot collaboration method is verified. Compared with the traditional adaptive admittance control, the proposed collaboration method shows better interaction performance, and its assembly time and assembly success rate are improved.