Google Files Patent for Radar Gestures That Learn From Your Second Try
Wave at your smart display, get ignored, wave again: Google's patent treats that second try as a free lesson. The device saves the failed attempt so it can recognize your version of the gesture next time.
What Google's self-correcting gesture radar does
Today, a gesture-controlled device either catches your wave or it doesn't. When it misses, you wave again, a little bigger or slower, until it responds. Nothing about that miss helps the device do better next time, so the same thing happens tomorrow.
Google's filing changes that. When a radar sensor sees a motion it can't match to a known gesture, and then you try again within a couple of seconds and the device does recognize it, the device assumes your first attempt was meant to be the same thing. It saves the radar signature of that failed try as another example of how you make that gesture.
Over time, your quirky version of a swipe gets recognized on the first go. There is no training session to sit through. The learning happens in the background as you use the device.
… comparing the first radar-signal characteristic to one or more stored radar-signal characteristics, the comparing not effective to correlate the ambiguous gesture to a first gesture; detecting another gesture performed by the user, the other gesture associated with a second radar-signal characteristic …
Translation: The system initially fails to recognize a confusing first wave pattern until a clearer follow up wave is performed.
How the radar matches a miss to a retry
Google's filing covers a method for a radar-equipped device to learn from its own misses. The device sends out radio waves and reads the reflections as a radar-signal characteristic (basically a fingerprint of a motion's shape and speed over time). It compares each new motion against fingerprints it has stored for known gestures, like a swipe or a tap.
If a motion lands in the gray zone, the filing calls it an ambiguous gesture: the device can't match it to any known gesture with enough confidence, or it matches several about equally. When the user then performs another gesture that the device does recognize, the system concludes the earlier one was the same gesture and stores the fingerprint of the failed attempt too.
The filing adds ways to make that guess safer:
- A time window: the retry counts when it comes two seconds or less later, with no other gesture in between.
- A confidence weight that is higher for quick retries and lower for slow ones.
- Extra context saved with the sample, such as where you stood and which way you faced, or readings from a camera or microphone.
- A prompt asking you to repeat the motion, or a non-gesture command that confirms what you meant.
It also describes watching continually, with no wake-up cue, and letting a second device in the home supply the successful retry.
Techniques and devices for continual in-line learning for radar-based gesture recognition are described in this document. Through continual in-line learning, a computing device may improve recognition of even the hardest-to-recognize gestures by gradually storing characteristics of ambiguous gestures performed by a user.
Translation: Google's new radar system gets better at catching messy hand movements over time by learning from your follow up attempts.
Why gesture controls on Google devices could stick
If you have ever given up on touch-free controls because a device kept ignoring your wave, this is aimed at you. Misses are what make people drop gestures, and the filing says accuracy can improve through normal use, without a separate training session. Because people wave differently, learning from your own retries could fit the device to you.
The description uses a Nest Hub style smart display with 60 GHz radar as an example, and lists watches, TVs, glasses and cars as other possible hosts. Radar also works without a camera, which matters in a bedroom or kitchen. It is a patent filing, so nothing here says any product will do this.
Google's 26th filing we've tracked since May in our wearables that read your body watchlist follows their tips that sense chewing application and one on fixing sleep misreads.
The design bets that a quick second try means the first try was the same gesture. That bet pays off when you wave twice at the same lights, and it can go wrong when you change your mind and do something different. A bad guess gets saved as a lesson, and the device could slowly get worse at the very gesture it was supposed to learn.
The filing hedges against this. It trusts a retry less as more time passes, and it can ask you to repeat the motion or give a non-gesture command to confirm. Each of those hedges costs a little convenience, since the device may occasionally pester you.
The trade is worth it. A gesture feature that misses you three times in a row gets abandoned, and a small risk of mislearning is cheaper than losing the user.
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.
There are more where this came from
We read every patent application Big Tech publishes and send you the ones worth knowing. Plain English, free, every week.
The drawings
46 drawing sheets from US 2026/0349520 A1 · click any drawing to enlarge
Show the other 38 sheets
Want this weekly breakdown for a company we don't cover? Patentlyze Pro →
Be the first to weigh in