Smart glasses

Smart glasses are electronically equipped eyewear that provides digital functions while being worn. Depending on the design, they incorporate cameras, microphones, speakers, motion sensors, wireless connectivity, and near-eye displays. Some operate principally as hands-free accessories to a smartphone, while others contain substantial onboard computing. A display is not essential: camera-and-audio glasses and glasses that present visual information belong to the broader category, although their capabilities differ considerably. (developers.meta.com)

Display-equipped systems may present a small information panel or overlay digital imagery on the visible environment. More advanced augmented-reality systems combine near-eye optics with environmental sensing and tracking to position virtual content relative to physical surroundings. Other smart glasses emphasize photography, communication, or voice-based artificial intelligence rather than visual augmentation. These approaches connect wearable computing with optical engineering, computer vision, and human–computer interaction. (learn.microsoft.com)

Their significance lies in bringing digital information and communication to an eye-level, potentially hands-free interface. Documented uses include first-person recording, remote technical assistance, access to work instructions, and contextual interaction with software services. Development involves trade-offs among optical performance, weight, power consumption, and interaction capabilities, alongside questions about data privacy and the recording of nearby people. (realwear.com)

Definition, scope, and device categories

Smart glasses combine eyewear, or electronic hardware mounted on eyewear, with functions for acquiring, processing, communicating, or presenting information. The defining feature is this information interface, not corrective lenses or a particular display technology. Prescription glasses and sunglasses can therefore serve as the physical platform for smart glasses; ordinary corrective eyewear does not qualify merely because it changes how the wearer sees. For this classification, electrically adjustable focus or tint alone is also insufficient: changing a lens’s optical properties is distinct from recording a scene, receiving a message, or presenting computer-generated information. Neither a camera nor a visual display is universal. Amazon’s original Echo Frames, announced in 2019, provided voice-assistant access, calls, audio playback, and spoken notifications without either component; Meta’s developer documentation likewise distinguishes display-equipped from displayless glasses. (press.aboutamazon.com)

Camera-and-audio glasses acquire images and sound while presenting information principally through speakers. Displayless Ray-Ban Meta devices exemplify this category: their cameras and microphones supply inputs to recording or connected applications, but the wearer does not see a digital image through the lenses. Audio-only eyewear forms an adjacent category, retaining microphones and speakers without requiring image capture. Notification-display glasses, by contrast, place short messages, prompts, or other limited visual information near the eye. Vuzix introduced the Z100 in January 2024 with a transparent, monochrome, monocular display for information supplied by an Android or iOS device. These categories describe input and output capabilities rather than mutually exclusive product families: a notification display can coexist with cameras and audio, and the presence of a screen does not by itself establish spatial augmented reality. (developers.meta.com)

Assisted-reality viewers present task information in an accessible panel without necessarily registering it to physical objects. The distinction is between reading a maintenance checklist near the eye and seeing an annotation remain attached to a particular machine component as the wearer moves. Industrial systems may use an adjustable display arm rather than an ordinary spectacle frame. RealWear’s Navigator 500, for example, uses a single opaque, non-see-through display; camera imagery can appear within that panel without making the display itself transparent. Spatial augmented reality glasses instead estimate the wearer’s pose and the surrounding environment so that virtual content can occupy positions relative to the physical world. HoloLens 2 illustrates this capability through six-degrees-of-freedom tracking and real-time spatial mapping, although its larger headset construction places it among related systems rather than ordinary eyewear. (realwear.com)

Display arrangement and computing dependence provide separate classification axes. Monocular presentation supplies an image to one eye; binocular presentation supplies imagery to both, while stereoscopic presentation additionally uses different views to convey depth. These distinctions do not determine whether content is spatially anchored. Likewise, glasses may depend on a smartphone, act as a wired external display, or contain the computing resources needed to run applications themselves. The Z100 requires pairing with an Android or iOS device through its companion application. XREAL Air 2 connects to compatible external video sources through USB-C, whereas HoloLens 2 is an untethered computer with its own processor, storage, and operating system. “Standalone” therefore describes onboard operation, not an absence of networking or optional external services. (ir.vuzix.com)

A head-mounted display is a broader category defined by the placement of visual output on the head: display-equipped smart glasses belong within it, but displayless glasses do not. Immersive virtual-reality headsets ordinarily block the direct view of the surroundings and substitute a rendered environment. Optical see-through systems preserve a direct view while adding imagery; camera-mediated systems instead reproduce the surroundings on otherwise opaque displays. These viewing modes should not be confused with physical form factor. HoloLens 2, specified at 566 grams and fitted with an adjustable headband and visor, demonstrates why a spatial mixed-reality headset can share functions with smart glasses without being equivalent to spectacle-style eyewear. Neither a head-worn mounting nor a mixed-reality label alone makes a device “glasses.” (learn.microsoft.com)

Historical development and representative systems

In 1968, Ivan Sutherland published “A Head-mounted Three Dimensional Display,” describing an experimental system that presented separate computer-generated views to the two eyes. Miniature cathode-ray tubes and partially reflecting optical elements allowed line drawings to appear alongside the physical surroundings. Head-position measurements controlled the perspective calculation, so a virtual object could appear stationary while the observer moved. The project investigated both mechanical and ultrasonic tracking; its mechanical apparatus connected the head-mounted equipment to an overhead support. This was a laboratory installation dependent on external computing and tracking hardware, not a self-contained spectacle frame. Its contribution to later augmented reality was the combination of see-through imagery, stereoscopic presentation, and viewpoint-dependent rendering. (web.stanford.edu)

Google Glass represented a different approach: a compact information display rather than a complete binocular spatial view. The 2013 Explorer Edition placed its display above the wearer’s right eye, making information available without occupying the entire visual field. The Explorer program subsequently informed an enterprise-oriented redesign developed with industrial partners, including Boeing, DHL, and AGCO. Google announced Enterprise Edition 2 on May 20, 2019, emphasizing access to checklists, instructions, inspection images, and remote collaboration. Its Android-based software and support for enterprise device management facilitated integration into organizational systems. The transition was therefore not simply a smaller version of Sutherland’s apparatus, but a shift toward task-specific information delivery and managed workplace deployment. Sales of Glass Enterprise Edition ended on March 15, 2023, and Google’s support ended on September 15, 2023; existing devices and software could continue operating, but no further Google software updates were planned. (x.company)

Industrial assisted-reality devices and spatial mixed-reality headsets developed along distinct paths. RealWear’s HMT-1 used a boom-mounted monocular display, a camera, and voice-controlled navigation to provide access to documents and applications without requiring a handheld screen. Its manual documented spoken application names and commands as the principal navigation mechanism. Microsoft’s HoloLens instead provided an untethered, see-through platform for placing digital content within a three-dimensional environment. The HoloLens Development Edition began shipping to developers and enterprise partners on March 30, 2016; HoloLens 2 followed in 2019 with articulated hand tracking and eye tracking. Whereas the HMT-1 served as a head-worn interface to task information, HoloLens supported spatially positioned imagery and interaction with virtual objects. The latter was a larger headset rather than ordinary spectacle-shaped eyewear. (support.realwear.com)

Camera-and-audio eyewear established another commercially supplied category. Meta and EssilorLuxottica announced Ray-Ban Meta glasses on September 27, 2023, with retail availability beginning October 17 and a starting US price of $299. The launch model combined a 12-megapixel camera, five microphones, open-ear speakers, and voice access to Meta AI, but did not include an image display. In April 2024, Meta began rolling out a beta visual-assistance feature in the United States and Canada, allowing spoken questions to be combined with camera input—for example, to translate text on a menu. Here, artificial intelligence extended photography and audio interaction rather than supplying a visible spatial overlay. (about.fb.com)

Meta’s Orion, unveiled on September 25, 2024, demonstrated a separate display-equipped development path. Its prototype incorporated silicon-carbide lenses, optical waveguides, and micro-LED projectors to present digital content within the wearer’s view of the physical world. Demonstrations included messaging, video calls, and an assistant responding to visible objects. At unveiling, Orion was designated a product prototype, with access restricted to Meta employees and selected external participants—not a retail offering. Its status differed both from commercially sold Ray-Ban Meta glasses and from a purchasable developer platform such as the original HoloLens Development Edition. (about.fb.com)

Hardware architecture, connectivity, and power

The hardware chain begins with sensor acquisition and passes through processing, working memory and persistent storage to output components, with a battery and power-management circuitry supplying the system. A system-on-chip can combine general-purpose computing with specialized image processing, audio processing and neural-network accelerators. Qualcomm’s Snapdragon AR1 Gen 1 documentation, for example, describes dual image-signal processors, a neural processing unit, a low-power sensing hub, and interfaces for microphones, stereo speakers and optional displays. These are platform capabilities rather than a sensor package fitted to every product. In a larger standalone system, Microsoft’s HoloLens 2 hardware documentation, updated on March 13, 2023, lists a Snapdragon 850 processor, a separate custom holographic processing unit, 4 GB of working memory and 64 GB of storage. Working memory supports active computation, whereas persistent storage retains applications and recorded media. (docs.qualcomm.com)

Component placement depends on whether the glasses perform computation locally or share it with another device. Cameras need an unobstructed view, while electronics, speakers and batteries must fit within the limited frame volume. Meta’s February 2023 account of Ray-Ban Stories describes integrating cameras, microphones, audio drivers and touch controls into an eyewear-sized enclosure. Qualcomm’s Snapdragon AR2 Gen 1 architecture, announced on November 16, 2022, instead distributes an AR processor, a sensor-aggregating coprocessor and connectivity hardware across the glasses to balance weight and reduce temple width. It retains latency-sensitive perception processing on the eyewear while offloading more complex workloads to a compatible smartphone or computer. Thus, an onboard processor does not necessarily make a device computationally independent. (about.fb.com)

Sensors are selected for particular functions. Camera-and-audio glasses require image capture and acoustic input but need not include depth sensing or eye tracking: the Ray-Ban Meta model announced on September 27, 2023, used a 12-megapixel camera and five-microphone array. Multiple microphone channels provide inputs for directional capture and noise processing. Spatial systems may add tracking cameras and an inertial measurement unit, combining accelerometers and gyroscopes to measure motion. HoloLens 2 illustrates a more extensive configuration: four visible-light head-tracking cameras, two infrared eye-tracking cameras, a 1-megapixel time-of-flight depth sensor, and an inertial unit containing an accelerometer, gyroscope and magnetometer. Its separate photographic camera records 8-megapixel stills and 1080p video at 30 frames per second. Depth and eye-position sensing support spatial interaction, but are not prerequisites for ordinary recording, calls or audio playback. (about.fb.com)

Connections serve distinct purposes rather than acting as interchangeable links. Ray-Ban Meta glasses use Bluetooth for pairing and connected audio functions; their companion application manages device settings and media imports. For transfers such as importing recordings or downloading firmware, the glasses temporarily use an internal Wi-Fi network. A wired connection can carry both output signals and electrical power: XREAL’s One-series guide specifies a USB-C host supporting DisplayPort video output, with the glasses drawing power from the connected phone, tablet or laptop. Consequently, a USB-C socket alone does not establish compatibility; the host must supply the required video protocol as well as power. Wireless distributed computing, exemplified by Snapdragon AR2, adds another role for Wi-Fi: exchanging processing data with an external host. (ray-ban.com)

Battery capacity and endurance are separate quantities: usable stored energy is consumed at a workload-dependent rate, so continuous capture, audio processing and radio transmission cannot be equated with intermittent use. Ray-Ban’s specifications distinguish the 2023 Ray-Ban Meta model’s maximum four hours of moderate use from three hours of continuous audio streaming and voice assistance. Meta’s launch announcement specified eight additional case charges, giving up to 36 hours including recharging intervals—not uninterrupted wear. The glasses recharge through contacts in the case, whose own battery is charged through USB-C. Microsoft’s February 2019 HoloLens 2 fact sheet specified two to three hours of active use and passive cooling. Its later hardware documentation specifies an 18-watt supplied charger and at least 15 watts to maintain or increase charge during operation. Fanless cooling makes dissipation of processor heat through the enclosure a design constraint; workload allocation therefore affects both electrical endurance and thermal demands on the head-worn hardware. (ray-ban.com)

Near-eye displays and optical systems

An optical see-through display transmits light from the surroundings directly to the eye while an optical combiner introduces generated imagery into the same view. The wearer therefore sees the physical scene through the optics, rather than as a camera recording. In camera-mediated video passthrough, outward-facing cameras capture the surroundings, and displays present those images together with computer-generated content. This distinction concerns the visual path: an optical see-through device may still contain cameras for recording or tracking. Microsoft’s HoloLens 2, for example, combines see-through waveguide lenses with separate imaging sensors. (global.canon)

The image engine and combiner perform different functions. Microdisplays generate or modulate an image: organic light-emitting diode and micro-LED arrays emit light, whereas liquid-crystal-on-silicon and digital micromirror devices modulate external illumination. Relay lenses then form a virtual image at a designed optical distance. Combiners can use partially reflecting mirrors, curved reflectors, or prisms combining reflection and refraction; the folded “birdbath” arrangement is one example. A waveguide instead transports image light through a thin transparent plate, allowing much of the projection hardware to sit beside the eye rather than directly in front of it. (lcd.creol.ucf.edu)

In many waveguides, an input coupler redirects projected light into angles that support total internal reflection at the plate’s boundaries. An output coupler extracts light toward the eye. Geometric waveguides employ partially reflective surfaces; diffractive designs use diffraction gratings, including surface-relief or volume gratings. Pupil expansion replicates the projector’s exit pupil across one or two directions, enlarging the region from which the image is visible without enlarging the original projector aperture. Repeated extraction distributes light over that region, making brightness uniformity and transmission losses important design constraints. These mechanisms are not universal: a 2024 experimental geometric lightguide used partially reflective mirrors for pupil expansion but propagated light without total internal reflection. (nature.com)

Field of view is the angular extent of the displayed image; horizontal, vertical, and diagonal values describe different dimensions and are not interchangeable. Eye relief is the working distance between the eye and the final optical surface. The eyebox is the region within which the pupil can move while retaining the specified image; outside it, clipping or disappearance can occur. These quantities jointly determine permissible eye position. The 2024 lightguide experiment reported a $40^\circ$ diagonal field of view, a $13 \times 6.5\ \mathrm{mm}$ eyebox, and $18\ \mathrm{mm}$ eye relief. Angular sampling, commonly expressed in pixels per degree, describes how densely an image is sampled across its visual angle rather than merely counting panel pixels. Microsoft specifies HoloLens 2’s image density as more than 2,500 light points per radian. (dispelix.com)

Luminance, measured in $\mathrm{cd/m^2}$, characterizes image brightness in a viewing direction; transparency describes how much environmental light passes through. Contrast depends on both the generated image and the transmitted background. Ordinary additive see-through displays cannot create opaque black simply by switching pixels off: black becomes transparent, and bright backgrounds reduce the visibility of overlays. Tinting can suppress background light but also darkens the physical scene. Optical efficiency describes how much source light reaches a specified output or receiving aperture, so comparisons require consistent measurement boundaries. A bright microdisplay does not necessarily produce equally bright imagery at the eye after coupling, propagation, and pupil-expansion losses. (learn.microsoft.com)

Monocular presentation supplies an image to one eye; binocular presentation becomes stereoscopic when different views provide binocular disparity. Stereoscopic depth does not necessarily change the virtual image’s focal distance. Microsoft documents HoloLens imagery at an optical distance of approximately $2.0\ \mathrm{m}$, even when rendered objects appear nearer or farther away. The eyes then converge toward the depicted depth while accommodating to the display’s fixed focal distance—the vergence–accommodation conflict. A fixed information panel need not provide binocular disparity or world registration. Conversely, spatial imagery requires viewpoint-dependent rendering and registration, discussed separately, and a fixed-focus stereoscopic display still does not reproduce the changing focus cues of physical objects. (learn.microsoft.com)

Tracking, environmental perception, and spatial registration

Spatial tracking estimates the wearer’s head pose relative to the surroundings. Orientation-only tracking measures three rotational degrees of freedom—pitch, yaw, and roll—but does not determine translation. Position-and-orientation tracking adds movement along three spatial axes, producing six-degree-of-freedom pose estimates. This distinction matters when the wearer leans sideways or approaches an object: a world-registered overlay must change perspective with translation, not merely with head rotation. Visual–inertial systems combine observations from outward-facing cameras with an inertial measurement unit (IMU), whose gyroscope measures angular velocity and whose accelerometer measures specific force. Inertial observations provide rapid motion information, while visual observations constrain movement relative to environmental features. (developers.google.com)

In simultaneous localization and mapping (SLAM), computer vision algorithms track recognizable image features across successive frames while estimating both device motion and a map of their three-dimensional locations. Combining these observations with inertial measurements helps constrain errors that would accumulate from inertial integration alone. Microsoft’s HoloLens 2 provides a documented example of this architecture, although it is a larger mixed-reality headset rather than an ordinary eyewear frame. Microsoft’s 2020 Research Mode description identifies four grayscale tracking cameras: two form a forward-facing stereo pair that obtains feature depth by triangulation, while the others extend the area over which features remain visible. These synchronized global-shutter cameras capture images at up to 30 frames per second; the device also exposes accelerometer, gyroscope, and magnetometer measurements for research. (microsoft.com)

A localization map and a surface reconstruction serve different purposes. Sparse feature locations can support pose estimation, whereas spatial mapping represents environmental surfaces for object placement, collision testing, and occlusion. Depth measurements, including those from a dedicated depth camera, support reconstruction of surrounding geometry. In Microsoft’s spatial-mapping interface, each surface is represented by a triangle mesh attached to a world-locked coordinate system. An application can intersect a ray with this geometry to position an object on a wall or table, or suppress virtual pixels that should lie behind a real surface. Such geometry is not necessarily a semantic interpretation: a reconstructed flat region need not have been identified as a “table.” Meshes also require updating; an unobserved area can retain obsolete geometry after a person or object moves away. (microsoft.com)

Accurate registration additionally requires calibration between the tracking coordinate system and the virtual scene as viewed through the display. The wearer’s eyes are distinct viewpoints from the tracking cameras, so estimating head pose alone does not establish the correct projection. HoloLens 2 uses eye-position measurements internally to adjust rendering, including when the visor shifts slightly during use; this function is distinct from reporting gaze direction for interface input. In a 2017 study, Ehsan Azimi, Long Qian, Nassir Navab, and Peter Kazanzides investigated calibration by aligning virtual points with physical counterparts. Their experiments with HoloLens reported average reprojection error of approximately 4 mm under the evaluated conditions, illustrating a measured alignment result rather than a universal tracking specification. (learn.microsoft.com)

Registration error can appear as a persistent offset, slow drift, jitter, or sudden jumps when the system revises its environmental estimate. Latency creates another error source: the displayed image may correspond to an earlier head pose. Pose prediction and late-stage reprojection reduce this discrepancy, but cannot completely correct every change in perspective. Microsoft’s HoloLens rendering documentation specifies 60 frames per second as a stability target and notes that reducing latency reduces prediction error. Environmental ambiguity remains consequential: darkness provides insufficient visual information, excessive brightness can saturate cameras, blank walls provide few features, and repetitive workstations can be mistaken for one another. Reflective surfaces and moving crowds further undermine stable correspondences, causing overlays to wander, jump, or lose their intended alignment with physical objects. (learn.microsoft.com)

Input methods and interface design

Physical controls provide discrete commands without requiring speech or a visual cursor. Ray-Ban Meta camera-and-audio glasses, for example, combine a capture button with a touchpad on the right temple: a single tap pauses or resumes audio, double and triple taps move between tracks, and forward or backward swipes adjust volume. Calls use context-dependent mappings, including double-tap to answer and press-and-hold to reject. Voice commands provide another route to these functions, such as “Hey Meta, take a photo.” These are documented controls for particular devices, not a universal input vocabulary for smart glasses. (ray-ban.com) Industrial interfaces can instead expose command words directly on screen. RealWear’s Ari OS uses a “say what you see” approach: labelled controls respond to their displayed names, while unnamed items receive numbers selected by saying “SELECT ITEM” followed by the number. “SHOW HELP” reveals available commands, addressing discoverability without requiring the wearer to memorize every screen’s vocabulary. (support.realwear.com)

Head-directed selection and eye gaze are distinct targeting mechanisms. Head gaze projects a ray along the head’s forward direction and identifies the object intersected by that ray; it approximates attention rather than measuring the eyes’ direction. Eye tracking provides a separate gaze vector, supported on HoloLens 2 but not the first-generation HoloLens. Both can be paired with an explicit confirmation, such as an air tap, the spoken command “select,” or a button press on the separate HoloLens Clicker peripheral. A cursor or subtle target highlight communicates what will receive the command. (learn.microsoft.com) Looking alone does not establish an intention to activate something: people also inspect labels, read text, and scan their surroundings. Microsoft’s eye-gaze guidance therefore distinguishes attention from deliberate selection and combines gaze with hand or voice input to confirm intent. (learn.microsoft.com)

Dwell selection substitutes sustained looking for a separate confirmation action. This permits interaction when hands are occupied, but creates a timing trade-off: short thresholds can trigger unwanted selections, whereas long thresholds slow navigation. Microsoft’s eye-gaze-and-dwell guidance specifies an initial delay of 150–250 milliseconds before displaying activation feedback, followed by another 650–850 milliseconds of fixation before selection. A progressively contracting indicator makes the pending action visible. These figures are interface-design recommendations, not measured universal response times or requirements shared by all eyewear. (learn.microsoft.com) Hand gestures offer more explicit control where supported. Microsoft’s August 2022 HoloLens 2 documentation describes a ray extending from the palm toward distant content, with thumb–index air taps confirming selection and changing the ray and cursor’s appearance. This articulated-hand interface differs from the first-generation HoloLens gesture model. (learn.microsoft.com)

Output channels determine how commands and system states become perceptible. Ray-Ban Meta glasses can announce callers and read messages aloud; their companion smartphone application provides switches for announcements and automatic read-out. (ray-ban.com) Display-equipped systems can instead show menus, target highlights, and captions. Meta Ray-Ban Display, for example, presents live speech transcriptions on its display; captions can be started through its menu or a voice command and stopped through voice, touchpad input, or a gesture using the separate Meta Neural Band. Accessory-mediated gestures therefore should not be conflated with camera-based hand tracking. (ray-ban.com) In spatial interfaces, spatial audio can associate feedback with a virtual object’s location. Microsoft documents distinct sounds for button contact, completed actions, voice-command success or failure, and notifications, partly compensating for the absence of tactile feedback when manipulating virtual controls. (learn.microsoft.com)

Hands-busy operation still requires attention to targets, command states, and feedback. Microsoft’s head-gaze-and-dwell documentation uses engine repair as an example in which tools occupy the hands and workshop noise makes voice commands difficult; it also identifies repeated upward and downward head movements caused by poorly placed controls as an undesirable interaction pattern. (learn.microsoft.com) RealWear combines speech recognition with microphone noise cancellation, but also provides a microphone-mute command; once muted, listening must be restored with the physical Action button rather than another spoken command. (support.realwear.com) Interruption management likewise involves explicit interface choices: message read-outs can be disabled, while Microsoft’s sound guidance calls for repeated notification tones to remain unobtrusive rather than harsh. (ray-ban.com)

Software platforms and AI-enabled functions

Smart-glasses software can run directly on the eyewear or extend an application running on a connected smartphone. Industrial viewers illustrate the first arrangement: RealWear documents an Android 13–based operating system with its WearHF hands-free interface, while its Auto Voice Engine derives voice commands from application interface elements. Developers can therefore adapt Android applications rather than build every interaction from device-specific primitives. Consumer eyewear can instead expose its hardware to mobile software: Meta’s Wearables Device Access Toolkit connects iOS and Android applications to compatible glasses. Its documented modules cover camera capture, speech, displays, input, and motion, although available functions vary by glasses model; a display API does not give displayless glasses a visual output channel. (support.realwear.com)

Hardware access is mediated by application programming interfaces and permission flows, not simply by the presence of a sensor. Meta’s Android toolkit separates application registration, session management, camera streaming, and device permissions; its sample application explicitly requests permission to access the glasses’ camera. The phone’s operating system imposes additional controls: Android 12 and later provide system-wide camera and microphone toggles on supported devices. Raw sensor access can also be restricted to research interfaces. In the related HoloLens platform, Research Mode exposes tracking-camera, depth, and inertial streams after explicit activation through the Device Portal, but Microsoft distinguishes this facility from interfaces intended for deployed enterprise or store applications. Such distinctions separate ordinary application capabilities from experimental access to underlying hardware. (github.com)

Processing may be distributed across the glasses, a phone, and cloud services. Ray-Ban’s documented visual-question workflow sends a photograph to Meta’s cloud, returns an audio answer through the glasses, and makes the request, image, and response available in the companion application. The phone also supplies service connections, settings, and media-management functions. Meta combined the earlier Meta View companion application with its Meta AI application in April 2025. Cloud-dependent interpretation adds network transfer and remote processing to the response path; consequently, local capture and access to an online assistant are distinct capabilities. Local neural-network inference provides another arrangement: at AWE USA on June 10, 2025, Qualcomm demonstrated an assistant running a one-billion-parameter Llama model on RayNeo X3 Pro glasses without a phone or cloud connection. This was a demonstrated implementation, not evidence that all functions of commercially supplied glasses operate offline. (ray-ban.com)

AI-assisted visual questions combine spoken instructions with camera-derived context. Functionally, the pipeline involves speech recognition, interpretation of the request through language processing, and visual interpretation, followed by answer generation and audio or display output. These functions need not correspond to separate models: a multimodal model can combine image and text representations. For example, Meta’s Llama 3.2 Vision architecture uses a separately trained vision adapter integrated with a pretrained language model; this architectural example does not establish which model a particular glasses service uses. On April 23, 2024, Meta began rolling out visual-question functionality in beta to Ray-Ban Meta users in the United States and Canada, illustrating it with camera-based translation of a French menu. The captured image supplies context for the question rather than constituting continuous, comprehensive knowledge of the wearer’s surroundings. (developers.meta.com)

Contextual assistance also depends on integration with existing services. Documented Ray-Ban Meta functions include camera-view sharing through WhatsApp and Messenger, phone-based Apple Music playback, and Shazam music identification; these connect eyewear input and output to established applications rather than replace them. Their availability depends on service configuration, supported languages, and markets. Generated descriptions and translations remain different from verified observations: language-model documentation acknowledges the possibility of inaccurate responses, including hallucinations. Evaluation therefore distinguishes successful capture, correct recognition of visible text or objects, factual correctness of the generated answer, and successful execution of any requested service action; fluent speech alone does not establish that the interpretation is accurate. (about.fb.com)

Consumer, industrial, and accessibility applications

Consumer camera-and-audio glasses support first-person photography and video recording, telephone calls, voice messages, and music playback without requiring a near-eye display. Ray-Ban Meta illustrates this category: its camera can capture the wearer’s perspective or share it during a video call, while speakers deliver calls, media, and spoken assistance. Meta’s September 2024 updates introduced voice-set reminders and functions for remembering information such as a parking location. Contextual assistance adds another use: a wearer can ask a question about the scene captured by the camera and receive an audio response. These functions do not constitute a visual overlay; reading captions or viewing instructions inside the glasses requires display-equipped hardware. (about.fb.com)

Industrial remote assistance connects an on-site worker’s camera and microphone to a specialist who can inspect equipment remotely and return spoken guidance, documents, or annotated images. On June 11, 2019, BMW of North America announced TSARAVision, combining RealWear HMT-1 viewers with Ubimax software, for implementation at 347 BMW Centers and selected MINI dealers that month. Engineers could send technical bulletins and schematic drawings to the technician’s display, enlarge images, and capture screenshots; technicians could open documents by voice while working on a vehicle. The announcement documents the intended deployment and workflow, rather than independently establishing a reduction in repair time attributable to the glasses alone. (press.bmwgroup.com)

Workflow applications instead guide workers through ordered tasks. RealWear describes displaying maintenance procedures, digital checklists, and step-by-step instructions, while its document viewer provides voice-controlled access to PDF manuals and images. Inspection software can associate checklist responses with photographs, timestamps, and pass/fail decisions: Applivity’s Inspection AI, for example, describes on-device defect flagging followed by human review and an inspection record for audit. In logistics, “vision picking” presents warehouse picking guidance while leaving workers’ hands available for handling goods. DHL reported a 25% efficiency increase in a Netherlands pilot conducted with Ricoh and Ubimax. That figure is the operator’s reported result for a particular pilot, not a general performance estimate for smart glasses or an independently replicated finding. (realwear.com)

Applications in education include displaying guidance beside a practical task and connecting learners with instructors. Workplace training can use the same remote-support mechanism as maintenance: a novice performs a procedure while an expert observes the transmitted view and provides feedback. A 2022 study by Shih-Yeh Chen, Pei-Hsuan Lin, and Wei-Che Chien investigated drawing and colour instruction with fourth-grade pupils in Taitung, Taiwan, using experimental and control groups of 30 pupils each. The system analysed outlines with OpenCV, generated colour suggestions with pix2pix, and returned recommended colours and pigment-mixing ratios to the glasses. Suggestions appeared when a learner pressed a controller button, rather than continuously. The authors reported higher assessed painting performance in the experimental group; the intervention combined glasses, software, and instruction, so it did not isolate the contribution of eyewear alone. (realwear.com)

Accessibility applications use different output channels for different tasks. XRAI’s November 2022 launch paired an Android application with phone-tethered Nreal glasses to present live conversational captions, with transcription and translation functions in nine languages. This display-based approach was designed for deaf people and people with hearing loss. Visual assistance can instead operate through audio: Ray-Ban Meta offers spoken descriptions of camera-observed surroundings, while its Be My Eyes integration connects blind or low-vision wearers to sighted volunteers. Meta documented expanded description settings in May 2025 and described the volunteer service as available in November 2025. Human volunteer assistance and automatically generated descriptions are distinct services, even though both use the glasses’ camera as their visual input. (xrai.glass)

Surgical visualization belongs to a separate medical-device context. The US Food and Drug Administration cleared Augmedics’ xvision Spine system through the 510(k) pathway on December 20, 2019. Its documentation describes a near-eye see-through display presenting spinal images, three-dimensional models, instrument positions, and trajectory guidance in relation to the patient’s anatomy. This is a specialized navigation system with tracking and image-registration requirements, not an ordinary consumer visual-assistance application. The clearance and documented display functions do not, by themselves, establish improved patient outcomes. (fda.gov)

Ergonomics, performance assessment, and deployment

Fit involves both the load carried by the wearer and the positioning of electronic components. In spectacle-style devices, the bridge or nose pads and the temples support the frame; prescription compatibility therefore concerns lens availability as well as frame adjustment. Ray-Ban’s Meta Scriber Optics documentation, for example, specifies flexible nose pads, adjustable temple tips, and hinges permitting 10 degrees of additional opening. Weight distribution is a separate variable from total mass: a 2021 study of modified Vuzix glasses compared front-, side-, and back-weighted configurations during warehouse picking and putting tasks. The side-weighted arrangement received the best comfort ratings, but participants still reported discomfort, and the configurations produced no significant productivity differences. These results distinguish tolerable task performance from comfortable prolonged wear. (ray-ban.com)

Industrial assisted-reality viewers use different fitting arrangements from ordinary eyewear frames. RealWear Navigator devices can be supported by a workband or attached to compatible protective equipment, with a movable boom positioning the monocular display independently of prescription glasses. The manufacturer’s fitting procedure places the display slightly below the dominant eye, approximately 1.5–2 inches in front of it, then rotates and tilts the pod until all four screen corners are visible. Its Navigator 500 user guide explicitly permits continued use of eyeglasses or eye protection. Thus, physical compatibility with a helmet or safety glasses does not by itself establish correct display alignment; mounting and visual adjustment are separate operations. (support.realwear.com)

Performance assessment separates component specifications from the information successfully obtained during a task. Display resolution is not a direct measure of instruction readability, just as camera pixel count does not establish whether a remote expert can identify a component. RealWear’s specifications separately list display resolution, camera and video formats, speaker output, and speech recognition in noise; its remote-assistance documentation additionally identifies bandwidth, latency, packet loss, and jitter as determinants of call quality. Task-oriented evaluation therefore distinguishes reading or identification accuracy, speech intelligibility and command success, capture usefulness, completion time, errors, and comfort. Battery endurance likewise requires a defined operating workload: capacity in watt-hours describes stored energy, whereas an endurance result concerns elapsed operation under stated display, recording, processing, and connectivity conditions. These are different measurements, not interchangeable performance claims. (support.realwear.com)

Organizational deployment includes configuration, application administration, and training in addition to distributing hardware. RealWear’s configuration application can obtain network settings by scanning a QR code generated through its companion application or a supported management tool. Its Collaborate integration with Microsoft Teams requires administrator-enabled permissions; version 2.0, introduced in March 2025, added inbound and outbound calling with additional permission requirements. Training also includes physical fitting and visual adjustment. In spatial-display systems, wearer-specific calibration can affect task accuracy: Microsoft’s Dynamics 365 Guides documentation states that an incorrectly configured interpupillary distance can displace HoloLens overlays by 1–2 cm. HoloLens is a related mixed-reality headset rather than an ordinary spectacle frame, but the example illustrates why calibration status belongs in deployment records and evaluation protocols. (support.realwear.com)

Laboratory demonstrations, manufacturer specifications, and independently measured operational outcomes answer different questions. Specifications document supported functions and declared operating limits; controlled experiments isolate effects of hardware or interface design; workplace studies test those effects within actual procedures. A simulated warehouse study involving 16 participants compared monocular and binocular displays, four interfaces, and a paper-list baseline. Interface design affected performance more than display type: graphic-based instructions reduced completion time by approximately 13% and errors by approximately 59%. A separate 2024 laboratory study reported mean picking times of 7.28 minutes with smart glasses and 12.81 minutes with printed lists, with error rates of approximately 1% and 10%, respectively. Its small-scale simulated warehouse and short task duration constrain interpretation: those figures establish results for the tested procedure, not full-shift endurance, organization-wide productivity, or clinical effectiveness. (pubmed.ncbi.nlm.nih.gov)

Privacy, security, and social use

Privacy concerns extend to both wearers and people within their cameras’ and microphones’ range. Photographs and video can contain bystanders, while audio can include nearby conversations rather than only the wearer’s commands. Capture, storage, transmission, and subsequent processing are distinct stages: collecting a recording does not necessarily upload it, and uploading it does not by itself specify its later use. Meta’s published prototype-testing policy illustrates these distinctions: media normally remains on testers’ devices, but may be sent to company servers for feedback or bug reporting; stored voice commands may be reviewed to improve and train speech recognition. These are conditions for that testing program, not universal properties of smart glasses. (meta.com) Spatial systems can additionally expose gaze, movement, and location information. Microsoft’s Dynamics 365 Guides documentation identifies separate permissions for eye tracking, camera, microphone, background spatial perception, and location; when optional location access is enabled, location data is stored in Microsoft Dataverse. (learn.microsoft.com)

Gaze deserves particular attention because interaction data can also support identification. In a 2021 study, Brendan David-John, Diane Hosfelt, Kevin Butler, and Eakta Jain demonstrated re-identification risks using eye-movement datasets, including data collected during natural viewing in virtual reality. Their experimental privacy mechanisms modified the gaze stream before passing it to applications. In one evaluated dataset, adding Gaussian noise reduced mean identification accuracy from 85% to 30%, with effects on application utility assessed separately. The result demonstrates a possible privacy–utility trade-off, rather than establishing that every eye-tracking device identifies its wearer or implements such protection. (arxiv.org)

Capture indicators make recording activity more observable, but do not themselves establish consent. Ray-Ban Meta documentation distinguishes the inward-facing notification light from the outward-facing capture LED, which signals photography or video recording. (ray-ban.com) On September 17, 2021, Ireland’s Data Protection Commission, together with Italy’s Garante, questioned whether the small indicator on the earlier Facebook View glasses effectively notified bystanders. The regulators requested evidence of field testing and a public information campaign; their statement concerned notice, not a finding that an illuminated LED supplied permission to record. (dataprotection.ie) Application permissions provide a different safeguard by controlling software access to sensors: for example, HoloLens settings allow eye-tracking permission to be changed under Settings > Privacy > Eye Tracking. Such controls authorize an application’s access from the wearer’s device, rather than obtain agreement from people in its surroundings. (learn.microsoft.com)

Security encompasses device authentication, companion-phone access, account association, and continuing software maintenance. HoloLens 2 provides a concrete enterprise example: Microsoft documents device encryption and, for Microsoft Entra ID users, Windows Hello for Business authentication through a PIN or iris recognition. (learn.microsoft.com) Ray-Ban Meta glasses pair with one Meta account at a time; their documented factory reset erases captures on the glasses and removes account associations. This is a device-level operation, distinct from managing recordings already transferred elsewhere. (ray-ban.com) Support termination is another lifecycle boundary. Google ended Glass Enterprise Edition sales on March 15, 2023, and support on September 15, 2023. Its announcement stated that existing devices and software could remain usable, but no further Google software updates were planned, while the unsupported Meet on Glass application could subsequently stop working. Operational hardware therefore need not remain supported hardware. (support.google.com)

Recording practices also intersect with workplace data governance and jurisdiction-specific law. The United Kingdom’s Information Commissioner’s Office explicitly includes wearable cameras and location-tracking body-worn devices in its worker-monitoring guidance. It calls for an identified lawful basis, clear purposes and privacy information, justified retention periods, and restricted access to collected information; employer authorization alone does not replace these requirements. (ico.org.uk) In California, Penal Code section 632 generally prohibits intentionally recording a confidential communication without all parties’ consent, subject to statutory exceptions. Confidentiality depends on the circumstances; the statutory definition excludes public gatherings and situations in which participants may reasonably expect to be overheard or recorded. This rule concerns confidential communications, not an undifferentiated prohibition on every photograph or recording made with glasses. (courts.ca.gov)

Engineering limitations and research directions

Display-equipped smart glasses face coupled optical constraints: a thin frame must accommodate a wide field of view, a sufficiently large eyebox—the region within which the pupil receives the image—and enough brightness for viewing against ambient light. Enlarging the eyebox distributes available light over more possible eye positions rather than increasing source power; widening the field of view also increases the range of propagation angles that the optical system must handle. Waveguides reduce bulk but introduce losses at coupling, propagation, and extraction stages. Consequently, improving viewing freedom can conflict with optical efficiency, while compensating through brighter illumination increases electrical demand. These constraints are particularly important in spatial displays, rather than camera-and-audio glasses without near-eye imagery. (pmc.ncbi.nlm.nih.gov)

Metasurfaces investigate whether subwavelength structures can perform optical functions otherwise requiring thicker or multilayer components. Their patterned elements control phase and diffraction, redirecting light into and out of a waveguide while managing wavelength-dependent behavior. In 2022, Hyunpil Boo and colleagues reported a single-layer, full-colour metasurface-waveguide prototype with 1080-pixel resolution, a field of view exceeding 40°, and overall input–output optical efficiency exceeding 1%. The work combined numerical design, nanofabrication, and device measurements; it did not demonstrate that these figures could be maintained in a complete battery-powered consumer product. Its focal-free implementation was a particular viewing strategy, not evidence that every metasurface waveguide supplies natural depth-dependent focus. (pmc.ncbi.nlm.nih.gov)

Holographic approaches instead reconstruct an image-bearing wavefront by modulating the phase or amplitude of illumination. Unlike presenting a flat image at one optical distance, wavefront reconstruction can assign different focal depths to different scene points. In 2024, Nature published a full-colour three-dimensional augmented-reality demonstration combining metasurface waveguide couplers with a phase-only spatial light modulator. Placing the modulator close to the input coupler reduced the need for a conventional projection-lens assembly. The researchers co-designed the waveguide geometry, couplers, and computational image formation, illustrating that compactness depends on the complete optical architecture rather than simply substituting a thinner component. The reported system was an experimental demonstration, not a specification for commercially supplied glasses. (nature.com)

Focus and aberration correction are also computational research problems. In 2017, Andrew Maimone, Andreas Georgiou, and Joel Kollin demonstrated holographic calculation at at least 90 Hz using a graphics processing unit, with per-pixel focal control and a unified model for focus, optical aberrations, and vision correction. Calibration incorporated defects between the light source and retina into the hologram calculation. An eyeglasses-like prototype achieved an 80° horizontal field of view, but its driving electronics were external, and the paper evaluated capabilities across several prototypes rather than one fully integrated device. Thus, measured optical performance and real-time computation did not establish simultaneous achievement of untethered operation, compact power supply, and all demonstrated imaging functions. (microsoft.com)

Pupil replication addresses another limitation: an image can disappear when the eye moves outside a narrow exit pupil. Waveguide extraction structures or computational holograms can produce laterally displaced copies of that pupil. Chenliang Chang and colleagues’ 2019 holographic Maxwellian display formed a $3 \times 3$ pupil array with 1 mm spacing by multiplexing differently directed plane waves. Its narrow beams produced an extended-depth-of-field image, distinct from reconstructing natural accommodation cues at multiple depths. Replication broadens the viewing region but does not create additional optical power; replica spacing and overlap also affect visibility and image duplication. Eye-tracked pupil steering provides an alternative, introducing dependence on tracking accuracy and response time. (pmc.ncbi.nlm.nih.gov)

System integration must accommodate sensor packages, interconnections, specialized computing, and heat removal within the same frame. Qualcomm’s Snapdragon AR2 architecture, announced on November 16, 2022, distributed functions among an AR processor, sensor-processing coprocessor, and connectivity subsystem, retaining latency-sensitive perception on the glasses while offloading heavier processing to a host. This architecture reduces on-frame computational requirements but introduces host and wireless dependencies. Thermal distribution is independently consequential: a 2020 steady-state modelling study, under a specified 5 W total load and 25 °C ambient temperature, calculated that dividing heat-generating components between both sides reduced ear-contact temperature from 51.0 °C to 43.6 °C. These were modelled conditions, not universal operating temperatures; smaller electronics alone do not eliminate the need to spread and dissipate their heat. (qualcomm.com)

References

1. Amazon Introduces 8 New Echo Devicespress.aboutamazon.com
2. Frequently Asked Questionsdevelopers.meta.com
3. Build Apps that See, Hear, and Respond: Explore What’s Possible With Wearables Device Access Toolkitdevelopers.meta.com
4. Vuzix Introduces Z100 Smart Glasses to Seamlessly Connect Workers with AI Optimization Toolsir.vuzix.com
5. 5 Benefits of Assisted Realityrealwear.com
6. First Look at the Future of Assisted Reality with RealWear Navigator 500realwear.com
7. HoloLens 2 hardwarelearn.microsoft.com
8. How to Connect to the Z100?support.vuzix.com
9. XREAL Air 2us.shop.xreal.com
10. What is mixed reality?learn.microsoft.com
11. A head-mounted three dimensional displayweb.stanford.edu
12. Google Glass - A Google X Moonshotx.company
13. Glass Enterprise Edition 2: faster and more helpfulblog.google
14. Glass Enterprise Edition Announcement FAQ - Glass Enterprise Edition Helpsupport.google.com
15. HMT-1 Device Overviewsupport.realwear.com
16. Using Voice Commandsrealwear.setupmyhmt.com
17. Satya Nadella and Terry Myerson: Build 2016news.microsoft.com
18. Microsoft HoloLens 2news.microsoft.com
19. Introducing the New Ray-Ban | Meta Smart Glassesabout.fb.com
20. New Ray-Ban | Meta Smart Glasses Styles and Meta AI Updatesabout.fb.com
21. Introducing Orion, Our First True Augmented Reality Glassesabout.fb.com
22. Orion: Llegan las verdaderas gafas de realidad aumentadaabout.fb.com
23. Snapdragon AR1 Gen 1 Platform Product Briefdocs.qualcomm.com
24. Ray-Ban Stories: How We Built Our First Smart Glassesabout.fb.com
25. Qualcomm Launches Snapdragon AR2 Designed to Revolutionize AR Glassesqualcomm.com
26. User Guide_XREAL One Seriesus.shop.xreal.com
27. Mixed Reality (MR) systemsglobal.canon
28. Review: Advanced light engines for AR/VR displayslcd.creol.ucf.edu
29. Waveguide holography for 3D augmented reality glassesnature.com
30. Design of a large field-of-view see-through near to eye display with two geometrical waveguidesopg.optica.org
31. Geometric lightguide for near-eye light field displaysopg.optica.org
32. Transparent XR displays from near-eye to head-updispelix.com
33. Color, light, and materialslearn.microsoft.com
34. Enhancing the Optical Efficiency of Near-Eye Displays with Liquid Crystal Opticsapi.creol.ucf.edu
35. Comfortlearn.microsoft.com
36. Hologram stabilitylearn.microsoft.com
37. Fundamental concepts | ARCore | Google for Developersdevelopers.google.com
38. HoloLens 2 Moving Platform Modelearn.microsoft.com
39. Sensors - Windows appslearn.microsoft.com
40. Enabling interaction between mixed reality and robots via cloud-based localization - Microsoft Researchmicrosoft.com
41. Visual-Inertial Monocular SLAM with Map Reusearxiv.org
42. Microsoft HoloLens 2: Improved Research Mode to facilitate computer vision research - Microsoft Researchmicrosoft.com
43. Spatial mapping - Mixed Realitylearn.microsoft.com
44. HoloLens environment considerationslearn.microsoft.com
45. Improve visual quality and comfortlearn.microsoft.com
46. Alignment of the Virtual Scene to the Tracking Space of a Mixed Reality Head-Mounted Displayarxiv.org
47. How Voice Control Workssupport.realwear.com
48. Gaze and commit - Mixed Reality | Microsoft Learnlearn.microsoft.com
49. Eye-gaze-based interaction - Mixed Reality | Microsoft Learnlearn.microsoft.com
50. Eye-gaze and dwell - Mixed Reality | Microsoft Learnlearn.microsoft.com
51. Point and commit with hands - Mixed Reality | Microsoft Learnlearn.microsoft.com
52. Spatial audio best practices - Mixed Reality | Microsoft Learnlearn.microsoft.com
53. Head-gaze and dwell - Mixed Reality | Microsoft Learnlearn.microsoft.com
54. Device Featuressupport.realwear.com
55. Device Specificationssupport.realwear.com
56. Building Custom Apps for RealWear Just Got a Whole Lot Easier - RealWear Blogrealwear.com
57. Bring your mobile app to AI glassesdevelopers.meta.com
58. Meta Wearables Device Access Toolkit for Androidgithub.com
59. CameraAccess MainActivity.ktgithub.com
60. Explain access to more sensitive informationdeveloper.android.com
61. HoloLens Research Mode - Mixed Realitylearn.microsoft.com
62. Introducing the Meta AI App: A New Way to Access Your AI Assistantabout.fb.com
63. On-glass GenAI demoqualcomm.com
64. Llama 3.2 Vision Model Cardgithub.com
65. New Ray-Ban Meta Smart Glasses Styles and Meta AI Updatesabout.fb.com
66. How We’re Advancing Accessibility at Metaabout.fb.com
67. Ray-Ban | Meta Glasses Are Getting New AI Features and More Partner Integrationsabout.fb.com
68. BMW of North America Introduces Smart Glasses and New Technician Communication Systems at U.S. Dealerships to Reduce Repair Times and Improve Customer Satisfaction.press.bmwgroup.com
69. Manufacturing Solutionsrealwear.com
70. Document Viewersupport.realwear.com
71. Inspection AI - Hands-Free Voice-Guided Inspections on RealWearapplivity.com
72. DHL successfully tests augmented reality application in warehousedhl.com
73. Children’s Digital Art Ability Training System Based on AI-Assisted Learning: A Case Study of Drawing Color Perceptionfrontiersin.org
74. Ever Wanted To Put Subtitles On Your Daily Life?xrai.glass
75. How AI Glasses Are Empowering People With Disabilitiesabout.fb.com
76. Augmented Reality and Virtual Reality in Medical Devicesfda.gov
77. xvision Spine System: K190929 clearance letter and 510(k) summaryaccessdata.fda.gov
78. Fit Adjustmentssupport.realwear.com
79. RealWear Navigator 500 User Guideamd.sysgration.com
80. Collaborate Technical Infosupport.realwear.com
81. Calibrate your HoloLens to use with Dynamics 365 Guideslearn.microsoft.com
82. Influences of augmented reality head-worn display type and user interface design on performance and usability in simulated warehouse order pickingpubmed.ncbi.nlm.nih.gov
83. Exploring the impact of augmented reality smart glasses on worker well-being in warehouse order pickingdoi.org
84. How we test new productsmeta.com
85. Permissions required to use the Dynamics 365 Guides HoloLens applearn.microsoft.com
86. A privacy-preserving approach to streaming eye-tracking dataarxiv.org
87. Data Protection Commission statement concerning Facebook View (glasses)dataprotection.ie
88. HoloLens security overviewlearn.microsoft.com
89. Data protection and monitoring workersico.org.uk
90. Judicial Council of California Civil Jury Instructions (2026)courts.ca.gov
91. Toward the next-generation VR/AR optics: a review of holographic near-eye displays from a human-centric perspectivepmc.ncbi.nlm.nih.gov
92. Metasurface wavefront control for high-performance user-natural augmented reality waveguide glassespmc.ncbi.nlm.nih.gov
93. Full-colour 3D holographic augmented-reality displays with metasurface waveguidesnature.com
94. Holographic Near-Eye Displays for Virtual and Augmented Realitymicrosoft.com
95. Computational holographic Maxwellian near-eye display with an expanded eyeboxpmc.ncbi.nlm.nih.gov
96. Computational holographic Maxwellian near-eye display with an expanded eyeboxnature.com
97. Thermal Model and Countermeasures for Future Smart Glassespmc.ncbi.nlm.nih.gov