A smartwatch can count your steps, but researchers at Cornell University have shown that it may also be able to estimate how your entire body is moving. Their EchoMotion project uses sound from an ordinary watch rather than cameras placed around a room. It is a laboratory research result, not a feature that has suddenly appeared on everyone’s wrist, yet it points to an intriguing future for smartwatch body tracking, exercise feedback and accessible technology.
The Cornell team’s research announcement, published on 11 November 2026, describes a system that sends out sound waves and listens to their echoes. AI software interprets the returning signals to estimate the positions of body joints. In tests using a standard Samsung Galaxy Watch 6, new participants’ joint-position estimates were on average within four inches of the camera-based reference measurements. A short personal calibration improved that result further.
How smartwatch body tracking works with sound
Anyone who has watched a bat negotiate a dark space already understands the rough idea. Sound travels out, strikes objects and returns with information about the surroundings. EchoMotion brings a related principle to a watch’s speaker and microphone, using ultrasonic signals that people cannot normally hear. The pattern of echoes changes as a person’s arms, legs and torso move. A trained model learns to associate those patterns with likely body positions.
There is an important difference between tracking movement and taking a photograph. A camera records an image that can reveal the room, clothing and other people. Acoustic measurements capture a much less direct signal. That makes the approach interesting for activities where a person may not want a camera watching them. It does not mean the technology is automatically free from privacy concerns, because movement patterns can still reveal habits and behaviour when recorded or analysed over time.
The researchers initially worked with a custom watch-like device containing additional microphones and speakers. They collected echoes from ten volunteers performing eighteen exercises, then built a model to connect acoustic patterns with positions measured by a reference camera system. Moving to a conventional smartwatch was a separate test. That progression matters: a clever sensor arrangement is one thing, but making it work with the limited hardware already on consumers’ wrists is a more demanding test.
On an unmodified Galaxy Watch 6, fifteen new volunteers tried the system. Cornell reports that average joint-position error was below four inches. After a one-time twelve-minute calibration, average error fell to about 2.4 inches. This describes the distance between predicted and reference positions of measured joints, not a blanket claim that every gesture was recognised correctly or that the watch could replace a medical movement assessment.
Why tracking more than steps matters
Step counts are helpful for showing broad activity, but they tell us very little about the quality of a movement. Two people can record the same number of steps while walking with different gaits or recovering from different injuries. Similarly, a fitness watch might know someone is exercising without recognising whether their arm or knee is in the intended position. Estimating more of the body’s posture could add a different kind of feedback.
One potential application is exercise technique. A watch that can distinguish patterns of movement might eventually count repetitions more intelligently or draw attention to an unusual motion. The Cornell researchers discuss fitness tracking as an area they are exploring. That is not the same as demonstrating a reliable personal coach for every exercise. Body shape, watch placement, loose clothing and the environment all complicate the problem, and the consequences of a mistaken correction would vary by activity.
Another possibility is studying movement outside a specialised laboratory. Traditional body tracking can involve multiple cameras, markers or dedicated equipment. Those tools may be extremely accurate, but they change the conditions in which people are observed. A watch-based method could make it easier to collect movement information during ordinary activities, provided people understand what is being measured and agree to it. Researchers would still need to establish accuracy for the question being asked.
The idea connects with existing attempts to make wearables more informative without making them cumbersome. LiveAIWire has looked at battery-free devices that analyse walking patterns, an approach focused on a different sensing method. Both reflect the desire to understand movement with smaller devices. Neither means a consumer should treat a prototype’s estimates as a clinical diagnosis or a substitute for a trained professional.
What happened when people left the laboratory
Cornell did not confine the evaluation to a silent, empty room. The team also asked ten new participants to perform exercises in a small room, a narrow hallway and beside a noisy road. Its account says ordinary furniture and outdoor noise did not defeat the approach. Performance suffered somewhat in the hallway, where echoes from other people complicated the signal. That is a useful result because acoustic sensors interact with their environment rather than observing the wearer in isolation.
A crowded train, gym or office would pose further challenges. Several people may be moving nearby, surfaces may reflect signals in unexpected ways and the wearer’s arm position may change the signal path. The research does not establish that the system will perform just as well in all those settings. Nor does it show that any current smartwatch, regardless of design, can run the full experience without modification or a suitable software model.
The team reports that the same trained model also worked on a Galaxy Watch 7 and a Pixel Watch 3 without additional model training. That is encouraging for compatibility, but a research demonstration on selected devices is different from an official consumer rollout. Battery use, long-term stability, processing requirements and integration with operating systems would need consideration before a manufacturer could promise everyday use. Users should not go looking for an EchoMotion button in their existing fitness app on the strength of this research alone.
The privacy advantage needs a closer look
The absence of a camera may make movement tracking feel less intrusive, particularly in the home. It avoids creating ordinary video footage of a person doing exercises or moving through a room. However, privacy depends on how the measurements are processed, stored, shared and explained. A history of movement could still reveal when somebody is at home, how active they are and which activities they repeat. The appropriate protection would depend on the final product and the sensitivity of its use.
It also matters who controls the analysis. A watch giving immediate movement feedback locally is a different proposition from an employer or insurer receiving regular reports about an individual’s behaviour. The technology does not imply either business model. Yet the potential to infer more detail from familiar sensors makes it important to build consent and limits into design from the beginning, rather than assume there is no privacy issue because no photographs are taken.
For broader context, LiveAIWire has explored interfaces that adapt to people’s changing needs. Sensing and responding to movement could, in principle, support more accessible experiences, although the Cornell study does not itself demonstrate such a deployed application. It is equally important that a system’s inferences remain understandable: someone should know whether a warning reflects a confident measurement or an uncertain guess.
A different way of listening to movement
EchoMotion has a predecessor in the same research area. Cornell reported earlier in 2026 on using smartwatch sonar to follow hand movements. That earlier project addressed finer hand gestures, while EchoMotion attempts to estimate posture across the body. The common thread is the use of everyday audio hardware in ways its original designers may not have anticipated.
That is part of what makes the research appealing. It does not require attaching a camera to every wall or wearing a complicated collection of sensors to produce a useful signal. But the results should be judged by what was actually tested: selected exercises, a relatively small number of people and measured positioning errors. Researchers will need larger and more varied assessments before confidence can extend to different ages, mobility needs and environments.
An ordinary smartwatch that understands more than a step count might eventually offer richer exercise feedback and new accessibility tools. At present, EchoMotion offers a clear proof of concept with meaningful limitations. The sensible expectation is not that a watch can see through clothes or know everything the wearer is doing. It is that sound from a familiar device can reveal more about movement than its owner might ever have imagined.
LiveAIWire has also reported on phone captions that add information about the direction of a speaker, a separate example of digital devices turning sensory signals into practical context. Whether through sound, motion or speech, the best outcome is technology that helps the person using it without demanding excessive attention or quietly collecting more information than they intended to share.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, cybersecurity and the social impact of emerging technology. LiveAIWire is an independent, human-led technology publication using AI-assisted research, editorial production and original AI-assisted editorial illustrations under his direction.