Stroke rehab robot learns from therapists in early tests

Lifting an arm, reaching for a cup, buttoning a shirt: after a stroke, these gestures can become the hardest work of the day. Each year, strokes leave 5 million people with long-term impairments, and the shortage of physical therapists who could help them keeps growing. Johannes Lachner, now an assistant professor at Purdue University after a postdoc at MIT, and Noah Geiger, a former visiting student at MIT, are trying a different kind of help.

Their robotic system combines generative AI with real-time force feedback. It learns from real therapists, and it measures in real time how hard the patient pushes or pulls. From that, it adjusts how much assistance to give, so that the patient gets just the right amount of help during exercises like reaching, lifting the arms and performing tasks of daily living. The robot does not work alone: what it does is built on what therapists already do with their hands.

Lachner's stated goal is "not to replace therapists, but to extend their reach." He reports that initial robotic experiments have demonstrated the feasibility of the approach. Those experiments are the first step. The team has so far described feasibility only, and the next phase is still ahead.

That next phase aims to build therapist-specific models and test them in a long-term clinical study with the same patients who previously received manual therapy. That comparison, with patients who already know hands-on rehabilitation, is where the approach will meet its real test.

For patients, the promise is access: more guided exercise for each therapist, if the clinical study confirms the approach. The team gives no timeline for that study. Lachner also sees possible uses in contact-rich industrial settings or collaborative workspaces, a field the team does not detail.