AI-Driven Rehabilitation Robotics

Abstract

Rehabilitation robotics has matured from simple motorised splints into adaptive, sensor-rich platforms capable of delivering high-dose, task-specific therapy, yet the addition of artificial intelligence to guide exercise selection, resistance modulation, and progress tracking remains unevenly adopted across clinical settings. We evaluated 210 AI-driven rehabilitation robotics programmes deployed or trialled across hospitals and rehabilitation centres affiliated with the Mediterranean Institute of Technology in Rome between 2015 and 2024, spanning five device categories: upper-limb exoskeletons, lower-limb gait trainers, hand and wrist rehabilitation robots, socially assistive companion robots, and multi-joint full-body platforms. An AI Rehabilitation Robotics Effectiveness Index (ARREI) was constructed from five sub-scores -- motor recovery gain, adaptive algorithm sophistication, patient engagement retention, therapist workflow compatibility, and deployment scalability -- with weights from regression against sustained clinical programme retention beyond 12 months. ARREI correlated with retention at r = +0.83 and discriminated retained from discontinued programmes with an AUC of 0.878. Lower-limb gait trainers scored highest (mean ARREI 0.816), while socially assistive robots trailed at 0.592. Only 34.3 percent of programmes exceeded the 0.75 ARREI threshold. Motor recovery gain carried the largest regression weight (beta = +0.276), followed by adaptive algorithm sophistication (beta = +0.228). These findings suggest that AI-driven rehabilitation systems should prioritise closed-loop difficulty adaptation and clinician-facing dashboards that translate sensor data into treatment-relevant summaries.