Generalization problem

This is an extremely outstanding job! May I ask whether the office and outdoor navigation tasks displayed on the webpage were trained in these environments in advance, or are these zero-shot results?

<img width="1235" height="846" alt="Image" src="https://github.com/user-attachments/assets/6f49ac55-3d03-4f1a-827e-bd7b5d0260bd" />