Rutgers engineers have developed a smart shoe that automatically analyzes how a person walks, a technology that someday could help monitor people with Parkinson’s disease, spinal cord injuries, traumatic brain injuries, and other movement disorders.
The prototype analyzes movement with 95.4% accuracy while counting steps and estimating calories burned.
The research, led by Simiao Niu , a biomedical engineer at Rutgers University-New Brunswick, was published in Science Advances .
What makes the shoe unusual is that it powers the analysis with energy generated from the wearer’s footsteps. It has no battery to recharge.
“When you are walking or running, you automatically have biomechanical energy available, so you can harvest this energy,” said Niu, an assistant professor in the Department of Biomedical Engineering in the Rutgers School of Engineering.
A shoe was a natural target, he said, because walking produces the information the researchers want to analyze and the energy needed to study it.
“When you sit down, there is no energy available, but you don’t need gait monitoring,” Niu said.
A device embedded in the sole produces electricity from the pressure and friction of each step. The process, known as the triboelectric effect, is related to the static electricity created when different materials rub together.
The electricity arrives in irregular bursts that the electronics cannot use directly. The researchers designed a power-management circuit that converts it into a useful form, increasing the usable energy by as much as 120 times compared with a conventional method.
When studying walking information, scientists use “gait” to describe a person’s pattern of walking, including balance, speed, stride, and rhythm. Changes can offer clues about disease progression, fall risk or rehabilitation.
“Gait is one of the most significant biomarkers for a lot of diseases,” said Niu.
Doctors often evaluate gait by watching a patient walk briefly in a clinic or laboratory. A wearable device could eventually measure movement over longer periods as people go about their daily lives.
“If you are able to use what I call the ‘worry-free shoes’ we’ve developed, patients can just wear them, and the shoes can automatically collect their gait pattern,” Niu said.
With further development and clinical testing, similar shoes might one day assess fall risk, detect unusual walking patterns or follow recovery after a brain or spinal cord injury. The design also might be adapted to monitor heart activity, biochemical signals or other aspects of health.
The white athletic shoes conceal electronics in the heels. The current version is an early prototype, not a medical device, Niu said. It cannot yet diagnose disease, predict a fall, or determine whether a treatment is working, but it shows basic walking patterns can be analyzed without a rechargeable battery.
That could address a weakness in existing health wearables, he said. Smartwatches and other devices can gather large amounts of information, and artificial intelligence (AI) can turn those measurements into useful findings. But AI requires energy. As wearables become more intelligent, they may drain their batteries faster.
Niu calls this problem the “energy-intelligence bottleneck.”
“We want to solve the fundamental bottleneck in current wearable devices,” Niu said. “We are developing a smart wearable with integrated AI functionality that can harvest energy on its own, so you don’t need to worry about charging.”
Niu encountered the problem while working at Apple, where he helped develop an electrocardiogram sensor for the Apple Watch. Monitoring stops when a health device is removed for charging, and users may forget to put it back on.
“Once you put it onto the charger, you typically forget about it, and then you don’t wear it,” Niu said. “Those wearables cannot monitor your health if you just leave them in your drawer.”
An accelerometer measures the foot’s movement along three axes, labeled x, y and z. A tiny processor uses AI to analyze patterns in those measurements and classifies each 15-second segment as one of four activities: slow walking, fast walking, running, or climbing stairs. It displays the results on a screen attached to the shoe.
The analysis occurs inside the wearable, an approach known as “edge AI.” Because the shoe doesn’t continuously send raw information to a phone, computer, or cloud server, it requires significantly less energy.
Making the AI algorithm small enough to fit the tiny processor’s limited memory was another challenge. The original model examined 21 characteristics of movement and achieved 98.1% accuracy, but required more memory than the shoe’s processor could hold.
The team found that the variation in movement along the three axes provided most of the information the AI algorithm needed. The smaller model achieved 95.4% accuracy while running about 15 times faster and using about one-sixth as much current.
The sensor and AI algorithm together consume 86 microwatts, a fraction of the power used by many wearable AI systems. In laboratory tests, even slow walking generated enough electricity to keep the complete system operating.
Fuying Dong, a Rutgers biomedical engineering doctoral student and the first author of the study, said the project required the team to treat the shoe as one connected system.
“The idea of how to co-design the whole system is the best thing I learned from this project,” Dong said. “You break a huge project into smaller pieces, finish them one by one, and try to figure out what’s the biggest story behind it.”
The prototype was developed using data from four healthy volunteers ages 23 to 26. So far, the AI algorithm has been trained and tested only on the activities included in the study. It hasn’t been tested in older adults, people with movement disorders, or patients undergoing rehabilitation.
Niu, Dong, and Chi Han are inventors on a Rutgers provisional patent application related to the technology.
Explore more of the ways Rutgers research is shaping the future .
Science Advances
Data/statistical analysis
People
A biomimetic, ultralow-power edge-AI-empowered and self-sustaining gait analysis system
19-Aug-2026
S.N., F.D., and C.H. are inventors on a U.S. provisional patent application related to this work, titled “Wearable Gait Analysis System and Method” (U.S. Provisional Application No. 63/897,241; filed 10 October, 2025; Rutgers Reference No. P2025–279-01). The current assignee is Rutgers, The State University of New Jersey. All other authors declare that they have no competing interests.