An Attention-Based CNN-LSTM Model with Limb Synergy for Joint Angles Prediction

Estimation of lower limb movement is crucial in exoskeleton-assisted gait rehabilitation which can reduce the training load by recognizing the movement intention of patients, so as to realize the adaptive and transparent robotic assistance. Human locomotion has inherent synergies and coordination, and the dynamic mapping of the upper and lower limbs is beneficial to improve the prediction accuracy. Current prediction methods do not fully consider the correlation of gait data in time and space, resulting in a large amount of redundant data and low prediction accuracy. This paper proposes a gait trajectory prediction method based on attention-based CNN-LSTM model, which predicts the human knee/ankle joint trajectory based on upper and lower limb collaborative data. The attention mechanism is applied to determine which dimensions are essential in estimation of lower limb movement, so the accuracy can be improved by adopting key elements. Results show that, within a predicted horizon of 60 ms, prediction RMSE is as low as 0.317 degrees.