AI-powered smart clothing logs posture, exercises
Researchers at Cornell have developed a new type of smart clothing that can track a person’s posture and exercise routine but looks, wears – and washes – just like a regular shirt.
The new technology, called SeamFit, uses flexible conductive threads sewn into the neck, arm and side seams of a standard short-sleeved T-shirt. The user does not need to manually log their workout, because an artificial intelligence pipeline detects movements, identifies the exercise and counts reps. Afterward, the user simply removes a circuit board at the back neckline, and tosses the sweaty shirt into the washing machine.
The team envisions that SeamFit could be useful for athletes, fitness enthusiasts and patients engaged in physical therapy.
Most existing body-tracking clothing is tight and restrictive or embedded with chunky sensors, according to Catherine Yu, a doctoral student in the field of information science and lead researcher on the project.
“We were interested in how we can make clothing smart without making it bulky or unusable,” Yu said, “and to push the practicality, so that people can treat it the way they would usually treat their clothing.”
Alternatively, athletes can choose fitness trackers, like smartwatches or rings, but these are extra devices that people may not want to wear while exercising, and can’t track movement across the entire body.
“Not everyone is willing to try out a new wearable form factor, but people will have clothes on,” said co-author Cheng Zhang, assistant professor of information science in the Cornell Ann S. Bowers College of Computing and Information Science. “We provide a very neat form factor that is always on you.”
Their study,“SeamFit: Towards Practical Smart Clothing for Automatic Exercise Logging,” published in March in the Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies, and will be presented at the UbiComp/ISWC 2025 meeting in October.
Most mass-produced clothing has seams, which Yu realized could be exploited to make a comfortable, affordable piece of smart clothing. She constructed three SeamFit shirts – in small, medium and large – using a home sewing machine to attach conductive threads on top of the seams. The three sizes allowed users to choose a looser or tighter fit, but did complicate the process of interpreting each user’s movements.
To test the shirts’ performance, the team recruited 15 volunteers, who did a series of 14 exercises – including lunges, sit-ups and biceps curls – while wearing SeamFit. Without any calibration or training for each user, SeamFit’s model classified the exercises with 93.4% accuracy and successfully counted reps, with counts that, on average, were off by less than one.
SeamFit works because when people exercise, the threads’ capacitance – their ability to store charge – changes as the threads move, deform and interact with the human body. The circuit board at the back neckline measures the capacitances and transmits them through a Bluetooth connection to a computer. A customized, lightweight signal-processing and machine-learning pipeline then deciphers the movements.
After the workouts, Yu washed and dried the shirts at home.
The project is a new iteration of SeamPose, a previous effort to track body postures using conductive threads in eight seams of a long-sleeve T-shirt.
The team envisions that this type of unobtrusive smart clothing could be especially useful for athletes logging their exercise routines and for physical therapists monitoring the progress of patients at home.
More broadly, this type of technology could assist with human-AI interaction, because by tracking human movements and activities, AI can better understand when to interact and when to wait – such as when someone is eating or sleeping.
Enabling AI to understand human activity is the main focus of Zhang’s Smart Computer Interfaces for Future Interactions (SciFi) Lab, which develops new, AI-powered wearable sensing systems, to enable AI to comprehend human activities and intentions in everyday settings and provide support when needed.
“While this paper demonstrated the approach for a simple garment, we believe it can easily be adapted to a wide range of garments and could take advantage of the complex seam patterns of advanced sportswear,” said co-author François Guimbretière, professor of information science in Cornell Bowers CIS and the multicollege Department of Design Tech.
To create SeamFit, Yu set up a “little sewing factory” in the lab. However, she is currently exploring how the manufacturing process could be affordably scaled up, using industrial serger machines – which sew and make seams using three or four threads simultaneously – and more robust conductive threads.
“By just replacing a single thread in this mass manufacturing process, all of the clothing could easily become smart and be able to have this motion tracking capability,” Yu said. “I’m imagining one day, you open your closet and there’s really no difference between smart and nonsmart clothing.”
Additional authors on the study include Manru Mary Zhang ’25 and Luis Miguel Malenab ’25; Chi-Jung Lee and Ruidong Zhang, both doctoral students in the field of information science; and Jacky Hao Jiang, a visiting undergraduate from Rice University.
Funding for this work came from the National Science Foundation.
artificial intelligence Smart Clothing Cornell University SeamFit posture exercise
By Patricia Waldron, Cornell Ann S. Bowers College of Computing and Information Science
