EchoMotion uses smartwatch ultrasound to estimate full-body movement

A stock smartwatch can estimate a wearer's head-to-ankle movements without recording video. Cornell's EchoMotion research uses the watch's existing audio hardware to infer the positions of 12 body joints. The work appeared in a September 2026 journal issue, ahead of the university's November 11 account. It is a research system, not a feature already delivered to every watch owner.

The useful comparison is not with a step counter alone. Cornell's BodyTrak demonstrated wrist-based full-body tracking in 2022 by adding a small camera. A learning model interpreted the body silhouettes in its images. EchoMotion takes a different route: it substitutes reflected sound for that visual input, with the prospect of using hardware already inside a commercial watch.

Camera-free pose estimation also has a history. The 2023 IMUPoser study used inertial sensors in phones, watches and earbuds. Those components measure acceleration and rotation; the system could make a best-guess body pose from whatever subset of devices was available, potentially just one. EchoMotion's distinction is therefore active acoustic observation, not the first attempt to infer a whole person from sparse wearable data.

Think of clapping in a dark room and listening to the returning sound. Here, the watch emits ultrasound rather than a hand clap, and its microphone receives reflections from the body. A learned model translates changing echoes into estimated joint locations. It does not produce a photograph or directly see through the person.

Cornell reported tests involving 15 people on an unmodified Galaxy Watch 6. For new users, average joint-position error was below roughly ten centimetres. Twelve minutes of individual calibration brought it to about six centimetres. Those are centimetre-scale estimates, not precision measurements of an elbow or ankle.

Reflections do not come exclusively from the wearer. Corridor results deteriorated slightly because of echoes from other people. Crowd conditions remain relevant to whether the approach will work reliably outside a demonstration.

For everyday exercise, the appeal is obtaining more detailed movement information without adding a camera setup. The practical next step is dependable interpretation across people and surroundings. Making an on-screen figure move is only the starting point; useful guidance must also tell movements apart consistently.

Original paper issue: September 2026. Cornell University report: November 11, 2026.