The real-world deployment setup #1

Hi, thank you for sharing this great work!

I have four questions about the real-world deployment setup:

1. Are the depth images from both ZED 2i cameras resized directly from 1280 × 720 to 48 × 27, preserving the full field of view? Or is any cropping applied to the top, bottom, left, or right before resizing?

2. Where did you purchase the mounts used to attach the front and rear cameras to the G1 torso? If they are custom-designed, would you be willing to share the CAD files or provide some details about their design?

3. How did you align the positions and orientations of the front and rear cameras on the real robot with those used in training? Did you use a specific calibration procedure to measure the camera poses relative to the torso, or were they aligned approximately based on the mounting geometry?

4. Can other depth cameras, such as Orbbec cameras, be used instead of the ZED 2i for real-world deployment? What should we pay particular attention to—for example, field of view, camera intrinsics, mounting poses, depth range, frame rate, latency, and preprocessing? Would using a different camera require retraining or fine-tuning the student policy?

Thank you for your help!