Synergy-based knee angle estimation using kinematics of thigh.

Area of Science:

• Biomedical Engineering
• Robotics
• Machine Learning

Background:

• Lower limb assistive devices are crucial for amputees and stroke patients.
• Accurate prediction of knee joint motion is essential for effective device function.
• Existing methods require precise estimation of knee angles for optimal performance.

Purpose of the Study:

• To estimate human knee joint motion during walking using thigh kinematics from a single Inertial Measurement Unit (IMU).
• To test the hypothesis that the proposed method offers precise knee angle estimation and universality across different subjects, speeds, and strides.

Main Methods:

• Collected kinematic data (angular velocities, accelerations) from a thigh-mounted IMU on 8 healthy subjects walking at varying speeds.
• Utilized a long short-term memory (LSTM) neural network to model the relationship between thigh motion and knee joint kinematics.
• Trained the LSTM model to predict knee angles based on thigh movement data.

Main Results:

• The LSTM-based approach demonstrated superior estimation performance compared to existing methods.
• Achieved an average Root Mean Square Error (RMSE) of 3.89° across all subjects.
• The method showed adaptability to different walking speeds and stride patterns.

Conclusions:

• The proposed IMU-based LSTM model accurately predicts knee joint angles during walking.
• This method holds significant promise for developing advanced assistive robotic devices.
• It enables the generation of harmonious knee trajectories synchronized with thigh motion for improved user assistance.