Motivation
Physical and occupational therapists provide personalized assistance, adapted to each patient. The therapist guides the movement with one hand while supporting at particular points with the other. They feel when more help is needed, and — just as importantly — when to withdraw it so the patient does as much of the work as possible.
Vision and Impact
We build robotic systems that can emulate this behavior. Our algorithms emphasize physical interaction rather than pure trajectory generation. When the robot senses that a movement cannot be completed independently, it supports more; when less help is needed, it backs off.
Our goal is that more patients can be treated, and underserved communities can benefit.
Clinical pilot tests
The work has moved out of the laboratory into an ongoing study with Pfennigparade, an outpatient rehabilitation center in Munich, Germany.
During physical and occupational therapy sessions, therapists wear force-sensing gloves and are recorded on camera while treating patients. Sensors record both movement and force, and that data trains models that capture the physical interaction style of an individual therapist. In the next stage, a model is trained for each therapist, and the robot provides assistance as that therapist would.
References
- N. Geiger, T. Asfour, N. Hogan and J. Lachner, “Diffusion-Based Impedance Learning for Contact-Rich Manipulation Tasks,” IEEE Transactions on Robotics, vol. 42, pp. 3160–3177, 2026.
- Personalized physical therapy: Stroke rehabilitation powered by AI — MIT News, 5 August 2026.
- Forschungsteam entwickelt Therapie-Assistenz-Roboter — Stiftung Pfennigparade, 30 July 2026.
All research