Skip to content
Subscribe · $99.00

Sector · Data collection for AI

The machine never saw a pair of hands work. The glasses show it

Almost everything on this page is a dataset, not a product. Human first-person video is what robot learning was missing, and the glasses are the cheapest way to record it. Also here: the day Amazon announced its delivery glasses, and what the Chinese industry signed.

A robot arm can be taught by teleoperation: a person drives it, the machine copies. It works, and it is slow and expensive, because every minute of training costs a minute of someone sitting at the controls. The alternative is to record a person doing the task with their own hands, from their own eyes, and let the model learn from that. For twenty years nobody had that footage at scale. Glasses with a camera, gaze tracking and hand tracking are how it started to exist.

Meta's Project Aria is the visible engine. It is not for sale: Meta lends the glasses to research partners, more than 290 of them running over 1,000 devices in 27 countries, and the partners have logged more than 8,000 hours. Out of that came Ego-Exo4D, where 740 people cooked, played sport, repaired bicycles and performed health procedures with Aria glasses synced to four external cameras, publishing 1,286 hours. And EgoVerse, a consortium of Georgia Tech, Stanford, UC San Diego, ETH Zurich, MIT, Meta, Mecka AI and Scale AI: 1,362 hours, 1,965 tasks, 240 scenes, 2,087 demonstrators. Microsoft built a different shape with HoloAssist, where 222 people worked in instructor and performer pairs and the performer's headset recorded seven synchronised streams.

The one number that measures a result

Nearly all of these projects publish hours, not outcomes. The exception is EgoMimic at Georgia Tech: about 90 minutes of everyday human tasks recorded on Aria glasses, co-trained with a low-cost bimanual manipulator, reported as a 400% improvement over training on teleoperation data alone. Read that for what it is: a benchmark on the tasks the lab chose, published at CoRL 2024 and ICRA 2025 with the hardware, dataset and code open. It is not a robot working in a warehouse.

The part that is not research

Amazon showed AI glasses for delivery drivers at its Milpitas hub on 22 October 2025. They read packages, guide the route, flag hazards and dogs, and take the proof-of-delivery photo without the driver touching a phone. Every one of those readings is also training data. The announcement came one day after the New York Times reported internal plans to avoid hiring more than 600,000 workers by 2033, with an internal goal of automating up to 75% of operations. Amazon published no productivity figure with the glasses.

The industry knows how this looks. In June 2026 a Chinese alliance including Lebird, Rokid and ZTE signed a 15-rule self-regulation pact on trustworthy vision for AI glasses: clear notice and consent before a sensor turns on, face, iris and voice processed locally where possible, data compliance along the chain. It is a pact, not a law. In the same season Chinese education authorities banned AI glasses from gaokao exam rooms, treating mere possession, powered on or not, as cheating.

Cited cases

  • Who
    Meta Reality Labs Research, Project Aria partner program
    What they did
    Meta lends its Project Aria research glasses (egocentric camera, SLAM, eye tracking and hand tracking) to more than 290 academic and industrial partners, who operate over 1,000 devices in 27 countries to record first-person video used in machine perception, contextual AI and robot learning.
    What came out of it
    More than 8,000 hours of egocentric data recorded by the partners; the platform became the standard capture base for the datasets listed below (Ego-Exo4D, EgoMimic, EgoVerse). Aria Gen 2 was announced in Feb 2025, with applications open to researchers and a larger batch expected in Q2 2026.

    www.projectaria.com

  • Who
    Meta FAIR and Project Aria, with 15 partner universities (including the University of Bristol)
    What they did
    740 people recorded skilled human activities (cooking, sport, music, bicycle repair, health procedures) wearing Aria glasses synchronised with 4 external GoPro cameras, forming the egocentric plus exocentric Ego-Exo4D dataset for training multimodal perception models.
    What came out of it
    1,286+ hours of video published (released Nov 2023, expanded Mar 2024); it became a reference benchmark for egocentric video understanding and for robot learning research.

    ai.meta.com

  • Who
    Simar Kareer, PhD candidate at the Georgia Tech Robotic Learning and Reasoning Lab (EgoMimic)
    What they did
    Recorded about 90 minutes of everyday human tasks using Project Aria glasses (egocentric video plus 3D hand tracking) and combined that data with a low-cost bimanual manipulator to co-train a robot imitation policy, instead of relying on teleoperation alone.
    What came out of it
    A 400% improvement in robot performance on the tested tasks compared with training on teleoperation data alone; presented at CoRL 2024 and ICRA 2025, with hardware, dataset and code open.

    ai.meta.com

  • Who
    Georgia Tech, Stanford, UC San Diego, ETH Zurich, MIT, Meta Reality Labs, Mecka AI and Scale AI (EgoVerse dataset)
    What they did
    A consortium of universities and companies built EgoVerse, a collaborative egocentric manipulation dataset of 1,362 hours, 1,965 tasks, 240 scenes and 2,087 demonstrators, using Project Aria glasses as the standard capture platform (21-point hand pose and 6-DoF head pose) to train robotic manipulation policies.
    What came out of it
    Published in April 2026 as one of the largest egocentric datasets ever built specifically for robot learning.

    arxiv.org

  • Who
    Microsoft Research (HoloAssist dataset)
    What they did
    222 participants formed 350 instructor and performer pairs: the person performing the task wore a mixed reality headset that captured 7 synchronised data streams (egocentric video, gaze, hand pose, IMU, audio) while carrying out 20 object manipulation tasks under live remote instruction from a human.
    What came out of it
    More than 160 hours of egocentric interaction video published, with benchmarks for error detection, intervention prediction and hand motion forecasting, as input for physical AI assistants.

    www.microsoft.com

  • Who
    Ege Ozsoy, Arda Mamur and team (Technical University of Munich), EgoExOR dataset
    What they did
    Recorded 94 minutes of two simulated spinal surgical procedures (ultrasound-guided needle insertion and minimally invasive surgery), combining egocentric video from wearable glasses (RGB, gaze, hand tracking, audio) with exocentric RGB-D cameras and ultrasound imaging, annotating scene graphs with 36 entities and 22 relations.
    What came out of it
    The dataset was accepted at NeurIPS 2025 and published as a resource for training real-time surgical assistance AI and for surgeon skill assessment.

    arxiv.org

  • Who
    Amazon
    What they did
    Unveiled AI smart glasses for delivery drivers at its Milpitas Delivery Hub. The glasses read packages, guide routes, flag hazards and dogs in yards, and capture the proof-of-delivery photo without the driver using a phone. Every reading, navigation step and photo also becomes training data, alongside the Blue Jay robotics platform and the AI operations manager called Project Eluna.
    What came out of it
    Announced on 22 Oct 2025, one day after a New York Times report on internal automation plans to avoid hiring more than 600,000 workers by 2033, with an internal goal of automating up to 75% of operations.

    ppc.land

  • Who
    China's AI glasses industry alliance (with Lebird, Rokid and ZTE taking part)
    What they did
    At the 2026 AI Glasses Ecosystem Conference the alliance launched a Self-Regulation Pact on Trustworthy Vision for AI Glasses, with 15 rules covering user privacy protection, technical control at the terminal (clear notice and authorisation before a sensor is activated, local processing of face, iris and voice by preference) and data compliance along the whole chain, in answer to the worry that camera glasses become a tool of blanket surveillance.
    What came out of it
    The pact was adopted by the main manufacturers; in parallel, Chinese education authorities banned AI glasses from 2026 gaokao exam rooms, treating mere possession of the device, powered on or not, as cheating.

    finance.sina.com.cn

  • Who
    Mentra (MentraOS, an open system for smart glasses)
    What they did
    An open platform with a Miniapp Store that runs on compatible glasses from several brands, distributing apps such as live captioning, an AI note taker and a proactive AI mode that continuously processes the wearer's first-person camera and audio feed through third-party models (OpenRouter for Google, OpenAI, Anthropic, xAI, DeepSeek).
    What came out of it
    Raised USD 8 million and launched MentraOS 2.0 with a public app store and an open SDK, letting any developer build AI apps that consume egocentric data on ordinary glasses hardware.

    gamesbeat.com

Devices used

Talk to us

Back to the use cases

A portal about smart glasses.

Pricing

$99.00 per month

R$ 549,00 per month · The currency follows the account language.

Price checked on

2026 · All rights reserved.