A Novel Estimation Method of Lower Limb Continuous Motion by sEMG Signal Based on BiLSTM Neural Network

In human-machine collaboration (HMC) control for lower limb exoskeleton robot (LLER), accurately recognizing human motion intent is crucial for ensuring safe system operation. To enhance the recognition accuracy of multi-joint movement intentions, this study collects multi-channel surface electromyography (sEMG) signals from subjects during cycling and treadmill walking tasks and applies advanced signal processing techniques to preprocess the raw data. Based on this, a bidirectional long short-term memory (BiLSTM) neural network model is developed to estimate the angles of multiple lower limb joints. Compared to traditional models such as convolutional neural networks (CNN), backpropagation neural networks (BPNN), and radial basis function neural networks (RBFNN), the root mean square errors (RMSE) values achieved by the BiLSTM model in predicting lower limb joint angles are 2.9988°, 3.0981°, and 2.2297°,with coefficient of determination (R²) values of 0.8776, 0.9003 and 0.9090, respectively. These results highlight the BiLSTM model’s high efficiency and accuracy in recognizing multiple joint movement intentions, providing a solid theoretical foundation for the development of HMC control in LLER systems.