AI Navigation Assistance for Visually Impaired Users

Independent navigation is one of the most consequential accessibility challenges. A blind person who can navigate their city independently has fundamentally different life opportunities than one who depends on sighted assistance for every trip. AI is augmenting traditional mobility tools (white canes, guide dogs, memorized routes) with real-time environmental information that was previously available only through sight.

Current Tools

Smartphone Apps

Be My Eyes provides on-demand visual assistance through both AI (Be My AI, powered by GPT-4 vision) and human volunteers. Users point their phone camera at their surroundings and receive spoken descriptions. While not a turn-by-turn navigation tool, it provides environmental context that supports wayfinding.

Google Maps and Apple Maps both offer accessible turn-by-turn navigation with voice guidance. Google’s Detailed Voice Guidance mode provides additional callouts for blind users: announcing distances to turns further in advance, describing the direction of travel, and providing intersection information.

Soundscape (originally by Microsoft, now open-sourced) uses 3D spatial audio to place virtual beacons at destinations and points of interest, creating an audio “landscape” that helps users build a mental map of their surroundings.

Lazarillo provides GPS navigation supplemented with information about nearby points of interest, bus stops, and intersections, specifically designed for blind and low-vision users.

Wearable Devices

Research teams have developed wearable prototypes that combine:

RGB-D cameras (capturing both color and depth) mounted on glasses frames
Haptic feedback through wristbands, insoles, or belt-mounted vibrators
Bone-conducting earphones for spatial audio guidance
AI processing for obstacle detection and path planning

In controlled studies, users of these systems achieved navigation speeds comparable to cane-based navigation, with smoother turning and more efficient pathfinding. One prototype from a multidisciplinary research team integrates a 3D-printed glasses frame, ultrathin artificial skins for haptic feedback, and triboelectric smart insoles for training via a virtual reality platform.

Smart Canes

AI-enhanced canes combine traditional white cane functionality with electronic sensors. Products like WeWalk integrate ultrasonic sensors, GPS, and a smartphone connection to provide obstacle detection above waist height (which traditional canes miss) and navigation guidance.

What AI Enables

Obstacle Detection Beyond Cane Range

Traditional white canes detect obstacles at ground level within arm’s reach. AI-powered computer vision detects obstacles at a distance and at all heights: overhanging branches, head-height signs, approaching bicycles, and open car doors.

Scene Description

AI converts visual environmental information into spoken descriptions: “You are approaching a crosswalk. The pedestrian signal shows a walk sign. There is a coffee shop on your left and a bank on your right.” This contextual awareness helps users build mental models of unfamiliar areas.

GPS does not work indoors, where blind users face some of their greatest navigation challenges: airports, hospitals, shopping centers, and office buildings. AI combined with Bluetooth beacons, computer vision, and inertial sensors can provide indoor positioning and turn-by-turn guidance.

Transit Assistance

AI tools can identify approaching buses by route number, locate subway platform edges, and guide users to specific train doors or seats. Real-time transit information delivered through accessible apps reduces the uncertainty that makes public transit challenging for blind travelers.

Limitations

GPS accuracy in urban canyons (between tall buildings) degrades to 10-30 meters, which is insufficient for pedestrian navigation that requires meter-level precision.

Camera-based systems require adequate lighting and clear line-of-sight, and they consume significant battery power.

Latency between obstacle detection and user notification must be under one second for walking-speed navigation. Processing delays reduce safety margins.

Environmental complexity challenges AI in ways that would be trivial for a sighted person: distinguishing a curb from a step, reading construction detour signs, or interpreting temporary obstacles.

User trust develops slowly. Users must trust the system enough to act on its guidance but not so completely that they abandon other orientation techniques.

For obstacle detection specifics, see computer vision for accessibility: object detection. For haptic feedback approaches, read AI haptic feedback for accessibility.

Key Takeaways

AI navigation tools augment traditional mobility aids (canes, guide dogs) with real-time environmental information unavailable through other means.
Smartphone apps (Be My Eyes, Google Maps with Detailed Voice Guidance, Lazarillo) provide accessible navigation today.
Wearable prototypes combining cameras, haptic feedback, and spatial audio show promising results in controlled studies.
Indoor navigation remains a significant unsolved challenge, with Bluetooth beacon and computer vision approaches in development.
GPS accuracy, battery consumption, latency, and environmental complexity are ongoing technical limitations.

Sources

Microsoft Soundscape — open-source 3D audio navigation: https://github.com/microsoft/soundscape
Be My Eyes — AI and volunteer visual assistance: https://www.bemyeyes.com/
WeWalk — smart cane with ultrasonic sensors and GPS: https://wewalk.io/
WHO vision impairment and blindness fact sheet: https://www.who.int/news-room/fact-sheets/detail/blindness-and-visual-impairment