Wearable- and Sensor-Based Artificial Intelligence in Exercise Science: IMUs, Heart-Rate Variability, and EMG for Real-Time Coaching and Fatigue/Recovery Prediction

Abstract

Introduction and Objective. Wearable sensors and artificial intelligence (AI) are increasingly being used to 
monitor human movement, physiological responses, fatigue and recovery during exercise. Inertial 
measurement units (IMUs), heart-rate variability (HRV) and surface electromyography (sEMG) provide 
complementary information about movement, autonomic regulation and neuromuscular activity. This 
narrative review examines how machine-learning (ML) methods can integrate these signals to support 
real-time coaching, fatigue monitoring and recovery prediction. Material and method. The review 
considered applications involving wearable movement sensors, cardiovascular monitoring, EMG, 
training-load variables and individualized physiological responses. Particular attention was given to signal 
preprocessing, feature extraction, multimodal data fusion, temporal modelling, individualized baselines and 
real-time decision support. Results. IMU-based systems can identify movement patterns and changes in 
exercise technique, while HRV provides longitudinal information related to cardiovascular and autonomic 
responses. EMG can contribute information about muscle activation and neuromuscular fatigue. ML can 
combine these heterogeneous signals to estimate exercise intensity, detect changes in movement or fatigue, 
and support individualized recovery assessment. However, performance varies according to sensor quality, 
population characteristics, experimental conditions and validation strategy. Motion artifacts, individual 
variability, limited external validation, laboratory-to-field differences and model interpretability remain 
important limitations. Conclusions. Wearable AI represents a promising decision-support technology for 
exercise science, particularly when multiple sensor modalities are integrated longitudinally. Current 
evidence supports its use for monitoring and individualized feedback rather than autonomous high-stakes 
decision-making. Future research should prioritize multimodal validation, real-world longitudinal studies, 
transparent algorithms and integration with professional coaching and clinical expertise.