Continuous prediction of multi-joint angles based on Informer model using novel fused features

Abstract Continuous prediction of multi-joint angles based on surface electromyography (sEMG) has the potential to enhance the human-machine interaction performance. However, achieving accurate angle prediction remains a significant challenge during various motions. In this study, a method for extracting gait feature, named Gait-Time, is proposed, and different models for predicting multi-joint angles are established and compared, including Informer, gated recurrent unit, one-dimensional convolutional neural network, long short term memory (LSTM), as well as two hybrid models, LSTM-BP and LSTM-Attention. The results indicate that Informer model, utilizing left lower limb sEMG fused with Gait-Time features as input, outperforms the other mentioned models, demonstrating promising outcomes with the average correlation coefficient values exceeding 0.99 for all three joints. Furthermore, the mean absolute error (MAE) and root mean square error (RMSE) for the ankle and hip joints average below 1°, while for the knee joint, both the average MAE and RMSE fell within a range of 2°. It can be concluded that the accuracy of Informer model is adequate to support the continuous prediction of multi-joint angles in human motion scenarios.