For many stroke survivors, walking even a short distance can feel like running a marathon. Simple movements often bring fatigue, instability, and the constant reminder of lost coordination. To address this challenge, researchers at Georgia Tech have designed a robotic exoskeleton that does more than provide mechanical support. It learns from the user.
By integrating artificial intelligence (AI), the system adapts in real time to each person’s unique gait, reducing strain and restoring confidence. The Georgia Tech innovation, recently described in IEEE Transactions on Robotics, highlights a shift in thinking: instead of requiring people to adapt to rigid machines, the exoskeleton adapts to them.
From rigid to responsive support
Traditional exoskeletons, while promising, often demand tedious manual adjustments. Engineers must calibrate sensors, fine-tune settings, and repeatedly tweak the system for each user. Even then, the results may not fit the unpredictable gait of someone recovering from a stroke.
“Getting it right can be frustrating, sometimes nearly impossible,” explains Aaron Young, associate professor at the George W. Woodruff School of Mechanical Engineering. “With AI, the exoskeleton does the mapping itself. It learns timing and rhythm through a neural network, without the need for manual fine-tuning.”
The study focused on a hip exoskeleton, designed to provide torque at the hip joint. Sensors track every step, feeding data to the AI, which immediately adapts the level of support. Within minutes, the device aligns with the user’s walking style, aiding that feels natural rather than forced.
Fast, accurate adaptation
Walking after a stroke is rarely predictable. Strides may vary from day to day, or even from one step to the next. Conventional exoskeletons, modeled on the steady gait of healthy adults, often fail in this context. The Georgia Tech solution uses a neural network that quickly recognizes a patient’s movement patterns. “The speed really surprised us,” says Young. “In just one to two minutes of walking, the system had already learned a person’s gait with high accuracy.”
The results are striking: the AI-powered exoskeleton reduced errors in gait tracking by 70% compared to standard systems. According to lead clinician Kinsey Herrin, the AI doesn’t just learn once, it keeps learning, adapting stride by stride to ensure seamless support.
This adaptability has real-life impact. Patients walk farther with less effort, regaining not only mobility but also independence. “When you see someone able to walk without becoming exhausted,” Young adds, “you realize this isn’t just about robotics. It’s about giving people part of their life back.”
A universal solution
One challenge in exoskeleton research is that each device uses its own set of sensors, producing data in different formats. Typically, an AI system trained on one machine fails when transferred to another.
Young’s team addressed this by designing software that acts like a universal adapter plug. After just ten strides of calibration, the system reduced error rates by over 75%, no matter which exoskeleton it was connected to. “The goal is simplicity,” Young explains. “Someone should be able to strap on a device and, within a minute, feel like it was built specifically for them.”
Beyond stroke recovery
While the current research centres on stroke rehabilitation, the applications are much broader. Adaptive exoskeletons could support:
- Older adults coping with age-related muscle weakness.
- Patients with neurological conditions such as Parkinson’s disease.
- Individuals with orthopaedic challenges, including osteoarthritis.
- Children with mobility disorders, offering tailored support during growth.
Clinical trials are now underway to evaluate how well the system supports diverse everyday activities. “There is no such thing as an ‘average’ user,” says Young. “The real challenge is designing technology flexible enough to meet the full spectrum of human mobility.”
Toward lifelong adaptation
The long-term vision is clear: an exoskeleton that continues learning over time, adjusting as a user’s body and mobility change. Inseung Kang, lead author of the study and now assistant professor at Carnegie Mellon University, sees the potential as transformative.
“We’ve built a system that adapts to a person’s walking style in just minutes,” Kang explains. “But imagine a companion that keeps learning with you throughout your life, evolving as your needs change. A robot that understands your gait and provides the right assistance every step of the way.” With AI at its core, Georgia Tech’s exoskeleton points to a future where technology doesn’t simply help people walk, it walks with them.
Project MARCH
In the Netherlands, the TU Delft student team last year unveiled Project MARCH IX, the ninth version of their robotic exoskeleton, during a live demonstration at the Louwman Museum in The Hague. Designed to help people with spinal cord injuries stand and walk again, the new model emphasizes greater maneuverability, lighter weight, and improved control.
Key innovations included a passive tilting ankle joint that enables sideways walking, integrated cameras that scan the environment for obstacles, and a new motor controller that boosts responsiveness. The exoskeleton is also 25% lighter than its predecessor, enhancing usability. During the demonstration, pilot Daan van der Heyden successfully completed an obstacle course, showcasing its real-world potential.
