Rehabilitation robot learns to feel stroke patients' efforts

Researchers Johannes Lachner (Purdue University, formerly a postdoc at MIT) and Noah Geiger (Robert Bosch GmbH) published a bimanual robotic stroke rehabilitation system in IEEE Transactions on Robotics, where a diffusion model adjusts force and resistance in real time to tailor assistance to the patient's efforts.

Compiled by Sergey KostenchukPublished 2026-10-20Updated 2026-10-21

2026-10-20 Coding OpenAI

🦾 The rehab robot feels force

Former MIT postdoc Johannes Lachner and Noah Geiger (Robert Bosch GmbH) created a bimanual robot for stroke rehabilitation. Inside is a diffusion model that senses force and resistance and adjusts assistance in real time to fit the patient. They trained it via telemanipulation using Apple Vision Pro with gloves.

🌍 Diffusion models have moved from images to force interaction: impedance learning, which, according to the authors, is transferable to industrial tasks. Stroke leaves lasting impairments in approximately 5 million people annually.

👤 Patients with hand impairments receive more frequent sessions with support tailored to their capabilities. In Munich, physiotherapists are recorded in force gloves, followed by a clinical trial.

Source 1: https://www.technologyreview.com/2026/10/20/1145282/robotic-therapy-for-stroke-patients/ Source 2: https://news.mit.edu/2026/personalized-physical-therapy-stroke-rehabilitation-powered-by-ai-0805

Sources