AI Robot Transforms Stroke Therapy Rehabilitation

Fifteen million people worldwide have a stroke every year, and five million of them end up with a permanent disability, according to figures cited by MIT researchers behind a new robotic therapy project. Getting consistent rehab care to that many patients is hard enough on its own. It gets harder when you factor in a shortage of physical therapists that's been building for years, leaving many stroke survivors with less hands-on time than they need to recover motor function.

A team at MIT's Department of Mechanical Engineering thinks part of the fix lies in AI, not to replace therapists, but to stretch their expertise further. Johannes Lachner, who did the work as an MIT-Novo Nordisk Artificial Intelligence Postdoctoral Fellow before moving to Purdue University as an assistant professor, teamed up with Noah Geiger, a former MIT visiting student now at Robert Bosch GmbH, to build a dual-arm robot that learns directly from physical therapists. The system pairs transformer-based diffusion models, the same underlying tech behind AI image generators, with real-time force feedback, letting the robot adjust how much physical support it gives based on what it observes a therapist doing during training sessions.