AI-Powered VR Rehabilitation Using EMG Signals: A Systematic Review

The convergence of artificial intelligence (AI), virtual reality (VR), and electromyography (EMG) has enabled the development of next-generation rehabilitation systems that are adaptive, data-driven, and patient-centered. This systematic review synthesizes recent advances (2020–2026) in integrated AI-powered VR rehabilitation frameworks utilizing EMG signals. Following a structured review methodology, 17 studies were analyzed with respect to machine learning techniques, system architectures, sensing modalities, and evaluation practices. Several studies reported high within-subject classification performance, whereas subject-independent generalization remained more challenging where it was directly evaluated. However, the reported performance estimates were obtained from heterogeneous datasets, participant groups, movement sets, sensor configurations, and validation protocols and should therefore be interpreted as study-specific feasibility results rather than directly compared across studies. The integration of multimodal biosignals enables real-time assessment of user states, including fatigue and stress, facilitating closed-loop adaptive rehabilitation. Despite these technical advancements, only a minority of studies involved clinical populations. Key challenges include inter-subject variability, small sample sizes, lack of longitudinal evaluation, and limited generalizability of trained models. Emerging approaches such as transfer learning, multimodal sensor fusion, and real-time adaptive control show promise in addressing these limitations.