Robust Machine Learning Mapping of sEMG Signals to Future Actuator Commands in Biomechatronic Devices

• Published in: Journal of bionics engineering 2024, Vol.21 (1), p.270-287
• Main Authors: Nasr, Ali, Bell, Sydney, Whittaker, Rachel L., Dickerson, Clark R., McPhee, John
• Format: Article
• Language: English
• Subjects: Artificial Intelligence; Biochemical Engineering; Bioinformatics; Biomaterials; Biomedical Engineering and Bioengineering; Biomedical Engineering/Biotechnology; Engineering; Research Article
• ISSN: 1672-6529; 2543-2141
• DOI: 10.1007/s42235-023-00453-8

A machine learning model for regression of interrupted Surface Electromyography (sEMG) signals to future control-oriented signals (e.g., robot’s joint angle and assistive torque) of an active biomechatronic device for high-level myoelectric-based hierarchical control is proposed. A Recurrent Neural Network (RNN) was trained using output data, initially obtained from offline optimization of the biomechatronic (human–robot) device and shifted by the prediction horizon. The input of the RNN consisted of interrupted sEMG signals (to mimic signal disconnections) and previous kinematic signals of the assistive system. The RNN with a 0.1-s prediction horizon could predict the control-oriented joint angle and assistive torque with 92% and 86.5% regression accuracy, respectively, for the test dataset. This proposed approach permits a fast, predictive, and direct estimation of control-oriented signals instead of an iterative process that optimizes assistive torque in the inverse dynamic simulation of a multibody human–robot system. Training with these interrupted input signals significantly improves the regression accuracy in the case of sEMG signal disconnection. This Robust Predictive Control-oriented Machine Learning (Robust-MuscleNET) model can support volitional high-level myoelectric-based control of biomechatronic devices, such as exoskeletons, prostheses, and assistive/resistive robots. Future work should study the application to prosthesis control as well as the repeatability of the high-level controller with electrode shift. The low-level hierarchical controller that manages the human–robot interaction, the assistance/resistance strategy, and the actuator coordination should also be studied.