Rehabilitation Robot Therapy and ABB Cobots

A rehabilitation robot does part of a physiotherapist's work: it guides a weak arm or leg through the same movement hundreds of times, pushes back when the patient can do more, and records every repetition. After a stroke or a spinal injury, that volume of practice is what most patients cannot get from human therapists alone. This post traces the job from hands-on therapy and Victorian exercise machines to MIT-MANUS and the Lokomat, looks at this year's research on robots that learn from therapists, and asks where force-controlled cobots such as ABB's GoFa fit, in clinics now and perhaps at home later.

A patient walks in an EksoNR robotic exoskeleton while a therapist supports him from behind. Photo: AriceEB / Wikimedia Commons (CC BY-SA 4.0)

From therapists' hands to the first rehab robots

For most of the twentieth century rehabilitation meant a therapist's hands. Physiotherapy grew out of the medical gymnastics taught in Stockholm in the early 1800s, and it expanded fast after the First World War, when armies trained staff to help wounded soldiers walk and use their arms again. The polio epidemics of the 1940s and 1950s added thousands of children who needed their limbs moved and stretched every day. The method stayed the same throughout: a trained person holds the limb, feels the resistance and adjusts. It works well, but each therapist can treat only one patient at a time.

A physical therapist helps two children recovering from polio exercise at a rail, 1963. Photo: CDC / Charles Farmer via Wikimedia Commons (public domain)

Exercise machines came surprisingly early. The Swedish physician Gustav Zander opened his first institute in Stockholm in 1865 with 27 exercise machines of his own design. Some let patients work against adjustable levers and weights, while others were driven by an engine and moved the patient's body for them. Zander's devices won a gold medal at the 1876 Centennial Exhibition in Philadelphia, and by 1911 there were about two hundred Zander institutes around the world. They were the first attempt to make repetitive therapy mechanical, though the machines could not sense how the patient was doing.

A man demonstrates one of Gustav Zander's knee bending machines, built in Sweden around 1892. Photo: Tekniska museet / Wikimedia Commons (CC BY 2.0)

Real rehabilitation robots arrived in the 1990s. At MIT, Neville Hogan and Hermano Igo Krebs started the MIT-MANUS project in 1989 and introduced the robot in 1991. It guided a stroke patient's hand across a table while measuring force and position. In 1994 it went into a trial at Burke Rehabilitation Hospital in White Plains, New York, where one group of patients received an extra hour of robot therapy each day. For walking, the Swiss company Hocoma, a 2000 spin-off of Balgrist University Hospital in Zurich, launched the Lokomat in 2001: a treadmill, a body weight harness and powered leg braces.

An early prototype of the HAL powered leg suit from Cyberdyne and the University of Tsukuba on display in 2010. Photo: Yuichiro C. Katsumoto / Wikimedia Commons (CC BY 2.0)

Robots that learn from physiotherapists

The newest work goes back to the same MIT lab. In August 2026 MIT reported that Johannes Lachner and Noah Geiger, in Neville Hogan's Newman Laboratory, are training a dual arm KUKA robot with real-time force feedback to help with exercises such as arm lifting and reaching. The robot learns from therapists in two ways: through teleoperated sessions, and through force-sensing gloves worn while therapists treat patients. A clinical study is planned with the Technical University of Munich and the Pfennigparade rehabilitation centre. The team's stated goal is "not to replace therapists, but to extend their reach," as Newsweek also reported in September.

US Defense Secretary Ash Carter tries a device in a rehabilitation and neuromuscular robotics lab, March 2016. Photo: Clydell Kinchen / U.S. Department of Defense via Wikimedia Commons (public domain)

The other trend is measurement outside the clinic. In September 2026 a team led by the University of Massachusetts Amherst, with Shirley Ryan AbilityLab and Washington University in St. Louis, described a wrist-worn device whose machine learning model tracks how arm impairment changes after a stroke, so therapists could adjust treatment between visits. Clinics meanwhile keep adding robotic gait trainers and exoskeletons, such as Ekso Bionics' EksoNR, which a patient walks in while a therapist stands behind. Home devices mostly handle single joints, like connected knee units that therapists can set up and monitor remotely.

An EXO-Robot gait trainer at the Palon ReWalk neuro-rehabilitation centre in Guwahati, India, 2026. Photo: Shefali.assam / Wikimedia Commons (CC BY-SA 4.0)

ABB does not make a rehabilitation robot and has not announced one, but its GoFa CRB 15000 has the features rehab research depends on. It has torque sensors in all six joints, limits its power and force so it can work next to people, and can be moved by hand to teach it a path, the same lead-through idea therapists use when they demonstrate an exercise. ABB also runs a healthcare research hub in Houston, where GoFa and YuMi robots work on hospital lab automation. A force-controlled cobot like GoFa could in principle hold and guide a limb, but that is a possibility rather than a product, and it would need medical certification.

Lead-through teaching: moving an ABB GoFa arm by hand. Photo: ABB

Rehabilitation robots in clinics and homes after 2026

In clinics the near future is one therapist running several robot stations, with the robots doing the repetitions and the therapist deciding the plan. The evidence for that model is mixed. A 2010 US Veterans Affairs trial of MIT-MANUS-based robots, published in the New England Journal of Medicine, found results similar to an equally intensive programme run by therapists, and the large UK RATULS trial in 2019 found no clear gain over usual care for arm function. Robots make high-dose therapy cheaper to deliver, but nobody has shown that they beat the same dose given by people.

Trying on an Ekso Bionics upper-body exoskeleton in 2020. Photo: Office of Rep. Mark DeSaulnier / Wikimedia Commons (public domain)

Homes are harder. A robot that pushes on a patient's arm must never push too hard, and a home has no therapist standing by to stop it. Industrial cobots have dealt with the same problem for years. Power and force limiting, speed monitoring and safe zones are already standard on industrial cobots, and similar rules would have to be built into any home rehab arm and then certified as a medical device. Cost is the other barrier, because insurers pay for clinic sessions more readily than for equipment at home. We looked at wearable support in our post on exoskeletons and heavy lifting.

ABB's GoFa CRB 15000 limits its power and force so a person can guide it and work beside it. Photo: ABB

What will probably reach homes first is a mix: wearables that measure progress, simple powered devices for one joint, and remote sessions where a therapist watches and adjusts the programme. Full robot arms that guide a whole limb are likely to stay in clinics for several more years, with learning systems like MIT's trained on real therapists' movements. For engineers, the skills overlap with factory cobots: force control, safe motion and programming by demonstration, the same topics covered in our guide to programming robots without code.

A GoFa cobot at ABB's healthcare research hub. Photo: ABB

Innovation

AI solutions for effortless ABB robot control.

Automation

Robotics

ceojohntran@chatwithrobot.net

+84905311611

© 2025. All rights reserved.

qtran1215@gmail.com