TLDRocket
Sign in

StreetReaderAI: Towards making street view accessible via context-aware multimodal AI

Google Research

Google built an AI tool that lets blind users explore Street View by voice, describing streets, crossings, and landmarks in real time. It turns a purely visual product into something screen reader users can actually navigate and ask questions about.

Based on reporting by Google Research — read the original for the full story.

Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error

Google Street View has always been a visual medium in the strictest sense: 220 billion images across more than 110 countries, and not one pixel of it accessible to someone using a screen reader. That's the gap Google Research is trying to close with StreetReaderAI, a prototype detailed in a paper presented at UIST'25 that pairs Gemini's multimodal reasoning with a rebuilt, keyboard-and-voice navigation layer.

The mechanics are straightforward once you see them. Arrow keys pan the view and StreetReaderAI narrates the heading aloud — "Now facing: Northeast" — while also flagging whether there's a landmark in front of you or room to keep walking. Pressing up or down triggers a "virtual step," with the system reporting distance traveled and nearby points of interest as you go. Two Gemini-backed subsystems do the heavy lifting: AI Describer, which narrates a scene either in safety-focused mode or as a tour guide dropping in historical trivia, and AI Chat, which holds a running memory of everything you've seen in a session — up to 1,048,576 tokens, or roughly 4,000 images worth of context — so you can ask something like

Read more about this at: Google Research

Related stories

The daily briefing

Every AI story that matters, in your inbox by 8am.

TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the day in two minutes. Follow companies and topics for alerts, or get the briefing in Slack. Free, no spam, unsubscribe anytime.