One Exoskeleton Could Replace Three in Rehab Clinics

One Exoskeleton Could Replace Three in Rehab Clinics

Nature Machine Intelligence publishes a roadmap for an AI controller that lets a single lower-limb exoskeleton handle walking, lifting, and stairs, no mode switching required!

LS
Linsey Smith
Aug 27, 2026
2 min read

Researchers from Northeastern University, Georgia Tech, and Alphabet-backed Skip Innovations published a road map in Nature showing how a single AI controller could drive a lower limb exoskeleton across any human activity without reprogramming. The experiment is a shift that would let a stroke patient walk, lift, and turn on the same machine instead of being moved between three.

As of today, companies like ReWalk Robotics and Ekso Bionics sell exoskeletons locked to fixed gait profiles. This means, a clinic that wants to offer walking, lifting, and stair training today buys three separate units. Each one costs hundreds of thousands of dollars. A task agnostic controller collapses that into one device. That is a direct hit to the equipment budgets of every rehab center in the country.

The discovery is not a small operational shift. Under current technology, a stroke patient in a clinic gets moved between machines one by one. One for gait. One for lifting. One for balance. The therapist resets the mode each time. The patient's session is broken into fragments. However, if the AI controller is integrated into clinics as a task-agnostic system, the patient puts on one suit and stays in it. The system reads joint forces and shifts behavior on its own. The therapist's job changes from managing hardware to actually working with the patient.

The team already proved the controller works. In 2024, the same system ran across 28 tasks and cut users' energy cost by up to 19.7% compared to walking alone. That was the lab. This roadmap is the plan for getting it into a hospital corridor.

Three things still block real deployment. Training data is hard to collect without putting repetitive load on patients. Safety checks need to be built in from day one, not added later. And a person has to trust a machine that is physically pulling on their legs. No benchmark score fixes that.

Skip Innovations, a spinout from Google X, gives the roadmap a commercial path that pure academia would not have. The authors are upfront that the hardest problem is not technical. It is making a scared person feel safe.

The destination is simple. Exoskeletons that figure out what you need instead of waiting for someone to set a mode.

About the Writer

More from Mindplex

Keep reading

Three more ideas worth your time.

Browse MindBytes

Discussion

Join the discussion

Sign in to share a response with the community.

Type @ to mention someone Type / or use + to add a block Highlight text, then choose Link
Loading editor

Comments cannot be edited after posting because they become part of the reputation record. Give yours a quick review first.