"AgenticPose: Kinematic Telemetry and Autonomous Visual Intervention Dataset for Physical Rehabilitation"

"Recent advancements in real-time human-pose estimation provide stable telemetry for physical rehabilitation. However, existing systems remain largely passive; they require human clinicians to interpret kinematic data and issue corrective feedback. This paper explores an Agentic AI framework designed to move from passive perception to active, autonomous intervention within interactive visual systems. We detail an end-to-end multi-modal architecture that continuously monitors a patient\u2019s biomechanical movements. By dynamically reasoning over visual inputs and 100Hz supplementary IMU sensor data, the system applies mathematical models to counteract physical occlusion. Rather than merely rendering a static digital twin, an embedded reasoning engine evaluates kinematic deviations and selects an appropriate visual intervention. The system then manipulates the interactive graphics environment by deploying AR overlays, color-coded joint corrections, and directional vectors to guide the user in real-time. This framework suggests a viable model for human-AI collaboration in digital healthcare. We evaluate the system's efficacy in reducing biomechanical error, its rendering latency and its impact on user cognitive load, illustrating how Agentic AI might serve as an active participant in physical therapy."