People read maps with ease. We look around and know our place. For computers, this is hard. This problem slowed tech growth for years.
Old vision tools fail often. They get mixed up by new views. A photo and a floor plan look far apart. Machines cannot link them well.
 |
| New Tools of Maps |
A team at Cornell found a fix. They made a new vision tool. It links photos to floor plans. It works down to each pixel.
The study showed up at NeurIPS 2025. It also appeared on arXiv.
Most tools only match like images. They fail when views change. This new tool works across types. It links images that look very different.
The team named it C3Po. The name comes from Star Wars. Kuan Wei Huang led the work. Three other Cornell researchers helped.
The team also built a big dataset. It is called C3. It pairs photos with floor plans. This helps machines learn fast.
The impact may be huge. Indoor maps can get better. Robots can move with care. 3D maps can feel more real.
