Machine Learning-Driven Muscle Fatigue Estimation in Resistance Training with Assistive Robotics

Area of Science:

• Biomechanics
• Exercise Physiology
• Machine Learning

Background:

• Monitoring muscle fatigue is crucial for safe and effective resistance training.
• Current methods like EMG, IMU, and RPE have limitations, especially in automated or unsupervised settings.

Purpose of the Study:

• To develop a machine learning model for predicting ratings of perceived exertion (RPE) directly from force-time data during resistance exercise.
• To assess the relationship between biomechanical features and fatigue progression.

Main Methods:

• Thirty-two male participants performed isokinetic bench press sets at a 7RM load.
• Force-time data and RPE were recorded; biomechanical and engineered features were extracted.
• A Random Forest model was trained to predict RPE from these features.

Main Results:

• Muscle fatigue progression, not absolute force, strongly correlates with RPE.
• Engineered features significantly enhanced predictive accuracy.
• The Random Forest model achieved >93% accuracy within ±1 RPE unit.

Conclusions:

• A machine learning approach using force-time data can accurately predict RPE during resistance training.
• This method offers potential for integration into intelligent exercise machines for automated load adjustment.
• Applications include enhancing athletic training and rehabilitation programs.