Artificial Intelligence-Driven Virtual Reality in Physical Rehabilitation: A Review of Machine-Learning Kinematics, Educational Scaffolding, and Treatment Adherence Strategies.

Physical rehabilitation increasingly depends on interventions that are intensive, personalized, and sustained outside the clinic. Yet contemporary rehabilitation systems face persistent barriers, including workforce shortages, geographic inequities, rising costs, fragmented follow-up, and poor adherence to home exercise programs. This comprehensive review examines how artificial intelligence (AI) and virtual reality (VR) can function together as a digital therapeutic framework for physical rehabilitation. The review argues that the core problem is not simply the absence of technology but the absence of continuous, meaningful supervision between clinic visits. AI-driven pose estimation, multimodal sensing, low-latency feedback, explainable analytics, and adaptive exercise progression allow rehabilitation programs to move from episodic observation to real-time, data-informed guidance. At the same time, principles of adult learning and motor learning help explain why immersion alone is not enough unless the system also teaches, motivates, and gradually transfers responsibility to the patient. Across musculoskeletal, neurological, cardiovascular, oncological, and chronic pain populations, the evidence suggests that VR-supported rehabilitation can improve engagement, exercise capacity, movement quality, and patient satisfaction, particularly when paired with personalized coaching and home-based monitoring. This paper therefore proposes AI-VR rehabilitation not as a replacement for clinicians, but as a clinically governed co-pilot that extends supervision, strengthens adherence, and expands equitable access to therapy.