Integrating Wearable Sensor Data and Artificial Intelligence for Real-Time Yoga Performance Feedback: A Systematic Review

Abstract

Wearable sensors and artificial intelligence (AI) are increasingly combined to support unsupervised yoga practice by detecting posture, movement and physiological state and by returning corrective feedback in real time. This systematic literature review synthesizes peer-reviewed research at the intersection of wearable sensing, AI/machine learning, and real-time performance feedback for yoga and the closely related movement domains. Following the PRIZMA 2020 methodology, four databases (Scopus, Web of Science, IEEE Xplore, and PubMed) were searched, yielding 812 records, after removing duplicates and screening by title, abstract and full text, 45 studies were analyzed with respect to sensor modality (inertial measurable units, photoplethysmography, electromyography, pressure/textile sensors), AI technique (convolutional and recurrent neural network, tree-based classifiers, Bayesian and hybrid models), feedback channel (visual, auditory, haptic), and reported performance. Findings show that inertial sensing combined with deep learning achieves high posture-classification accuracy (typing 90-99%), that multimodal fusion of movement and physiological signals is emerging for breath- and stress-aware feedback and that genuinely low-latency, on-device (edge) inference remains uncommon. Key gap includes small and non-diverse datasets, limited sensor placement practicality for long-term use, inconsistent real-time evaluation protocols, and minimal user-experience validation. The review proposes research agenda centered on lightweight edge AI, multimodal fusion, standardized benchmarking, and longitudinal usability studies.