Prediction of Lower Limb Exoskeleton Gait Trajectories Based on Plantar Pressure Information

Abstract

Accurate gait prediction plays a crucial role in the functionality of lower limb exoskeleton robots. In this research, we delve into the implementation and comparison of two distinct deep learning algorithms, Transformer and Long Short-Term Memory (LSTM), for predicting individual gait trajectories utilizing plantar pressure data. Our objective was to achieve high precision in predicting the gait movements of subjects wearing these exoskeletons. To this end, thirteen healthy subjects were outfitted with lower limb exoskeletons equipped with inertial sensors for gait data collection, while plantar pressure insoles embedded within the exoskeletons' feet gathered real-time foot pressure data. The performance of the models was compared using leave-one-out cross-validation, which highlighted the Transformer model's superior predictive capabilities over the LSTM, thereby emphasizing its effectiveness in gait trajectory prediction. The models' predictive accuracy was evaluated using the root mean square error (RMSE) at each time step. The LSTM model achieved an RMSE of 0.87°, while the Transformer model demonstrated an RMSE of 0.63°.