Toggle search
Toggle menu
5.5K
6.1K
231
53.7K
VR & AR Wiki
Toggle preferences menu
Toggle personal menu
Not logged in
Your IP address will be publicly visible if you make any edits.

Project Aria

vrarwiki.com · VR & AR Wiki
More actions
Project Aria
Basic Info
VR/AR Augmented Reality
Type Research glasses
Creator Meta (Reality Labs Research)
Developer Meta (Reality Labs Research)
Manufacturer Meta
Announcement Date September 16, 2020 (Gen 1); February 27, 2025 (Gen 2)
Release Date 2020 (Gen 1, research distribution only); Gen 2 applications opened October 19, 2025, with a broad rollout to approved researchers targeted for Q2 2026
Price Not for sale (provided to approved researchers)
Website https://www.projectaria.com/
AI Wiki Project Aria
Versions Aria Gen 1, Aria Gen 2
System
Operating System Android 7.1 (Gen 1)
Chipset Qualcomm Snapdragon 835 (Gen 1); custom Meta low-power co-processor (Gen 2)
Storage
Storage 128 GB UFS flash (Gen 1)
Memory 4 GB (Gen 1)
Display
Display None (no see-through display)
Optics
Optics None
Tracking
Tracking Visual-inertial odometry (6DoF), SLAM
Eye Tracking Yes (two inward-facing cameras)
Hand Tracking Yes (Gen 2, on-device)
Audio
Audio Two open-ear force-cancelling speakers (Gen 2); none on Gen 1
Microphone Seven-channel spatial microphone array (both generations); contact microphone in the nosepad (Gen 2)
Camera RGB point-of-view camera, monochrome scene (SLAM) cameras: two on Gen 1, four on Gen 2
Connectivity
WiFi Wi-Fi 6E (Gen 2)
Bluetooth 5.3 (Gen 2)
Battery Capacity 2.5 Wh (Gen 1)
Battery Life About 1.5 hours continuous recording (Gen 1, measured with recording profile 0); 6 to 8 hours continuous use (Gen 2)
Device
Weight About 75 g (Gen 1); 74 to 76 g depending on size (Gen 2)
Sensors RGB camera, monochrome scene cameras, eye-tracking cameras, IMUs, magnetometer, barometer, GNSS, ambient light sensor (Gen 2), PPG heart-rate sensor (Gen 2), proximity sensor (Gen 2)


Project Aria is a research program and wearable sensor device built by Meta through its Reality Labs Research division. The device takes the form of a pair of glasses fitted with cameras and other sensors, but it has no display and shows nothing to the wearer: its purpose is to record egocentric (first-person) data so that researchers can study the perception, mapping and contextual artificial intelligence problems that future augmented reality glasses will need to solve.[1][2] Meta has released two generations: Aria Gen 1 in 2020 and Aria Gen 2, announced on February 27, 2025.[3] Neither generation is sold. As of October 2026, Aria Gen 2 is offered to approved academic and corporate researchers through a rolling application for the Aria Research Kit, after Meta targeted a broad rollout to qualified applicants in the second quarter of 2026.[4][5]

Meta announced the program on September 16, 2020 under its former name Facebook, and at launch stated that the glasses "are not a consumer product, nor are they a prototype".[1] The device records video and audio from the wearer's point of view along with eye-movement and location data, and the resulting datasets are used to train and evaluate algorithms for simultaneous localization and mapping, eye tracking, hand tracking and audio understanding.[6][7]

A first-generation device (Aria Gen 1) was distributed from 2020 to a small group of Meta employees and contractors and later to external academic and corporate partners. Aria Gen 2 adds on-device machine perception, open-ear speakers, a heart-rate sensor and a battery life of six to eight hours.[3] Meta says the glasses are used internally and by more than 200 academic and corporate partners.[2]

Reviewed 17 October 2026. October 2026 additions (new timeline rows, Gen 2 software releases table, tools, case studies and partner contributions) checked against Meta's Project Aria and blog pages, GitHub releases, arXiv papers and partner project pages. About review dates.

Background and purpose

Meta's Reality Labs Research group develops the hardware and software it expects all-day wearable AR glasses to require, including perception that runs on a small, low-power, head-worn device. Project Aria exists to gather the real-world, first-person data needed for that work before any consumer AR product is ready to ship. Because the device only senses and records, and does not render anything for the wearer, Meta's 2023 device paper describes it as an egocentric, multi-modal data recording and streaming device.[7] Meta's 2020 announcement said the glasses carry the full sensor suite used in VR headsets for spatial awareness, and that participants cannot view or listen to the raw data they record.[1]

The 2023 device paper, "Project Aria: A New Tool for Egocentric Multi-Modal AI Research," authored by Jakob Engel, Kiran Somasundaram, Michael Goesele and a large group of Reality Labs Research collaborators, frames the program around egocentric machine perception and contextual AI, with the stated goal of supporting "always available, context-aware and personalized AI applications" for future wearable devices.[7]

One early use was LiveMaps, Meta's concept of a shared, crowd-sourced 3D map of the world that AR glasses would download and then only update where something has changed. The 2020 announcement said Project Aria would test how such a map works in practice.[1]

History

Facebook (renamed Meta in 2021) introduced Project Aria on September 16, 2020. At launch the glasses were worn by about 100 Facebook employees and contractors, primarily in the San Francisco Bay Area and Seattle, who used them to collect data in everyday environments.[6] Alongside the internal rollout, Meta announced a pilot with Carnegie Mellon University's Cognitive Assistance Laboratory to build 3D maps of indoor spaces such as museums and airports, work aimed in part at navigation aids for people with visual impairments.[1] A September 2021 update to the announcement said the device had by then been used in public spaces in the United States for about a year, and that from September 2021 employees and contractors would also record at public landmarks in Singapore.[1]

Over the following years Meta opened the program to outside researchers through the Aria Research Kit and released a series of public datasets recorded with the device. The 2023 device paper formalized the Gen 1 hardware and its data formats for the research community.[7] In October 2025 Meta said that nearly 300 academic labs in 27 countries had used more than 1,000 Gen 1 devices.[4] According to Road to VR, Gen 1 was released in 2024 to third-party researchers across 300 labs in 27 countries.[8]

On February 27, 2025 Meta announced Aria Gen 2, calling it the next generation of egocentric research glasses and describing upgrades to sensing, comfort, on-device computation and audio interactivity.[3][9] In June 2025 Meta published a technical overview of the Gen 2 hardware and showed interactive demos at CVPR 2025 in Nashville; at that point Gen 2 was not yet available to external partners.[10][11]

On October 19, 2025 Meta closed Gen 1 applications (while still processing the backlog) and opened applications for Gen 2, with separate forms for academic and corporate researchers. It targeted a broad rollout of Gen 2 devices to qualified applicants in the second quarter of 2026, and released an Aria Gen 2 Device Whitepaper and the Aria Gen 2 Pilot Dataset ahead of it. Meta also offered Gen 2 demos at ICCV 2025 in Honolulu.[4][8]

Project Aria timeline
Date Event
September 16, 2020 Project Aria announced; about 100 employees and contractors begin recording in the San Francisco Bay Area and Seattle[6]
September 2021 Recording extended to public landmarks in Singapore[1]
August 2023 Gen 1 device paper published on arXiv[7]
2024 Gen 1 released to third-party researchers across 300 labs in 27 countries (per Road to VR)[8]
September 26, 2024 Aria Training and Evaluation Kit (ATEK) released[12]
September 30, 2024 Digital Twin Catalog released, with Aria recordings for a subset of its objects[13]
November 13, 2024 Meta publishes research kit case studies and a call for Aria Research Kit applications[14]
February 27, 2025 Aria Gen 2 announced[3]
June 2025 Gen 2 technical overview published; demos at CVPR 2025[10]
October 16, 2025 Project Aria Tools 2.0.0 adds Gen 2 data support[15]
October 19, 2025 Gen 1 applications close; Gen 2 applications open; pilot dataset and device whitepaper released[4]
April 21, 2026 Meta announces Boxer, an open model for lifting 2D object detections to 3D boxes, with Aria support[16]
Q2 2026 Target for a broad rollout of Gen 2 to qualified applicants[4]
August 13, 2026 Project Aria Tools 2.2.0 adds Meta Neural Band sEMG data access and ML eye-tracking results[17][18]
September 8, 2026 Aria Gen 2 Device OS v23 adds a new on-device eye-tracking model and WebRTC streaming[19]
October 1, 2026 Aria Gen 2 Client SDK v2.5.1 adds externally clocked multi-device recording and streaming[19]

Aria Gen 1

Aria Gen 1 weighs about 75 grams, comes in 147 mm and 152 mm frame widths, and carries five cameras: one rolling-shutter RGB point-of-view camera, two global-shutter monochrome scene cameras used for SLAM, and two inward-facing monochrome eye-tracking cameras.[20] The RGB camera reaches a maximum resolution of 2880x2880 pixels with a 110-degree horizontal field of view through an f-theta fisheye lens, while the two scene cameras run at 640x480 pixels with a 150-degree horizontal field of view.[20] The glasses run Android 7.1 on a Qualcomm Snapdragon 835 with 4 GB of RAM and 128 GB of flash storage.[20]

Non-visual sensing on Gen 1 includes two inertial measurement units (running at 1000 Hz and 800 Hz), a seven-channel spatial microphone array sampled at 48 kHz, a magnetometer, a barometer, GPS, and Wi-Fi and Bluetooth beacon sensing. Meta lists a 2.5 Wh battery that gives about 1.5 hours of continuous recording plus roughly 30 hours of standby, measured with recording profile 0 (two eye-tracking and two SLAM cameras at 10 fps, the RGB camera at 1 fps at 8 MP, plus IMU, Wi-Fi, GPS and seven-channel audio at nominal rates).[20] Meta's comparison table describes the Gen 1 device as "not programmable".[21]

Aria Gen 1 hardware
Component Detail
Processor Qualcomm Snapdragon 835, 4 GB RAM, Android 7.1[20]
Storage 128 GB UFS flash[20]
Cameras 5 total: 1 RGB, 2 monochrome scene (SLAM), 2 eye tracking
RGB camera Up to 2880x2880 px, 110-degree HFOV, rolling shutter, fisheye lens, 30 fps maximum
Scene cameras 640x480 px, 150-degree HFOV, 120-degree VFOV, global shutter, 30 fps maximum
Eye-tracking cameras 2 inward-facing, monochrome, global shutter, 640x480 px, 90 fps maximum[20]
IMUs 2 (1000 Hz and 800 Hz)
Microphones 7-channel spatial array, 48 kHz
Other sensors Magnetometer (10 Hz), barometer (50 Hz), GPS, Wi-Fi and Bluetooth beacons
Weight About 75 g
Battery 2.5 Wh; about 1.5 h continuous recording, about 30 h standby (recording profile 0)[20]
Display None

Aria Gen 2

Aria Gen 2 keeps the roughly 75-gram weight of the first generation (Meta cites about 74 to 76 grams depending on size, across eight size options) and adds foldable arms for easier storage and transport.[3][10] Battery life rises to six to eight hours of continuous use.[3][9]

The sensor suite is expanded relative to Gen 1. Gen 2 uses four computer-vision cameras instead of two, with global shutters, and Meta's technical overview gives a high dynamic range of 120 dB, up from 70 dB on Gen 1, and a stereo overlap of 80 degrees versus 35 degrees on the earlier device.[10] Meta's Gen 2 hardware specification page lists these cameras at 512x512 pixels with a 119 x 119 degree field of view and a dynamic range above 110 dB.[22] It retains an RGB camera, eye-tracking cameras, a spatial microphone array, IMUs, a magnetometer, a barometer and GNSS.[3] New additions include a calibrated ambient light sensor, a contact microphone embedded in the nosepad that helps separate the wearer's voice from that of bystanders, and a photoplethysmography (PPG) sensor, also in the nosepad, that allows estimation of the wearer's heart rate.[10][9] A proximity sensor detects when the glasses are being worn, and an onboard sub-GHz radio aligns timestamps across several Aria devices recording together.[22][10]

A custom low-power co-processor designed by Meta runs machine-perception tasks on the device itself, including 6-degree-of-freedom visual-inertial odometry, eye tracking and articulated hand tracking, as well as on-device speech recognition.[3][10] Gen 2's eye-tracking output covers gaze per eye, the vergence point, blink detection, pupil center and diameter, and corneal center.[10] The glasses also add audio output through open-ear speakers, which lets researchers prototype systems that respond to the wearer in real time, something the sensor-only Gen 1 could not do.[3] Unlike Gen 1, Gen 2 can run applications installed on the device, and it supports cloud streaming in addition to local PC and mobile streaming.[21]

Aria Gen 2 hardware (manufacturer specifications)
Component Detail
RGB camera Sony IMX681, 12 MP, rolling shutter; 4032x3024 at 24 fps; 133 x 99 degree field of view[22]
Computer-vision cameras 4, global shutter, 512x512 px, 119 x 119 degree field of view, dynamic range above 110 dB[22]
Eye-tracking cameras 2, 400x400 px infrared, 8 IR LEDs per eye; gaze output up to 90 Hz[22]
IMUs 2 six-axis units, up to 1600 Hz (800 Hz typical)[22]
Magnetometer and barometer Magnetometer up to 400 Hz; barometer up to 240 Hz[22]
PPG sensor In the nosepad, 128 Hz typical, 512 Hz maximum[22]
Microphones 7 spatial microphones up to 48 kHz; Knowles V2S200D contact microphone in the nosepad[22]
Speakers 2 force-cancelling speakers[22]
Ambient light sensor VD6281, red, green, blue, IR and UVA channels plus a clear channel for flicker detection[22]
GNSS Dual-frequency GPS (L1 + L5) and Galileo (E1 + E5)[22]
Wireless Wi-Fi 6E and Bluetooth 5.3; sub-GHz radio for time alignment (under 10 microseconds offset error)[22]
Co-processor Custom low-power Meta design; H.265 (HEVC) video and Opus audio encoding[22]
Weight 74 to 76 g depending on size; eight size variations[10]
Battery 6 to 8 hours of continuous recording use[22]
Display None[8]
Aria Gen 2 changes versus Gen 1
Feature Aria Gen 1 Aria Gen 2
Computer-vision cameras 2 4
Camera dynamic range 70 dB 120 dB
Stereo overlap 35 degrees 80 degrees
On-device perception No (offline processing) Yes (custom co-processor: VIO, eye and hand tracking, speech)
On-device applications No Yes[21]
Heart-rate (PPG) sensor No Yes (in nosepad)
Contact microphone No Yes (in nosepad)
Speakers No Yes (open-ear)
Battery (continuous) About 1.5 h 6 to 8 h
Arms Fixed Foldable

Role in VR and AR

Project Aria sits upstream of Meta's augmented reality hardware: it is the data-gathering and perception-testbed layer for the technologies that AR and mixed reality glasses depend on, including head and eye tracking, hand tracking, environment mapping and contextual AI. Because the glasses have no display and are never sold, they isolate the sensing and perception problem from the rendering and product problem, letting Meta and outside researchers study how a small head-worn device can localize itself and understand its surroundings without the cost or constraints of a full AR headset.[7][2]

Meta's Project Aria site says that research breakthroughs which "pre-empt products like Meta's Orion" were made possible by the first-generation glasses, referring to the Meta Orion AR glasses prototype.[23] Road to VR noted that Meta calls Aria "research glasses" rather than AI glasses or smart glasses, and framed Gen 2 as a way for Meta to guide a large external research community toward the perception problems future AR glasses must solve, without Meta having to coordinate directly with every team.[8]

Aria also differs in purpose from consumer smart glasses such as Meta's Ray-Ban Meta line, which Meta markets as AI glasses. Aria is a research instrument offered only to approved researchers, and its sensor suite (SLAM cameras, eye-tracking cameras and, on Gen 2, a PPG sensor and contact microphone) is aimed at algorithm development.[8][2][21]

Research uses

Meta's partner pages and announcements describe several external projects built on Aria:[24]

  • Robotics: Georgia Tech's Robotic Learning and Reasoning Lab, led by Danfei Xu, used Aria glasses to record "human data" for tasks a humanoid robot should learn, in a framework called EgoMimic developed by PhD student Simar Kareer. The glasses were also mounted on the robot as its real-time sensor package, streaming through the Aria Client SDK. Meta reported that with 90 minutes of Aria recordings the robot's task performance rose by 400% compared with previous methods. Kareer was scheduled to present the EgoMimic paper at the 2025 IEEE International Conference on Robotics and Automation (ICRA) in Atlanta.[25]
  • Accessibility: Carnegie Mellon's NavCog project used first-generation Aria glasses for indoor navigation; Meta's 2024 case studies said that working with Aria reduced the app's dependence on Bluetooth beacons and let it bring new indoor environments online faster.[14] Separately, Envision is exploring Aria Gen 2 to combine its Ally AI assistant with spatial audio, using on-device SLAM and the onboard speakers to guide blind and low-vision users through indoor spaces. Meta describes the Envision work as exploratory.[3]
  • Automotive: BMW used Aria Gen 1 glasses to study how wearable AR technology could be integrated into vehicles.[24][3]
  • Datasets: the University of Bristol captured egocentric data with Aria for its contributions to the Ego-Exo4D dataset.[24]
  • Hearing research: researchers at the University of Iowa used Aria's microphone array to locate the direction of sounds in environments where people with hearing loss may have difficulty.[14]
  • Driving: IIIT Hyderabad's Driver Intent Prediction Project used Aria eye-gaze data to determine where a driver is looking, as part of accident-prevention research.[14]

Meta's partner page also features datasets, models and tools released by its research community, which Meta encourages to open source the work it develops with Project Aria:[24]

Selected partner contributions listed by Meta
Project Description
EgoMimic Georgia Tech framework for scaling robot manipulation learning from egocentric human demonstrations[25]
EgoVerse A continuously expanded dataset and ecosystem of human demonstrations for robot learning, from a consortium of Georgia Tech, Stanford University, UC San Diego and ETH Zurich labs with industry partners including Meta; its current release lists 4,003 hours of demonstrations covering 1,965 tasks[26]
Hoi! Multimodal dataset for articulated robot manipulation that combines several camera viewpoints with tactile and force sensing[24]
LaMAria City-scale egocentric benchmark for visual-inertial odometry and SLAM, with centimeter-accurate ground truth from surveyed control points; collected with Aria glasses according to Meta and presented at ICCV 2025 by researchers from ETH Zurich, Meta Reality Labs Research and others[24][27]
HD-EPIC Detailed egocentric kitchen video dataset recorded with Aria Gen 1 glasses, with 41 hours of unscripted recordings and a visual question answering benchmark of 26.6K questions; presented at CVPR 2025[24][28]

Datasets and research kit

Researchers access the device through the Aria Research Kit, which Meta offers to selected partners through a rolling application process. The kit now includes Aria Gen 2 glasses, a companion app for starting and stopping recordings and choosing sensor profiles, machine perception that runs on the device or through Meta's Machine Perception Services, a Client SDK for building and testing experiences on the glasses, the open-source Project Aria Tools (projectaria_tools) for visualizing and processing recordings, and the Aria Studio desktop application.[5] Machine Perception Services compute derived data from recordings, such as 6-degree-of-freedom trajectories and 3D eye-gaze estimates.[5]

Project Aria Tools 2.0.0, released on October 16, 2025, put Gen 1 and Gen 2 recordings behind the same C++ and Python APIs and exposed the on-device machine-perception data stored in Gen 2 files.[15] Later releases in 2026 added Gen 2 eye-tracking and proximity-sensor support, and version 2.2.0 (August 13, 2026) added access to Meta Neural Band surface electromyography (sEMG) data with EMG calibration, ML eye-tracking results and a battery status stream; version 2.3.0 followed on August 27, 2026, with changes to Neural Band timestamps and new eye-tracking and sEMG tutorials.[17][18][29] Meta also lists open tools and models built around Aria data, including EgoBlur, SceneScript, EFM3D and Photoreal Reconstruction.[2] EgoBlur is an open-source model that detects and blurs faces and license plates in images and video, with separate models for Gen 1 and Gen 2 data.[19] The Aria Training and Evaluation Kit (ATEK), released in September 2024, converts Aria's VRS recording format into PyTorch-compatible data, offers preprocessed datasets and evaluation libraries, and came with a Project Aria space on Hugging Face.[12]

Photoreal Reconstruction is a method from Reality Labs Research that rebuilds real scenes from Aria RGB video as a Gaussian splatting representation. Its paper, by Zhaoyang Lv and colleagues and accepted as a SIGGRAPH 2025 conference paper, calibrates the rolling-shutter RGB camera with visual-inertial bundle adjustment and adds a physical image formation model; the authors evaluated it on the Project Aria device under varied indoor and outdoor lighting and also validated it on a Meta Quest 3.[30] Meta released an accompanying Aria Scenes benchmark of 12 VRS recordings from indoor and outdoor environments.[31] In April 2026 Meta released Boxer, a model that lifts open-world 2D bounding boxes into metric 3D oriented bounding boxes; Meta reported results on the Aria Digital Twin and NymeriaPlus datasets as well as external benchmarks, and the code ships with built-in Project Aria support.[16][32]

Aria Gen 2 software releases

The Gen 2 research kit is a set of separately versioned components. Meta's releases page listed the following latest versions as of October 2026:[19]

Aria Gen 2 software components (latest versions listed by Meta, October 2026)
Component Version Date Notes
Device OS v23 September 8, 2026 New on-device eye-tracking model with dedicated recording and streaming profiles; WebRTC recording and streaming; users are prompted to recalibrate eye tracking after the update
Companion App v285 September 7, 2026 Controls for remote WebRTC streaming; more visible eye-tracking recalibration prompt
Client SDK v2.5.1 October 1, 2026 Multi-device recording and streaming from an external clock source with one Companion App connection; Linux multi-device USB fix
Machine Perception Services (MPS CLI) v1.4.0 September 2, 2026 Eye-gaze processing for Gen 2 recordings, with optional personal calibration
Project Aria Tools v2.3.0 August 25, 2026 (GitHub release published August 27, 2026)[29] Gen 2 default central pupil frame aligned with eye-gaze output; eye-tracking tutorials
VRS Health Check v1.3.1 August 25, 2026 Duration-aware thresholds and reporting of consecutive sample drops
Aria Studio v2.0.1 September 1, 2026 Shared sign-in through Project Aria Login; more reliable Gen 2 imports
Aria Gen 2 Pilot Dataset Loader v1.1.0 August 16, 2026 Updated for Project Aria Tools 2.2.0 and Rerun SDK 0.33.0

The same page lists Timecode Bridge, an open-source hardware companion that puts Aria Gen 2 captures on an external LTC master clock so that glasses data can be aligned with motion capture, cameras and audio equipment.[19]

The program has released several public egocentric datasets recorded with Aria glasses:

Selected Project Aria datasets
Dataset Device Focus
Aria Everyday Activities (AEA) Gen 1 Everyday first-person activity recordings; formerly the Aria Pilot Dataset[33]
Aria Digital Twin (ADT) Gen 1 Ground-truth scenes (200 sequences across two instrumented indoor environments, 398 object instances)[34]
Aria Synthetic Environments (ASE) Gen 1 Procedurally generated synthetic indoor environments for machine-learning training[33]
Aria Everyday Objects Gen 1 Small real-world dataset with static 3D oriented bounding box annotations[33]
Ego-Exo4D Gen 1 First- and third-person skilled activity, a collaboration between Meta FAIR, Project Aria and 15 university partners; Meta's documentation cites 1,422 hours of video from more than 800 participants in 13 cities[35]
MMCSG Gen 1 Two-sided conversations recorded on Aria glasses with multi-channel audio, video and IMU data, for speech recognition and speaker diarization research (released February 2024)[36]
HOT3D Gen 1 (and Meta Quest 3) 3D hand and object tracking (more than 3.7M images, over 833 minutes, 19 subjects, 33 objects)[37]
Nymeria Gen 1 Full-body human motion in the wild (300 hours, 264 participants, 50 locations)[38]
Digital Twin Catalog (DTC) Gen 1 3D object reconstruction: over 2,400 sub-millimeter-accurate 3D object models, with DSLR and Aria recordings for a subset of the objects (released September 30, 2024)[33][13]
Reading Recognition in the Wild Gen 1 Over 100 hours of real-world data with high-frequency eye tracking[33]
Aria Gen 2 Pilot Dataset (A2PD) Gen 2 A primary wearer and three friends in five scenarios (cleaning, cooking, eating, playing, outdoor walking), with raw sensor data and machine-perception outputs; released in batches under CC BY-NC 4.0[39]

Meta describes the Ego-Exo4D dataset, collected with first-generation Aria glasses, as a foundational resource across modern computer vision and robotics research.[3][35]

The Aria Gen 2 Pilot Dataset was described in an October 2025 arXiv paper by Chen Kong, James Fort, Aria Kang and colleagues. Beyond the raw streams, it includes eye tracking, hand tracking and visual-inertial odometry computed on the device during every recording, offline SLAM and hand tracking from Machine Perception Services, and outputs from algorithms such as speaker diarization, directional speech recognition that separates the wearer's speech from others, hand-object interaction segmentation and NVIDIA's FoundationStereo depth estimation. The paper notes that ground-truth heart-rate measurements were not available for the first release, so the PPG-based heart-rate estimates were not validated against them.[4][40][39]

Privacy and responsible use

Because the device continuously records video, audio and location data in real environments, Meta published a set of responsible-innovation measures alongside the 2020 launch. The glasses show a white light while recording and have a physical mute control that stops data collection when pressed, and participants can delete recorded segments through a companion phone app. Captured data is encrypted, kept in a separate designated storage space, and held under a quarantine period (Meta cited three days at launch) before researchers can access it, and data captured in public places is automatically processed to blur faces and vehicle license plates.[6] Meta stated that wearers were trained and were not to record in private spaces such as restrooms, prayer rooms or locker rooms, or during sensitive meetings, and that recording in stores, restaurants and similar private venues required the owner's consent. Participants wore clothing and a lanyard badge identifying them as part of the project, and Meta said the data would not be used to inform ads.[6][1]

Current status

As of October 2026 Project Aria remains a research-only program. Aria Gen 1 applications closed on October 19, 2025, and Meta's Project Aria site lists Aria Gen 2 as available through the rolling Aria Research Kit application, whose kit includes Gen 2 glasses; Meta had targeted a broad rollout of Gen 2 to qualified applicants in the second quarter of 2026.[4][5][21] Meta has not announced any consumer sale of either generation; the glasses are positioned as a tool for the perception and contextual-AI research that underpins future AR glasses.[3][2]

See also

References

  1. ↑ 1.0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 "Announcing Project Aria: a research project on the future of wearable AR". Tech at Meta. September 16, 2020. https://tech.facebook.com/reality-labs/2020/9/announcing-project-aria-a-research-project-on-the-future-of-wearable-ar/. Retrieved 2026-06-21.
  2. ↑ 2.0 2.1 2.2 2.3 2.4 2.5 "Aria Gen 2, from Meta - Project Aria". Project Aria (Meta). https://www.projectaria.com/. Retrieved 2026-06-21.
  3. ↑ 3.00 3.01 3.02 3.03 3.04 3.05 3.06 3.07 3.08 3.09 3.10 3.11 3.12 "Introducing Aria Gen 2: Unlocking New Research in Machine Perception, Contextual AI, Robotics, and More". Meta. February 27, 2025. https://www.meta.com/blog/project-aria-gen-2-next-generation-egocentric-research-glasses-reality-labs-ai-robotics/. Retrieved 2026-06-21.
  4. ↑ 4.0 4.1 4.2 4.3 4.4 4.5 4.6 "Aria Gen 2 Applications Are Now Open". Meta. October 19, 2025. https://www.meta.com/blog/aria-gen-2-updates/. Retrieved 2026-06-21.
  5. ↑ 5.0 5.1 5.2 5.3 "Project Aria Research Kit". Project Aria (Meta). https://www.projectaria.com/research-kit/. Retrieved 2026-10-14.
  6. ↑ 6.0 6.1 6.2 6.3 6.4 "Privacy Matters: Project Aria". Meta Newsroom. September 16, 2020. https://about.fb.com/news/2020/09/privacy-matters-project-aria/. Retrieved 2026-06-21.
  7. ↑ 7.0 7.1 7.2 7.3 7.4 7.5 "Project Aria: A New Tool for Egocentric Multi-Modal AI Research". arXiv. August 2023. https://arxiv.org/abs/2308.13561. Retrieved 2026-06-21.
  8. ↑ 8.0 8.1 8.2 8.3 8.4 8.5 Scott Hayden (October 29, 2025). "Meta to Ship Project Aria Gen 2 to Researchers in 2026, Paving the Way for Future AR Glasses". Road to VR. https://roadtovr.com/meta-to-ship-project-aria-gen-2-to-researchers-in-2026-paving-the-way-for-future-ar-glasses/. Retrieved 2026-10-14.
  9. ↑ 9.0 9.1 9.2 "Meta's new AR glasses for research can measure heart rate". TechCrunch. February 27, 2025. https://techcrunch.com/2025/02/27/metas-new-ar-glasses-for-research-can-measure-heart-rate/. Retrieved 2026-06-21.
  10. ↑ 10.0 10.1 10.2 10.3 10.4 10.5 10.6 10.7 10.8 "Inside Aria Gen 2: Explore the cutting-edge tech under the hood". Meta AI. June 4, 2025. https://ai.meta.com/blog/aria-gen-2-research-glasses-under-the-hood-reality-labs/. Retrieved 2026-10-14.
  11. ↑ Scott Hayden (June 5, 2025). "A Look Inside Meta's 'Aria' Research Glasses Shows What Tech Could Come to Future AR Glasses". Road to VR. https://roadtovr.com/meta-aria-2-glasses-future-peek-ar-glasses/. Retrieved 2026-10-14.
  12. ↑ 12.0 12.1 "Introducing the Aria Training and Evaluation toolkit (ATEK)". Project Aria (Meta). September 26, 2024. https://www.projectaria.com/news/aria-training-and-evaluation-toolkit-atek/. Retrieved 2026-10-17.
  13. ↑ 13.0 13.1 "Introducing the Digital Twin Catalog from Reality Labs Research". Meta AI. September 30, 2024. https://ai.meta.com/blog/digital-twin-catalog-3d-reconstruction-shopify-reality-labs-research/. Retrieved 2026-10-17.
  14. ↑ 14.0 14.1 14.2 14.3 "Project Aria Research Kit: Case Studies + a Call for Applications". Meta. November 13, 2024. https://www.meta.com/blog/project-aria-research-kit-case-studies/. Retrieved 2026-10-17.
  15. ↑ 15.0 15.1 "Release 2.0.0 - Major Version Release - facebookresearch/projectaria_tools". GitHub. Meta. October 16, 2025. https://github.com/facebookresearch/projectaria_tools/releases/tag/2.0.0. Retrieved 2026-10-14.
  16. ↑ 16.0 16.1 "Bridging the Gap: Introducing Boxer for Open-World 2D-to-3D Lifting". Project Aria (Meta). April 21, 2026. https://www.projectaria.com/news/introducing-boxer/. Retrieved 2026-10-17.
  17. ↑ 17.0 17.1 "Release 2.2.0 - facebookresearch/projectaria_tools". GitHub. Meta. August 13, 2026. https://github.com/facebookresearch/projectaria_tools/releases/tag/2.2.0. Retrieved 2026-10-14.
  18. ↑ 18.0 18.1 "Project Aria Tools changelog - Aria Gen 2 Docs". Project Aria Docs (Meta). https://facebookresearch.github.io/projectaria_tools/gen2/research-tools/projectariatools/changelog. Retrieved 2026-10-14.
  19. ↑ 19.0 19.1 19.2 19.3 19.4 "Releases - Project Aria". Project Aria (Meta). https://www.projectaria.com/releases/. Retrieved 2026-10-17.
  20. ↑ 20.0 20.1 20.2 20.3 20.4 20.5 20.6 20.7 "Hardware Specifications - Aria Gen 1 Docs". Project Aria Docs (Meta). https://facebookresearch.github.io/projectaria_tools/docs/tech_spec/hardware_spec. Retrieved 2026-06-21.
  21. ↑ 21.0 21.1 21.2 21.3 21.4 "Introducing Aria Gen 2 glasses". Project Aria (Meta). https://www.projectaria.com/glasses/. Retrieved 2026-10-14.
  22. ↑ 22.00 22.01 22.02 22.03 22.04 22.05 22.06 22.07 22.08 22.09 22.10 22.11 22.12 22.13 22.14 "Hardware Specifications - Aria Gen 2 Docs". Project Aria Docs (Meta). https://facebookresearch.github.io/projectaria_tools/gen2/technical-specs/device/hardware. Retrieved 2026-10-14.
  23. ↑ "About Project Aria". Project Aria (Meta). https://www.projectaria.com/our-work/. Retrieved 2026-10-14.
  24. ↑ 24.0 24.1 24.2 24.3 24.4 24.5 24.6 "Partner with Project Aria". Project Aria (Meta). https://www.projectaria.com/partners/. Retrieved 2026-10-14.
  25. ↑ 25.0 25.1 "EgoMimic: Georgia Tech PhD student uses Project Aria Research Glasses to help train humanoid robots". Meta AI. February 19, 2025. https://ai.meta.com/blog/egomimic-project-aria-georgia-tech-ego4d-robotics-embodied-ai/. Retrieved 2026-10-14.
  26. ↑ "EgoVerse". EgoVerse. https://egoverse.ai/. Retrieved 2026-10-17.
  27. ↑ Anusha Krishnan, Shaohui Liu, Paul-Edouard Sarlin, et al. (September 30, 2025). "Benchmarking Egocentric Visual-Inertial SLAM at City Scale". arXiv (ICCV 2025). https://arxiv.org/abs/2509.26639. Retrieved 2026-10-17.
  28. ↑ "HD-EPIC". HD-EPIC project page. https://hd-epic.github.io/. Retrieved 2026-10-17.
  29. ↑ 29.0 29.1 "Release 2.3.0 - facebookresearch/projectaria_tools". GitHub. Meta. August 27, 2026. https://github.com/facebookresearch/projectaria_tools/releases/tag/2.3.0. Retrieved 2026-10-14.
  30. ↑ Zhaoyang Lv, Maurizio Monge, Ka Chen, Yufeng Zhu, Michael Goesele, Jakob Engel, Zhao Dong, Richard Newcombe (June 4, 2025). "Photoreal Scene Reconstruction from an Egocentric Device". arXiv (SIGGRAPH 2025 Conference Paper). https://arxiv.org/abs/2506.04444. Retrieved 2026-10-17.
  31. ↑ "Photoreal Reconstruction". Project Aria (Meta). https://www.projectaria.com/photoreal-reconstruction/. Retrieved 2026-10-17.
  32. ↑ Daniel DeTone, Tianwei Shen, Fan Zhang, Lingni Ma, Julian Straub, Richard Newcombe, Jakob Engel (April 6, 2026). "Boxer: Robust Lifting of Open-World 2D Bounding Boxes to 3D". arXiv. https://arxiv.org/abs/2604.05212. Retrieved 2026-10-17.
  33. ↑ 33.0 33.1 33.2 33.3 33.4 "Open Datasets - Project Aria". Project Aria (Meta). https://www.projectaria.com/datasets/. Retrieved 2026-10-14.
  34. ↑ Xiaqing Pan, Nicholas Charron, Yongqian Yang, et al. (June 2023). "Aria Digital Twin: A New Benchmark Dataset for Egocentric 3D Machine Perception". arXiv. https://arxiv.org/abs/2306.06362. Retrieved 2026-10-14.
  35. ↑ 35.0 35.1 "Ego-Exo4D Dataset - Aria Gen 1 Docs". Project Aria Docs (Meta). https://facebookresearch.github.io/projectaria_tools/docs/open_datasets/ego-exo4d. Retrieved 2026-06-21.
  36. ↑ "MMCSG Dataset". Meta AI. February 13, 2024. https://ai.meta.com/datasets/mmcsg-dataset/. Retrieved 2026-10-14.
  37. ↑ Prithviraj Banerjee, Sindi Shkodrani, Pierre Moulon, et al. (June 2024). "Introducing HOT3D: An Egocentric Dataset for 3D Hand and Object Tracking". arXiv. https://arxiv.org/abs/2406.09598. Retrieved 2026-10-14.
  38. ↑ Lingni Ma, Yuting Ye, Fangzhou Hong, et al. (June 2024). "Nymeria: A Massive Collection of Multimodal Egocentric Daily Motion in the Wild". arXiv. https://arxiv.org/abs/2406.09905. Retrieved 2026-10-14.
  39. ↑ 39.0 39.1 "Aria Gen 2 Pilot Dataset". Project Aria (Meta). https://www.projectaria.com/datasets/gen2pilot/. Retrieved 2026-10-14.
  40. ↑ Chen Kong, James Fort, Aria Kang, et al. (October 17, 2025). "Aria Gen 2 Pilot Dataset". arXiv. https://arxiv.org/abs/2510.16134. Retrieved 2026-10-14.
Retrieved from "https://vrarwiki.com/index.php?title=Project_Aria&oldid=52505"