A camera in each earbud: University of Washington's VueBuds scored level with Ray-Ban Meta glasses

At a glance

Skip it if: your question depends on colour, since the stills are greyscale.

Status: University of Washington research prototype, presented at CHI 2026; first answers took 2.95 seconds on a desktop Mac.

Hold a Korean noodle packet in front of University of Washington's VueBuds, ask for a translation, and the earbuds answer "cold noodles". This research prototype, from Shyam Gollakota's lab, puts a monochrome camera roughly the size of a grain of rice in each bud of a pair of Sony WF-1000XM3 earbuds. The team presented its paper at CHI 2026 in April, where it received an honorable mention. In that paper, 74 raters compared recorded answers from VueBuds and Ray-Ban Meta across 17 tasks and scored them level: 3.33 against 3.32 out of five.[1] [2] [3]

The camera modules are low-power: they add 3.8 mW when idle and 4.9 mW at 60 queries an hour, and the buds send only 324 by 239 pixel greyscale stills over Bluetooth. The authors model that draw as an 11-14% cut in battery life on Sony or AirPods Pro 2 earbuds; an always-on camera would cost 42%, leaving the Sony buds at 3.5 hours. Angling each camera 5 to 10 degrees outward gives a 98 to 108 degree field of view, the width of the scene the pair can take in.[1] [2]

In the in-person test, 16 participants ran 130 trials, and VueBuds got 94.3% right on reading book titles and authors, 83.8% on translating Korean text and 82.5% on objects. Most misses came down to hardware limits such as glare and fine print. In the rating of recorded answers, raters preferred VueBuds' translations, 4.1 against 2.8 for Ray-Ban Meta; the glasses won on counting, 4.5 to 1.7.[1] [4] [2]

The thinking happens on a desktop: a Mac Mini M4 Pro ran the vision model, the software that interprets each image. The prototype began answering after 2.95 seconds, the wait before its first words; the university's release says about a second, a figure the paper reaches (1.14 seconds) only with optimisations it identifies. Greyscale stills also rule out any question that depends on colour.[1] [2]

Other companies appear to be heading the same way. Qualcomm's chief executive said in June that the company is working on more than 40 AI wearable designs, including earbuds with cameras. In August, TechCrunch reported that a leaked video in a macOS release candidate, spotted by MacRumors, appears to show AirPods with a camera: a man holds up a book and talks with Siri about it, and code seems to warn when hair covers the lens. Timing reports conflict: Bloomberg's May account, as TechCrunch relays it, tied launch to September software, while The Verge's June summary pointed to late 2027.[5] [6] [7]

VueBuds' own privacy case rests on three things: still images instead of video, a spoken wake word and processing kept off the cloud. Bloomberg's description of Apple's version, as TechCrunch relays it, adds an LED that signals when data goes to the cloud. If those cameras ship, check where the answers are computed and how long the first words take.[2] [1] [6]