Waymo researcher presents EMMA, an end-to-end multimodal autonomous driving system built on Google's Gemini foundation model. The system uses only camera inputs and routing information to generate driving waypoints, achieving state-of-the-art performance on benchmarks while providing explainable reasoning for driving decisions. EMMA demonstrates superior generalization capabilities for handling rare driving scenarios compared to traditional modular autonomous driving systems, and shows promise for scaling Waymo's technology to new cities globally.
•17m watch time
Sort: