Request for Camera-to-Base Extrinsics for BimanualYAM (Missing Base Poses)

**Hi MolmoAct2 Team,**

First of all, thank you for releasing the MolmoAct2 and the amazing BimanualYAM dataset! It's a fantastic resource for the community.

I am currently trying to map the visual observations to the robot's kinematic frames for simulation and control. I noticed in Issue #2 that the `T_midpoint_from_camera` matrix was kindly provided. However, **the exact extrinsic transformations from the camera (or the midpoint) to the respective left and right YAM arm bases are still missing.**

From the hardware setup images, the two YAM arms are mounted on a custom aluminum extrusion rail. Because of this custom setup, simply assuming symmetry isn't enough for precise kinematics, as we don't know the exact 3D offsets (e.g., the depth offset from the midpoint to the rail, or the exact Z-height of the base links relative to the table). Without knowing how the arm bases are precisely positioned relative to that `Midpoint`, we cannot chain the transformations to get `T_camera_to_left_base` and `T_camera_to_right_base`.

**Could you please provide any of the following to help us accurately align the visual and kinematic frames?**

1. The full $4 \times 4$ transformation matrices `T_midpoint_from_left_base` and `T_midpoint_from_right_base`.
2. OR the direct `T_base_from_camera` matrices for both arms.
3. OR a combined URDF, MJCF, or CAD file of your bimanual table setup, which would naturally contain all the relative positional data.

Knowing the exact 6-DoF spatial relationship is crucial for re-implementing or simulating the tasks accurately without introducing misalignment errors. Any explicit matrices or environment models you could share would be hugely appreciated!

Thank you again for your time and your great work!