Gait Trajectory Prediction of Lower Limb Exoskeleton in Multiple Conditions Using a Convolutional Transformer Network and Contrastive Learning

Gait trajectory prediction is vital for the effective operation of lower limb exoskeletons, particularly in multiple environmental conditions such as level ground, ramps, and stairs. This paper proposes an innovative approach using a Convolutional Transformer Network (CTN) integrated with contrastive learning to enhance the prediction of gait intentions for lower limb exoskeletons. By leveraging the local feature extraction of Convolutional Neural Networks (CNNs) and the long-range dependency capabilities of Transformers, the model is capable of accurately forecasting gait trajectories. The proposed method was evaluated using gait data from inertial measurement units (IMUs) and goniometers, capturing both accelerometer and gyroscope readings from key limb segments. Results demonstrated improved performance in terms of prediction accuracy and robustness, outperforming traditional deep learning models such as CNNs and RNNs across various walking conditions. The model achieved on average 0.9565 for precision and 0.9558 for F1 score, demonstrating its potential for real-world applications in neurorehabilitation and assistive devices.