Embodied artificial intelligence in robotic orthopaedic surgery: from intelligent assistance to adaptive surgical autonomy.

Despite remarkable advances in robotic orthopaedic surgery over the past three decades, current commercial platforms remain precision-assistance technologies rather than truly adaptive intelligent systems. Embodied Artificial Intelligence (Embodied AI) represents a paradigm shift by enabling robots to perceive, reason, learn, and interact continuously with their physical environment through integrated perception-action loops. This structured narrative review examines the emerging role of Embodied AI in orthopaedic robotic surgery, highlights its conceptual differences from existing platforms, and explores its potential to transform surgical planning, intraoperative decision-making, and postoperative care. Six major commercial systems-MAKO (Stryker), ROSA (Zimmer Biomet), CORI/NAVIO (Smith+Nephew), VELYS (DePuy Synthes), ExcelsiusGPS (Globus Medical), and TSolution One (THINK Surgical)-are evaluated with respect to autonomy, adaptive capability, and clinical functionality. Recent advances in multimodal sensor fusion, computer vision, three-dimensional reconstruction, dynamic scene understanding, force and haptic sensing, foundation models, large language models, reinforcement learning, digital twins, and continual learning are synthesised with emphasis on translational relevance across arthroplasty, trauma, spine, and orthopaedic oncology. Critical challenges surrounding safety validation, regulatory governance, cybersecurity, real-time computational performance, ethical considerations, and surgeon-AI collaboration are discussed. Current evidence confirms that commercially available orthopaedic robotic systems are not fully embodied or autonomous, despite documented improvements in surgical accuracy, implant positioning, and workflow efficiency. Safe clinical implementation will require integration of Embodied AI with patient-specific digital twins, graduated autonomy frameworks, adaptive learning architectures, and human-in-command safety systems, supported by robust multicentre validation, standardised performance benchmarks, comprehensive regulatory frameworks, and multidisciplinary collaboration.