Using wearable physiological sensors to predict energy expenditure

Area of Science:

• Biomechanics
• Robotics
• Wearable Technology

Background:

• Evaluating lower-limb assistive robotic devices often relies on measuring user energy cost.
• Indirect calorimetry, the standard method, is unsuitable for real-time and long-term energy cost assessment.

Purpose of the Study:

• To develop a method using wearable sensor data to predict user energy cost with improved temporal resolution and reduced variability.
• To enable more effective real-time evaluation of lower-limb assistive robotic devices.

Main Methods:

• Collected physiological (heart rate, electrodermal activity, skin temperature) and mechanical (EMG, accelerometry) data from healthy subjects walking on a treadmill.
• Established ground truth energy cost using indirect calorimetry.
• Developed a multiple linear regression model incorporating processed sensor data (accelerometer magnitudes, linear envelope EMG, averaged signals).

Main Results:

• Processed mechanical signals, particularly when averaged, showed improved correlation with energy cost.
• A multiple linear regression model combining physiological and mechanical data accurately predicted energy cost across various walking conditions.
• The sensor-based prediction demonstrated superior temporal resolution and less variability compared to breath measurements.

Conclusions:

• Wearable sensor data can reliably predict user energy cost during locomotion.
• This approach offers a portable and effective alternative to indirect calorimetry for real-world assistive robotic device evaluation.
• Improved energy cost estimation can significantly enhance the design, control, and assessment of lower-limb assistive robots.