We present a physics-informed diffusion framework for therapist-like robot assistance in upper-limb physical therapy. Unlike purely data-driven methods, our approach integrates generative modeling with physically consistent impedance control, combining principles of human motor control with energy-based models to generate stable, therapist-like interactions. The diffusion model produces simulated trajectories in real time, while the physics-informed layer ensures safe compliance modulation during contact. Validated with healthy participants, the framework achieves therapist-like adaptability and is now progressing toward clinical studies with patients affected by neurological and orthopedic impairments.
Teleoperation is used to record patient movements with varying levels of support, enabling the model to learn the dynamic relationship between force and motion.
The robot autonomously assists with upper-limb rehabilitation by adjusting support in real time based on patient engagement.
Code will be released soon upon publication.
