BiLSTM-KNN network for locomotion mode recognition in lower limb exoskeleton robot

Abstract

Abstract Lower limb exoskeleton robots have become a research hotspot among various enterprises, universities, and institutions. This paper proposes a deep learning network based on the recognition of five movement patterns of lower limb exoskeletons, including walking on level ground, going upstairs, going downstairs, uphill, and downhill. The network uses BiLSTM to capture the forward and backward relationships of time series data, which has better resolution than the original features. In addition, the network uses the KNN algorithm based on shortest-distance voting to improve the recognition accuracy. The network achieved a recognition rate of 99.18% for five motion patterns. In summary, the proposed network has excellent performance in terms of accuracy and generalization.