EMG-Based Continuous Motion Decoding of Upper Limb with Spiking Neural Network

Surface electromyography (EMG), generated during muscle activities of human beings, allows intuitive control for human-robot interaction to happen. Decoding human movement intention from EMG accurately and instantaneously is one of the most important parts of the whole control task. Spiking neural network (SNN) with spiking neurons is more computationally powerful than networks with non-spiking neurons and contains temporal information (time-dependency). Compared with discrete motion classification task, motion regression is more meaningful and helpful for the underlying applications including assisting human beings' activities of daily living (ADLs). We proposed a novel method deploying SNN in human motion regression task. An SNN is built to decode elbow joint angle from preprocessed surface EMG signals and achieved satisfying accuracy compared with long short-term memory. According to the experiment results, SNN is competent to decode motion information from surface EMG.