Google Files Patent for a Smartwatch That Measures How Each Foot Walks
Your watch already counts steps. This Google patent application would have it estimate how long each of your feet touches the ground, using nothing but your swinging wrist and your height.

- Publication number
- US 2026/0363578 A1(US20260363578A1)
- Applicant
- Google LLC
- Filed · Published
- May 13, 2026 · Nov 19, 2026
- Inventors
- Amir Bahador Farjadian Bejestan, Alicia Yolanda Kokoszka, Jonathan Hsu, Lawrence Zhengyu Cai, Anupam Pathak, Kyle Grant DeHolton, Jonathan Wesor Wang, Ming-Zher Poh, Shun Liao
- Status when we published
- Waiting for an examiner
- Document
- 20 claims
- US classification
- 702/141
- Parent application
- Claims priority from a provisional application 63806529 (filed May 15, 2025)
What Google's wrist-based walking tracker does
Every time you walk with a watch on, your wrist swings back and forth, and that swing carries a surprising amount of information about how your legs are moving. Google's patent describes a way to read it. A smartwatch's motion sensors feed a trained AI model, along with your height, and the model estimates how fast you walk and how long each foot spends on the ground.
That level of detail has traditionally meant a lab with special equipment. Phones can do some of it, but they work best when held in one consistent spot, and a phone riding in a bag or a loose hand gives noisy readings. A watch stays put on your wrist.
The filing says the estimates could be compared over time to spot meaningful changes, which could flag things like mobility problems, fall risk, or cognitive decline. It also says the device would ask for your permission before collecting any of this.
… estimating, using the deep learning model, a set of walking metrics based on the motion data and the height parameter, the set of walking metrics including at least one of gait speed, double support time, unilateral step length, swing time, and stance time.
Translation: An artificial intelligence model calculates specific walking measurements using motion data and user height.
How the model turns arm swing into step length
The claims cover a method where a device collects motion data from a sensor, gets the user's height, feeds both into a deep learning model (a kind of AI trained on many examples), and gets back a set of walking metrics. Those metrics include at least one of gait speed, double support time (the share of a stride when both feet are on the ground), unilateral step length (the distance between one foot's landing and the other's), swing time (how long a foot is in the air), and stance time (how long a foot is planted).
In the version the filing describes, the watch's accelerometer and gyroscope (sensors that measure movement and rotation) produce readings that are chopped into five-second segments, each overlapping the last by about one second. The model is a temporal convolutional network (a type of neural network that is good at spotting patterns across time) with several output "heads," one for each metric, including separate left and right foot values.
Training data came from people walking a 100-foot track with pressure-sensitive mats that recorded the true numbers. The set included six-minute walk tests, fast walks, and walks while wearing a locked knee brace to fake a limp. Random rotations, flips, and added noise were applied to the training data to mimic different ways of wearing a watch.
- Height can be typed in by the user or pulled from stored data.
- Age, sex, or weight can be added as extra inputs.
- The model can run on the watch, a paired phone, or in the cloud.
This technology enables the estimation of walking metrics using data from a wearable device and a user's height.
Translation: The system figures out how you walk using information from a wearable gadget and your height.
Why walking speed on your wrist could matter
If this ended up in a health app, you could get a running picture of how your walking changes over weeks and months, without a clinic visit or sensors strapped to your legs. The filing describes comparing your numbers over time against thresholds, such as a change in walking speed of 10 to 20 centimeters per second or a step length shift of 5 centimeters or more, to flag meaningful changes.
It also works without your phone, which matters if you leave it on a counter at home. The model is small enough, at roughly 130,000 parameters (the adjustable settings an AI learns during training), to run on the watch itself, which the filing says saves battery and keeps results quick. Everything starts only after you give permission.
This is the 28th Google filing we've tracked since May on wearables that read your body, following applications like one testing new habits and one on radar learning from retries.
Source: US 2026/0363578 A1 at the USPTO (PDF). Full patent text and figures are in the official publication.
Get our take in your Top Stories
Liked this breakdown? Add Patentlyze as a preferred source on Google, and our plain-English take shows up more often in your Top Stories the next time Google patent news breaks.
We read every patent application Big Tech publishes and send you the ones worth knowing. Plain English, free, every week.
The drawings · 7 sheets
7 drawing sheets from US 2026/0363578 A1 · click any drawing to enlarge

No comments yet