SimulationBrief
Fei-Fei Li presents Atlas, an AI that rebuilds Stanford's plaza from 33 photos
Researcher Fei-Fei Li presented Atlas, an AI model that builds coherent virtual places from a few images, partly to train robots in simulation, to specialists in robot learning. The Chinese site Leiphone published a transcript of her talk.
The CoRL conference, a major gathering for research on robot learning, opened in Austin in November 2026.
To illustrate what Atlas does, Fei-Fei Li showed the main plaza on the Stanford campus. With 33 images, one of which shows the memorial chapel, the model rebuilt almost the whole plaza, she said.
The team says it has recently slimmed the model down so that it runs in real time, at 15 frames per second.
For robots, World Labs is testing Atlas to train navigation robots in a simulated world, and to link the real world to simulation. Fei-Fei Li speaks of potential applications.
Source
- 雷峰网 Leiphone · leiphone.com
Also worth reading
Illustration, not a photo of the robot