Context-aware computing
More actions
Context-aware computing is the design of computer systems that sense information about the situation in which they are used, such as the user's location, the people and devices nearby, the user's activity and the time, and change their behavior in response. Bill Schilit, Norman Adams and Roy Want of Xerox PARC described such software in 1994 as adapting "according to the location of use, the collection of nearby people, hosts, and accessible devices, as well as to changes to such things over time".[1] The idea grew out of ubiquitous computing research in the early 1990s, and much of its early vocabulary, including the most widely used definition of "context", came from work by Anind Dey and Gregory Abowd at the Georgia Institute of Technology.[2][3]
Context-awareness is closely tied to augmented reality (AR). Location has long been one of the most common context sources in AR, as in Columbia University's GPS-tracked Touring Machine campus guide of 1997. In 2017 Jens Grubert, Tobias Langlotz, Stefanie Zollmann and Holger Regenbrecht proposed "Pervasive Augmented Reality": a continuous, multi-purpose AR interface that is "aware of and responsive to the user's context" rather than an application opened for a single task.[4][5] Current headsets and smart glasses collect several kinds of context, including room geometry, the presence of people and what the wearer is looking at, and use it to place content, keep users safe and drive AI assistants.
Definitions
Context-aware software
Schilit, Adams and Want framed context-awareness as a property of "mobile distributed computing", in which a person uses many mobile, stationary and embedded computers over the course of a day. In that setting the execution environment changes constantly: the processors, input and display devices, network capacity and costs available to a task vary with time and place, and the user moves between locations and social situations.[1] They named three important aspects of context: "where you are, who you are with, and what resources are nearby". Context, they wrote, covers more than location, and includes lighting, noise level, network connectivity, communication costs and bandwidth, and "even the social situation; e.g., whether you are with your manager or with a co-worker".[1]
In his 2001 paper "Understanding and Using Context", Dey credits Bill Schilit and Marvin Theimer's 1994 IEEE Network article as the work that first introduced the term "context-aware", and notes that it described context as location, the identities of nearby people and objects, and changes to those objects.[2] That article, "Disseminating active map information to mobile hosts", described an active map service that supports context-aware computing by telling clients about located objects and how they change over time, and dealt mainly with the communication overload that occurs when many people move through the same region while many clients want to know about their movements.[6]
Dey and Abowd's definitions
Dey, and Dey with Abowd and Salber, argued that definitions which list examples of context, or which simply call it the "environment" or "situation", are hard to apply to a new type of information.[2][7] Dey's 2001 definition of context reads:
Context is any information that can be used to characterize the situation of an entity. An entity is a person, place, or object that is considered relevant to the interaction between a user and an application, including the user and applications themselves.[2]
The same paper defines the property itself: "A system is context-aware if it uses context to provide relevant information and/or services to the user, where relevancy depends on the user's task."[2] Dey illustrated the test with an indoor mobile tour guide. Weather does not affect a guide used indoors, so it is not context for that application, while the presence of other people is context because a visitor traveling with others may be interested in different sites.[2] A 2014 survey of context-aware computing for the Internet of Things by Charith Perera and colleagues describes the definitions of context and context-awareness proposed by Abowd et al. in 1999 as the ones "widely accepted by the research community today".[3]
In their 2001 Human-Computer Interaction journal paper, Dey, Abowd and Daniel Salber added that context-awareness "represents a generalized model of input, including both implicit and explicit input". Their definition deliberately does not distinguish between context that is sensed automatically and context that a user enters by hand, because in the real world most context cannot be sensed automatically and has to be supplied by the user.[7]
Categories of context
Several classifications are in common use.
| Source | Classification | Details |
|---|---|---|
| Schilit, Adams and Want (1994), as summarized by Dey, Abowd and Salber | Three environments | Computing environment (available processors, devices for input and display, network capacity, connectivity and costs); user environment (location, nearby people, social situation); physical environment (lighting and noise level)[7] |
| Dey, Abowd and Salber (2001) | Entities | Places (rooms, offices, buildings, streets), people (individuals or groups, co-located or distributed) and things (physical objects or software components such as an application or a file)[7] |
| Dey, Abowd and Salber (2001) | Four primary categories | Identity, location (including orientation, elevation and spatial relations such as proximity or containment), status or activity (for example a room's temperature or a person's tiredness or current activity) and time[7] |
| Baldauf and Dustdar (2004) | Sensor types | Physical sensors (hardware such as cameras, microphones, accelerometers, GPS or the Active Badge system), virtual sensors (software sources such as an electronic calendar or keyboard and mouse activity) and logical sensors that combine the two with other information, following a classification by Indulska and Sutton[8] |
| Perera et al. (2014) | Operational perspective | Primary context is retrieved without using other context (for example location data from GPS); secondary context is computed from primary context[3] |
Dey, Abowd and Salber chose their four categories because other context can be inferred from them. A person's phone number can be derived from her identity, and knowing that a room is occupied, how many people are in it, where they are and whether they are talking can indicate that a meeting is taking place.[7]
Types of context-aware applications
Schilit, Adams and Want sorted applications along two dimensions: whether the task is getting information or carrying out a command, and whether it happens manually or automatically. Each of the resulting four categories was prototyped on the PARCTAB, a palm-sized wireless computer with a 128 x 64 pixel display that communicated over an infrared network in which each room was one cell, sharing 19,200 bps among the tabs in it.[1]
| Category | Meaning | Example from the 1994 paper |
|---|---|---|
| Proximate selection | Nearby located objects are emphasized or made easier to choose | Printer dialogs sorted or highlighted by distance from the user |
| Automatic contextual reconfiguration | Components or connections of the system change with context | Entering a room binds the tab to that room's shared "virtual whiteboard" |
| Contextual information and commands | Queries and commands give different results depending on where they are issued | A location browser showing files tied to the current room, such as coffee-making directions near the kitchen |
| Context-triggered actions | Simple IF-THEN rules run commands automatically | "Watchdog", which ran Unix commands on Active Badge events, and "Contextual Reminders" such as "the next time I see Marvin" |
All rows: Schilit, Adams and Want (1994).[1]
The authors also listed open problems: rapid changes in context can make constant adaptation distracting or impractical, adaptations may confuse users if context is reported incorrectly or if they do not know what context an application uses, and rule-based triggers must balance timely execution against predictable behavior.[1] Dey later proposed a simpler grouping of three features: presentation of information and services to a user, automatic execution of a service, and tagging of context to information for later retrieval.[2]
History
Mark Weiser's ubiquitous computing program at Xerox PARC is the background to the field and is covered in Ubiquitous computing and Mark Weiser. Schilit, Adams and Want described their work as inspired by that vision and by the PARCTAB, which they called "the first mobile computing system built to explore and exploit context-aware software".[1]
The Active Badge system at Olivetti Research in Cambridge, England, published by Roy Want, Andy Hopper, Veronica Falcão and Jonathan Gibbons in 1992, was one of the indoor location systems that early context-aware applications built on; the PARC "Watchdog" program listed in the table above reacted to Active Badge events. Each badge emitted a unique pulse-width modulated infrared code for about a tenth of a second every 15 seconds, which a network of sensors around the building picked up, and the main application helped a telephone receptionist route calls to wherever staff members were.[9] According to Schilit, Adams and Want, the original Olivetti system did not take context into account; badge wearers asked to control call forwarding by who they were with, where they were and the time of day, and "personal control scripts" were added in a later version.[1]
At Georgia Tech, Abowd's Future Computing Environments group, formed in April 1995, built Cyberguide, a family of handheld tour-guide prototypes that used the visitor's current location and history of past locations. The indoor version ran on the Apple MessagePad for visitors to monthly open houses at the GVU Center and located users with an array of television remote controls hung from the ceiling as infrared beacons; an outdoor version attached a Trimble GPS unit to the MessagePad's serial port. One outdoor prototype, CyBARguide, covered about 12 square miles of midtown Atlanta. The architecture divided a tour guide into four roles: cartographer (maps), librarian (information), navigator (positioning) and messenger (communications).[10] Dey later called the indoor mobile tour guide "the canonical context-aware application".[2]
Wearable computing brought the sensors onto the body. In 1998 Thad Starner, Bernt Schiele and Alex Pentland of the MIT Media Lab presented an assistant that tracked a wearer's location and current task through computer vision from body-mounted cameras, without infrastructure off the body, arguing that wearable computers could "see" and "hear" as the user does.[11] Later research on first-person sensor data is covered in Egocentric vision. In 1999 Salber, Dey and Abowd introduced the Context Toolkit at the CHI conference, with "context widgets" that mediate between the environment and an application in the way graphical widgets mediate between the user and an application.[12] Later surveys extended the concept to the Internet of Things. Perera and colleagues evaluated 50 research and commercial projects from 2001 to 2011 and argued that, with billions of sensors connected to the Internet, context-awareness would decide which sensor data needs to be processed at all.[3]
Architectures and toolkits
The Context Toolkit
Dey, Abowd and Salber modeled their framework on graphical user interface toolkits, with the aim of separating the acquisition of context from its use, much as GUI toolkits separate applications from input devices.[7] It has five kinds of component:
- Context widgets give applications access to one piece of context, such as who is present in a room, its temperature or sound level, through queries and callbacks. Unlike GUI widgets they run independently of any application, can serve several applications at once and keep a history of the context they sense.
- Interpreters raise the level of abstraction, for example turning coordinates into a street name or combining room occupancy and sound level to infer that a meeting is under way.
- Aggregators collect related context about one entity, such as a person, into a single component that applications can query.
- Services act on the environment through actuators, for example turning on a light or displaying a message.
- Discoverers keep a registry of the components that are running and what they can do.[7]
The toolkit was written in Java, with components running as separate processes that exchanged XML messages over HTTP so that devices such as wearable computers, pagers and handhelds only needed TCP/IP support. One application, an In/Out Board, used a location widget fed either by Dallas Semiconductor iButtons docked into a reader or by the PinPoint 3D-iD radio-frequency indoor positioning system, plus an interpreter that converted badge or tag IDs into names.[7]
Layered architectures and the context life cycle
Baldauf and Dustdar's 2004 survey, later published with Florian Rosenberg in the International Journal of Ad Hoc and Ubiquitous Computing, distinguishes three ways of building context-aware systems: direct sensor access, in which drivers are built into the application; middleware that hides low-level sensing; and a context server that lets many clients share remote sensors and moves heavy processing off small mobile devices.[8][13] Their abstract architecture has five layers: sensors, raw data retrieval, preprocessing (interpretation, aggregation and resolution of conflicts between sensors), storage and management with synchronous polling or asynchronous subscription, and the application.[8]
Perera and colleagues describe a four-phase context life cycle: acquisition from physical or virtual sensors, modeling, reasoning to derive high-level context from low-level sensor data, and distribution to the consumers that need it. They compare six modeling techniques (key-value, markup schemes, graphical, object based, logic based and ontology based) and six families of reasoning techniques: supervised learning, unsupervised learning, rules, fuzzy logic, ontological reasoning and probabilistic reasoning.[3]
Context-aware augmented and virtual reality
Early location- and situation-aware AR
At Sony Computer Science Laboratory, Jun Rekimoto and Katashi Nagao presented NaviCam at UIST '95: a "situation-aware portable device" that recognized the user's situation by detecting color-code IDs in the environment and superimposed situation-sensitive messages on a video see-through screen.[14] The Touring Machine, built by Steven Feiner, Blair MacIntyre, Tobias Höllerer and Anthony Webster at Columbia University, combined a head-tracked, see-through, head-worn 3D display with a handheld 2D display and overlaid information about items of interest on the university campus. It tracked the user's position with differential GPS and orientation with a magnetometer and inclinometer; because that tracking was imprecise, the interface was designed around approximate rather than precise registration of virtual and real objects.[4]
Filtering and adapting AR interfaces
Information clutter became a problem for mobile AR displays. In 2000, Simon Julier, Marco Lanzagorta, Yohan Baillot, Lawrence J. Rosenblum, Steven Feiner, Tobias Höllerer and Sabrina Sestito described a region-based information filtering algorithm for mobile AR that takes account of the user's location and intent and the state of the objects that could be annotated, and uses simple temporal, distance and angle cues to smooth transitions between information sets.[15] Other systems adapted to the quality of their own sensing. Grubert and colleagues describe work by Hallaway et al. in which an indoor-outdoor AR system switched between a high-precision ceiling-mounted ultrasonic tracker, less accurate pedometer and infrared tracking, and GPS with inertial sensors outdoors; in coarse tracking mode it showed a world-aligned world-in-miniature, which is less affected by tracking error, instead of labels registered in the environment.[5]
Context can also change how users give input. Steven Henderson and Feiner's "Opportunistic Controls" (2008) mapped virtual buttons and other widgets onto unused physical features of the work environment, so that the surfaces provide passive haptic feedback; in a user study of a simulated aircraft engine inspection, participants finished tasks significantly faster with them than with a simpler passive-haptics baseline, and preferred them.[16] In 2019 David Lindlbauer, Anna Maria Feit and Otmar Hilliges presented an optimization-based system that decides which mixed reality applications are shown, how much information each displays and where it is placed, based on the user's current cognitive load and knowledge of the task and environment. It combines rule-based decisions with combinatorial optimization that runs in real time, and in a dual-task evaluation it reduced interactions with the secondary task by 36 percent.[17]
Pervasive augmented reality
Grubert, Langlotz, Zollmann and Regenbrecht extend Ronald Azuma's definition of AR and define Pervasive Augmented Reality as "a continuous and pervasive user interface that augments the physical world with digital information registered in 3D while being aware of and responsive to the user's context". They contrast it with typical AR applications, which serve a single purpose and are used for short periods, and name head-mounted hardware and car-integrated head-up displays as enablers of a continuous AR experience.[5]
Their taxonomy of context-aware AR has three parts:[5]
| Part | Question | Categories and examples |
|---|---|---|
| Context sources | What the system adapts to | Human factors: personal factors (physiological, perceptual, cognitive and affective states, preferences, activity) and social factors (social networks and places). Environmental factors: physical factors (including time and the spatial configuration of a scene), digital factors (the type, quality and quantity of digital information) and infrastructure factors (such as network reliability and bandwidth). System factors: system state (platform, computing power, battery, sensors) and the available output and input devices |
| Context targets | What is adapted | System input (for example switching from speech to gestures depending on ambient noise or whether the space is public or private), system output (which content is shown and how it is laid out or filtered) and system configuration (for example which trackers run, or offloading computation to a server) |
| Context controller | How it is adapted | Implicitly by the system (adaptivity) or explicitly through user input (adaptability) |
The survey found that research on context-aware AR had concentrated on adapting the system configuration (mainly tracking) and the system output (view management and scene composition), while adapting input had been studied by very few works. Output examples include museum guides that show less detail to children than to adults, and one that combined pulse readings, eye fixations and audio sensing to estimate a visitor's interest and steer the tour away from noisy, crowded locations. Adaptive configuration examples include Verbelen et al.'s mobile AR system, which offloads parts of tracking to a remote server depending on processor load, network quality and server availability.[5] The authors judged the context controllers in AR at the time "too simplistic to model or infer detailed contexts" and pointed to user modeling and machine learning as possible sources of richer models.[5]
Context in current XR devices
Current headsets model the user's surroundings in much more detail than the room-level location cells of the PARCTAB and Active Badge systems. On Meta's Meta Horizon OS, the scene model is a "geometric and semantic representation of the user's space": the system stores scene anchors with labels such as floor, ceiling, wall, table and couch, and applications that have been granted spatial data permission can query them, for example to attach a virtual screen to a wall or to handle physics and occlusion.[18] Microsoft's scene understanding runtime, supported on HoloLens 2, similarly gives developers "a structured, high-level environment representation designed to make developing for environmentally aware applications intuitive", built on top of the less structured spatial mapping mesh.[19]
Context also drives safety features in VR. Space Sense, which arrived with the Quest v34 software update in November 2021, shows outlines of people, pets and objects that enter the user's room-scale boundary, up to about 9 feet away and only in front of the headset.[20]
AI assistants on glasses use the wearer's visual context directly. In December 2024 Meta added "live AI" to Ray-Ban Meta glasses through its Early Access Program in the US and Canada; during a session Meta AI "can see what you see continuously", and Meta said the feature would eventually "give useful suggestions even before you ask".[21] Google's announcement of Android XR the same month said that advances in AI let XR devices "understand your intent and the world around you", and that headset users could have conversations with Gemini about what they are seeing.[22] Meta presents its Project Aria research glasses, which record first-person sensor data, as a platform for research in "egocentric and contextual AI, and robotics".[23]
Privacy and acceptance
Context-aware systems collect data about people, and the early papers addressed this directly. The Active Badge authors reported that the usual first reaction to a staff location system was "one of horror". At Olivetti Research, a laboratory of 32 staff, badges were compulsory for a two-week trial and optional afterward, and most people kept wearing them. The authors noted that the system tracks badges, not people, so anyone who does not want to be located can leave the badge on a desk, and they described this escape mechanism as possibly important for acceptance.[9] Schilit, Adams and Want expected businesses and government agencies to export contextual information and commands to reach potential customers, and warned that users would need to verify the security and authenticity of such third-party information.[1]
For AR, Grubert and colleagues wrote that a pervasive interface must be "socially and privacy aware", citing the experience of Google Glass. Depending on context, such a system might switch off its camera to protect people nearby, or avoid voice commands and hand gestures where the social situation does not permit them.[5] The same concerns apply to today's camera-equipped glasses and headsets; see Privacy in virtual and augmented reality.
See also
References
- ↑ 1.0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 Bill N. Schilit, Norman Adams, Roy Want (1994-12). "Context-Aware Computing Applications". 1994 First Workshop on Mobile Computing Systems and Applications (WMCSA), Santa Cruz, pp. 85-90. https://doi.org/10.1109/WMCSA.1994.16. Retrieved 2026-10-17.
- ↑ 2.0 2.1 2.2 2.3 2.4 2.5 2.6 2.7 Anind K. Dey (2001). "Understanding and Using Context". Personal and Ubiquitous Computing, vol. 5, no. 1, pp. 4-7. https://doi.org/10.1007/s007790170019. Retrieved 2026-10-17.
- ↑ 3.0 3.1 3.2 3.3 3.4 Charith Perera, Arkady Zaslavsky, Peter Christen, Dimitrios Georgakopoulos (2014). "Context Aware Computing for The Internet of Things: A Survey". IEEE Communications Surveys and Tutorials, vol. 16, no. 1, pp. 414-454. https://doi.org/10.1109/SURV.2013.042313.00197. Retrieved 2026-10-17.
- ↑ 4.0 4.1 Steven Feiner, Blair MacIntyre, Tobias Höllerer, Anthony Webster (1997-10). "A Touring Machine: Prototyping 3D Mobile Augmented Reality Systems for Exploring the Urban Environment". Digest of Papers, First International Symposium on Wearable Computers (ISWC '97), Cambridge, MA, pp. 74-81. https://doi.org/10.1109/ISWC.1997.629922. Retrieved 2026-10-17.
- ↑ 5.0 5.1 5.2 5.3 5.4 5.5 5.6 Jens Grubert, Tobias Langlotz, Stefanie Zollmann, Holger Regenbrecht (2017). "Towards Pervasive Augmented Reality: Context-Awareness in Augmented Reality". IEEE Transactions on Visualization and Computer Graphics, vol. 23, no. 6, pp. 1706-1724. https://doi.org/10.1109/TVCG.2016.2543720. Retrieved 2026-10-17.
- ↑ Bill N. Schilit, Marvin M. Theimer (1994-09). "Disseminating active map information to mobile hosts". IEEE Network, vol. 8, no. 5, pp. 22-32. https://doi.org/10.1109/65.313011. Retrieved 2026-10-17.
- ↑ 7.0 7.1 7.2 7.3 7.4 7.5 7.6 7.7 7.8 Anind K. Dey, Gregory D. Abowd, Daniel Salber (2001-12). "A Conceptual Framework and a Toolkit for Supporting the Rapid Prototyping of Context-Aware Applications". Human-Computer Interaction, vol. 16, no. 2-4, pp. 97-166. https://doi.org/10.1207/S15327051HCI16234_02. Retrieved 2026-10-17.
- ↑ 8.0 8.1 8.2 Matthias Baldauf, Schahram Dustdar (2004-11-30). "A Survey on Context-aware systems". Technical Report TUV-1841-2004-24, Distributed Systems Group, Vienna University of Technology. http://www.infosys.tuwien.ac.at/Staff/sd/papers/TUV-1841-2004-24.pdf. Retrieved 2026-10-17.
- ↑ 9.0 9.1 Roy Want, Andy Hopper, Veronica Falcão, Jonathan Gibbons (1992). "The Active Badge Location System". ACM Transactions on Information Systems, vol. 10, no. 1, pp. 91-102 (Olivetti Research version hosted by the University of Cambridge). doi:10.1145/128756.128759. https://www.cl.cam.ac.uk/research/dtg/attarchive/pub/docs/att/tr.92.1.pdf. Retrieved 2026-10-17.
- ↑ Gregory D. Abowd, Christopher G. Atkeson, Jason Hong, Sue Long, Rob Kooper, Mike Pinkerton (1997-10). "Cyberguide: A mobile context-aware tour guide". Wireless Networks, vol. 3, no. 5, pp. 421-433. https://doi.org/10.1023/A:1019194325861. Retrieved 2026-10-17.
- ↑ Thad Starner, Bernt Schiele, Alex Pentland (1998). "Visual Contextual Awareness in Wearable Computing". Digest of Papers, Second International Symposium on Wearable Computers (ISWC '98), pp. 50-57. doi:10.1109/ISWC.1998.729529. https://faculty.cc.gatech.edu/~thad/p/031_20_Activity/visual-context-awareness_ISWC98.pdf. Retrieved 2026-10-17.
- ↑ Daniel Salber, Anind K. Dey, Gregory D. Abowd (1999). "The context toolkit: aiding the development of context-enabled applications". Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI '99), pp. 434-441. https://doi.org/10.1145/302979.303126. Retrieved 2026-10-17.
- ↑ Matthias Baldauf, Schahram Dustdar, Florian Rosenberg (2007). "A survey on context-aware systems". International Journal of Ad Hoc and Ubiquitous Computing, vol. 2, no. 4, pp. 263-277. https://doi.org/10.1504/IJAHUC.2007.014070. Retrieved 2026-10-17.
- ↑ Jun Rekimoto, Katashi Nagao (1995). "The World through the Computer: Computer Augmented Interaction with Real World Environments". UIST '95 (ACM Symposium on User Interface Software and Technology), pp. 29-36. doi:10.1145/215585.215639. https://sites.cc.gatech.edu/gvu/people/Masters/Ian/UIST95/abstracts/Rekimoto.html. Retrieved 2026-10-17.
- ↑ Simon Julier, Marco Lanzagorta, Yohan Baillot, Lawrence Rosenblum, Steven Feiner, Tobias Höllerer, Sabrina Sestito (2000). "Information filtering for mobile augmented reality". Proceedings IEEE and ACM International Symposium on Augmented Reality (ISAR 2000), pp. 3-11. https://doi.org/10.1109/ISAR.2000.880917. Retrieved 2026-10-17.
- ↑ Steven J. Henderson, Steven Feiner (2008-10). "Opportunistic controls". Proceedings of the 2008 ACM Symposium on Virtual Reality Software and Technology (VRST '08), pp. 211-218. https://doi.org/10.1145/1450579.1450625. Retrieved 2026-10-17.
- ↑ David Lindlbauer, Anna Maria Feit, Otmar Hilliges (2019-10). "Context-Aware Online Adaptation of Mixed Reality Interfaces". Proceedings of the 32nd Annual ACM Symposium on User Interface Software and Technology (UIST '19), pp. 147-160. https://doi.org/10.1145/3332165.3347945. Retrieved 2026-10-17.
- ↑ "Unity Scene Overview". Meta Horizon OS developer documentation. Meta Platforms. https://developers.meta.com/horizon/documentation/unity/unity-scene-overview/. Retrieved 2026-10-17.
- ↑ "Scene understanding". Microsoft Learn. Microsoft. https://learn.microsoft.com/en-us/windows/mixed-reality/design/scene-understanding. Retrieved 2026-10-17.
- ↑ Scott Hayden (2021-11-02). "Quest v34 Update Brings 'Space Sense' Guardian, Android Notifications & MR App Publishing". Road to VR. https://roadtovr.com/oculus-quest-2-space-sense-guardian/. Retrieved 2026-10-17.
- ↑ "Ray-Ban Meta Glasses Add Live AI, Live Translation, & Shazam Support". Meta. Meta Platforms. 2024-12-16. https://www.meta.com/blog/ray-ban-meta-v11-software-update-live-ai-translation-shazam/. Retrieved 2026-10-17.
- ↑ Shahram Izadi (2024-12-12). "Android XR: The Gemini era comes to headsets and glasses". The Keyword (Google blog). Google. https://blog.google/products/android/android-xr/. Retrieved 2026-10-17.
- ↑ "Introducing Aria Gen 2: Unlocking New Research in Machine Perception, Contextual AI, Robotics, and More". Meta. Meta Platforms. 2025-02-27. https://www.meta.com/blog/project-aria-gen-2-next-generation-egocentric-research-glasses-reality-labs-ai-robotics/. Retrieved 2026-10-17.