Human Upper Limb Trajectory Prediction Based on MSFA-ESN

With the advent of Industry 4.0, human upper limb trajectory prediction has become a key research area in robotics, particularly in human-robot collaboration. To address this, this paper proposes the multi-scale feature fusion and adaptive activation function echo state network(MSFA-ESN). MSFA-ESN leverages multi-scale feature fusion(MSFF) to independently capture short-term dynamics and long-term trends in motion trajectories, combining them through weighted fusion for enhanced accuracy. Additionally, an adaptive activation function(AAF) adjusts the network’s nonlinear transformation capabilities to better handle complex, multi-scale time series data. Experiments on upper limb trajectory datasets demonstrate that MSFA-ESN outperforms traditional echo state network(ESN) and long short-term memory(LSTM) models in prediction accuracy while maintaining high computational efficiency. This approach offers a robust solution for trajectory prediction in dynamic, nonlinear environments, with significant potential for real-world human-robot collaboration.