The Second Skin: A Wearable Sensor Suite That Enables Real-Time Human Biomechanics Tracking Through Deep Learning.

Area of Science:

• Biomechanics
• Wearable Technology
• Machine Learning

Background:

• Accurate real-time human motion analysis is crucial for biomechanics research, biofeedback, and exoskeleton control.
• Existing methods often struggle with task and user independence, limiting their real-world applicability.

Purpose of the Study:

• To investigate a task-independent, user-independent method for precise real-time estimation of lower-body joint kinematics and kinetics.
• To establish deep learning as a viable sensing approach for generalizable biomechanical analysis.

Main Methods:

• Developed a sensing suit with inertial measurement units (IMUs) and pressure insoles.
• Collected data from 10 participants performing 33 diverse tasks.
• Trained deep learning models to estimate joint kinematics and dynamics from sensor data.

Main Results:

• Deep learning models achieved significantly lower root-mean-squared errors (RMSE) for angle estimation compared to analytical methods across lower back, hip, knee, and ankle joints.
• Normalized moment estimation RMSEs were also substantially improved using the deep learning approach.
• Model performance demonstrated generalization across different users and tasks.

Conclusions:

• Deep learning offers a viable and accurate sensing approach for real-time biomechanical analysis.
• The developed method is comparable to state-of-the-art wearable sensing systems.
• This work has significant potential for advancing biofeedback systems and wearable robot control.