A Knowledge Transfer-Based Personalized Human-Robot Interaction Control Method for Lower Limb Exoskeletons

Accurate intent recognition by patients while wearing exoskeletons is crucial during their rehabilitation exercises. In this article, a transfer learning framework for human-robot interaction (EMGTnet-KTD) is proposed to predict human movement intentions in human-robot interactions through surface electromyography (sEMG) signals. EMGTnet-KTD consists of a pretrained EMGTnet model and a knowledge transfer module. First, EMGTnet is designed based on a Transformer network. A temporal and spatial domain feature fusion module has been introduced on top of the Transformer network, and the inputs have been reconfigured to enable it to utilize the relationship between before and after human actions. In addition, the knowledge transfer module is composed of a feature extraction layer, a noise reduction layer, and the personalized human lower limb dynamics controller. To evaluate the effectiveness of the proposed method, an experimental validation of our self-collected dataset from seven subjects is performed. The results show that our method achieves better results than other continuous motion prediction methods. Finally, to validate that the generation angle conforms to human physiology, walking experiments involving the use of an exoskeleton are conducted. The experiments demonstrate the effectiveness of the proposed framework and its implementability for exoskeletons.