Researchers at Cornell University have developed a “smart t-shirt” that logs exercise and posture data with high accuracy without compromising on appearance, comfort, or washability. Unlike many existing smart garments, the shirt—dubbed SeamFit—doesn’t rely on bulky sensors or a tight, restrictive fit. The shirt has potential applications across various fields, including medicine and athletics.
SeamFit uses flexible conductive threads sewn into the neck, arm, and side seams of an otherwise standard cotton shirt. “When the wearer’s body pose changes, the seam electrodes, i.e. the conductive threads, deform and change their self-capacitances, and the coupling between the wearer’s body, a large conductor, and the seam electrodes alters,” says Catherine Yu, a doctoral student in the field of information science at Cornell. “By monitoring the changes in the capacitances, we can infer the body poses/movements that induce these changes.”
A small, detachable circuit board at the neckline records the capacitance data and sends it via Bluetooth to a laptop. There, an AI model analyzes the signal to recognize exercises and count repetitions.
The researchers tested the shirt’s performance in 15 volunteers and described the results in a paper published in March in the Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies.
Engineering High Accuracy Exercise Detection
In the study, each volunteer performed 14 different exercises, including lunges, sit-ups, and biceps curls. Without needing user-specific calibration, SeamFit demonstrated an exercise detection accuracy of 89 percent, exercise classification accuracy of 93.4 percent, and was off by less than one rep on average when counting reps within a set—independent of users, washes, and fits.
SeamFit’s short-sleeve design means some joints, like elbows and wrists, aren’t directly covered by electrodes. Still, their movements can be indirectly tracked, says Yu. And SeamFit can pick up motions from other parts of the body—while isolated leg movements generate weak signals from the shirt, most lower-body exercises involve enough upper-body motion to produce consistent, trackable patterns.
SeamFit still requires some preparation before it goes into the wash. The researchers incorporated a detachable circuit board that must be removed before the shirt is thrown into a washing machine. The conductive threads sewn into the garment, however, retain their conductivity through multiple laundry cycles—although Yu says their performance degrades slightly over time, which introduces variations in signal quality after enough washes. To address this, the team trained their data processing pipeline on datasets collected after multiple washes, enabling the system to adapt to these changes and maintain reliable accuracy.
Researchers used conductive thread sewn into the neck, arms, and side seams of a standard cotton shirt to track a wearer’s movements.Louis DiPietro/Cornell University
“Practical design choices like the removable circuit board and machine washability directly address common user concerns around maintenance and comfort,” says Rong Yin, an assistant professor in the department of textile engineering, chemistry and science at North Carolina State University, who was not involved in the research. “This user-centric approach significantly increases the potential for widespread adoption, not only among fitness enthusiasts but also in clinical settings for physical therapy and rehabilitation.”
From Smart Clothes to Robotics?
Beyond fitness tracking, SeamFit could have applications in robotics. Because it provides an unobtrusive and comfortable way to collect data over extended periods, SeamFit could help train humanoid robots to better mimic human movements, Yu says. By capturing data on human motion, the shirt could also improve the safety of human-robot interactions. Yin adds that SeamFit could also have applications in human-computer interaction “where understanding subtle human movements is crucial.”
To make sense of the signals coming from the smart t-shirt, the researchers developed a custom AI system made up of three parts. A lightweight random forest model detects whether the wearer is exercising or at rest. A deep learning encoder-decoder model figures out what exercise is being performed. And a heuristic peak detector algorithm, which required no training, counts the number of repetitions the wearer performs.
The researchers used a “leave-one-participant-out” approach to test how well their AI models generalize to new users. This involved training the models on data from all but one participant, then testing them on the data from the participant who was left out. By repeating this process for each individual in the study, the team ensured that the system would work even with different people, different shirt fits, and the signal changes from washing.
One day, wearing a smart t-shirt may be as ordinary as wearing cotton is today—and, according to Yu and her colleagues, SeamFit marks a step toward that future.
