Synthetic Hamiltonian Energy Prediction for Motor Performance Assessment in Neurorehabilitation Procedures: A Machine Learning Approach with TimeGAN-Augmented Data

This study presents an assessment scheme for haptic interaction systems based on Hamiltonian energy prediction, which contributes to procedures applied to neurorehabilitation. It focuses on robotic systems involving human participation in the control loop, where uncertainty may compromise both stability and task performance. To address this, a regression-based model is proposed to predict total mechanical energy using the robot’s position and velocity signals during active interaction. Synthetic data generated via TimeGAN are used to enhance model generalization. Advanced machine learning techniques—particularly Gradient Boosting—demonstrate outstanding accuracy, achieving an MSE of 0.628×10−10 and R2=0.999976. These results validate the use of synthetic data and passive-mode-trained models for assessing motor performance in active settings. The method is applied to a patient diagnosed with Guillain-Barré Syndrome, using the Hamiltonian function to estimate energy during interaction and objectively assess motor performance changes. The results obtained show that our proposal is of great relevance since it solves a current field of opportunity in the area.