An Online Human Dynamic Arm Strength Perception Method Based on Surface Electromyography Signals for Human–Robot Collaboration

In human–robot collaboration, the perception of human arm strength is significant for robots to correctly understand human work intentions. Current perception methods lack a standardized scheme for online feature extraction and suffer from high model complexity and poor perceptual real-time performance. In this article, an online human dynamic arm strength perception method based on surface electromyography (sEMG) signals is proposed. This method studies online adaptive sEMG feature extraction and lightweight arm strength perception, aiming to help robots establish a high-precision and real-time arm strength perception interface. Specifically, an adaptive sEMG amplitude extraction method based on iteration learning algorithm and fractal dimension is first proposed to ensure the adequacy of online sEMG information extraction. Then, to perceive changes in human arm strength from extracted sEMG amplitude features, a dynamic arm strength perception model based on parallel convolutional structure, lightweight temporal capture module, and attention-weighted strategy is proposed, which can improve the strength perception accuracy of robots and reduce the complexity of perception model. The experimental results show that compared to representative methods, the proposed online perception method achieves improvements in perception accuracy and real-time performance. The human–robot collaborative sawing experiments verify the engineering feasibility of the proposed method.