Multi-Agent AI Framework for Personalized Physiotherapy
A research article available on arXiv (2604.21154) introduces a Multi-Agent System (MAS) framework that leverages Generative AI and computer vision to enhance adherence to physiotherapy at home. This system features four micro-agents: the Clinical Extraction Agent, which transforms unstructured medical notes into kinematic constraints; the Video Synthesis Agent, responsible for creating tailored exercise videos through foundational video generation models; the Vision Processing Agent, which performs real-time pose estimation; and the Diagnostic Feedback Agent, offering corrective guidance. This innovative approach seeks to complete the tele-rehabilitation process by delivering personalized feedback, overcoming the drawbacks of static pre-recorded videos and generic 3D avatars. The paper elaborates on the architecture and prototype pipeline of the system.
Key facts
- arXiv paper 2604.21154 proposes a Multi-Agent System for physiotherapy
- Four specialized micro-agents: Clinical Extraction, Video Synthesis, Vision Processing, Diagnostic Feedback
- Clinical Extraction Agent parses medical notes into kinematic constraints
- Video Synthesis Agent uses generative AI to create personalized exercise videos
- Vision Processing Agent performs real-time pose estimation
- Diagnostic Feedback Agent issues corrective instructions
- Addresses low compliance due to lack of personalized supervision
- Prototype pipeline using La (incomplete in source)
Entities
Institutions
- arXiv