StreetViewAI: Google’s Multimodal AI Brings Conversational Street View Navigation to the Visually Impaired

According to Google Research and DeepMind, StreetViewAI is a multimodal AI system designed to overcome street view maps’ visual dependency for visually impaired users. The system enables access to Google Street View’s repository of over 220 billion images across more than 100 countries through natural, conversational AI interaction. StreetViewAI is built on the Google Gemini Flash 2.0 multimodal model and comprises three core subsystems: AI Describer, AI Chat Agent, and AI Tour Guide.

In real-world testing, 11 visually impaired participants performed destination search and free exploration, interacting with the AI Chat Agent 917 times and the AI Describer 136 times. The researchers reported an 86.3% accuracy rate in responses with 3.9% incorrect answers, and more than 90% of participants preferred voice commands, with testers noting that StreetViewAI can describe appearances and accessibility and “lead you to the door.”