Feng-Yan Liang1, Fei Gao2, Wei-Hsin Liao3. 1. Department of Biomedical Engineering, The Chinese University of Hong Kong, Shatin, Hong Kong, China. 2. Department of Mechanical and Automation Engineering, The Chinese University of Hong Kong, Shatin, Hong Kong, China. 3. Department of Mechanical and Automation Engineering, The Chinese University of Hong Kong, Shatin, Hong Kong, China. Electronic address: whliao@cuhk.edu.hk.
Abstract
BACKGROUND: Lower limb assistive devices have been developed to help amputees or stroke patients. To precisely mimic the required function, researchers focused on how to estimate/predict the required knee angle for knee devices. RESEARCH QUESTION: The objective is to estimate the motion of the human knee joint during walking using the kinematics of wearer's thigh measured by a single Inertial Measurement Unit (IMU). The hypotheses are that the proposed method can precisely estimate knee angle and have good universality on different subjects, speeds and strides. METHOD: 8 healthy subjects walked on the level ground at three different speeds. An IMU mounted on the thigh was employed to collect the kinematic information of the thigh including angular velocities and accelerations. A long short-term memory (LSTM) neural network model was adopted to model intra-limb synergy between the motion of thigh and the knee joint. Such that with the trained LSTM model, the knee angle can be precisely predicted. RESULTS: Compared with the existing studies, the proposed approach based on an LSTM model has better estimation performance. The average RMSE for eight subjects can be limited to 3.89°. The proposed method has speed and stride adaptability. SIGNIFICANCE: The proposed method is promising to generate a desired and harmonious knee trajectory in line with thigh motion for assistive robotic devices.
BACKGROUND: Lower limb assistive devices have been developed to help amputees or strokepatients. To precisely mimic the required function, researchers focused on how to estimate/predict the required knee angle for knee devices. RESEARCH QUESTION: The objective is to estimate the motion of the human knee joint during walking using the kinematics of wearer's thigh measured by a single Inertial Measurement Unit (IMU). The hypotheses are that the proposed method can precisely estimate knee angle and have good universality on different subjects, speeds and strides. METHOD: 8 healthy subjects walked on the level ground at three different speeds. An IMU mounted on the thigh was employed to collect the kinematic information of the thigh including angular velocities and accelerations. A long short-term memory (LSTM) neural network model was adopted to model intra-limb synergy between the motion of thigh and the knee joint. Such that with the trained LSTM model, the knee angle can be precisely predicted. RESULTS: Compared with the existing studies, the proposed approach based on an LSTM model has better estimation performance. The average RMSE for eight subjects can be limited to 3.89°. The proposed method has speed and stride adaptability. SIGNIFICANCE: The proposed method is promising to generate a desired and harmonious knee trajectory in line with thigh motion for assistive robotic devices.