Abnormal Gait Phase Recognition and Limb Angle Prediction in Lower-Limb Exoskeletons

Area of Science:

• Robotics
• Biomechanics
• Machine Learning

Background:

• Controlling lower-limb exoskeletons faces challenges in detecting abnormal gait phases and predicting lower-limb angles.
• Abnormal gaits like scissor, foot-drop, and staggering gait require accurate phase recognition for effective assistance.

Purpose of the Study:

• To develop and validate a robust method for abnormal gait phase detection and lower-limb angle prediction.
• To enhance the control capabilities of lower-limb exoskeletons by improving motion prediction accuracy.

Main Methods:

• Simulated three abnormal gaits: scissor, foot-drop, and staggering.
• Proposed a four-discrete-phase division (pre-swing, swing, swing termination, stance) for single-leg gait phase recognition.
• Utilized Convolutional Neural Network (CNN) and Support Vector Machine (SVM) for discrete phase recognition.
• Employed an adaptive frequency oscillator for continuous phase estimation.
• Developed an innovative input scheme integrating three-axis ankle joint angles and continuous gait phases for motion angle prediction.
• Evaluated prediction accuracy using a CNN-Long Short-Term Memory (LSTM) network.

Main Results:

• Successfully recognized discrete gait phases using CNN and SVM.
• Achieved continuous gait phase estimation via an adaptive frequency oscillator.
• The proposed information fusion scheme significantly improved lower-limb angle prediction accuracy.
• CNN-LSTM network demonstrated superior performance in predicting limb motion angles.

Conclusions:

• The proposed four-discrete-phase division enhances abnormal gait recognition.
• Integrating ankle joint angles and continuous gait phases is effective for limb angle prediction.
• The developed methods show promise for improving the control and functionality of lower-limb exoskeletons.