DROID background asset contains a baked-in static robot, causing duplicate arms

Description

The bundled DROID scene background contains a static robot arm, while OmniGibson also instantiates the controllable simulated robot (robot0). This produces two visually overlapping robot arms: one remains completely static and the other moves with the simulation.

This appears to be a background asset / real-to-sim preprocessing issue rather than a duplicate robot instance in the scene registry.

Steps to reproduce

From the repository root, run:

python scripts/pipeline/C_application/stages/0_smoke_random_actions.py \
  --scene-json "$(pwd)/assets/scenes/DROID/droid_desk_serve_fruits/droid_desk_serve_fruits_scene_state_latest.json" \
  --external-sensors-config nv_franka_droid_high_res \
  --n-steps 20

Inspect either external-camera view in the generated smoke-test video.

The same behavior is visible with the standard nv_franka_droid camera configuration; high resolution only makes it easier to see.

Actual behavior

• The simulated robot0 moves in response to random actions.
• A second robot arm remains fixed in the reconstructed background throughout the video.
• In a first/middle/last-frame comparison, the background arm keeps exactly the same pose while the simulated arm changes pose.

Expected behavior

The reconstructed background should be robot-free, so only the controllable simulated robot is visible.

Asset-level diagnosis

The scene JSON declares only one robot, robot0. It also loads:

mesh_background_0
  usd_path: ../../../backgrounds/mesh_backgrounds/droid_v1.usd
  fixed_base: true

droid_v1.usd references droid_v1.usdz. The static arm appears to be baked into this mesh/texture reconstruction. Adding the simulated robot on top therefore creates the duplicate-arm artifact.

A robot mask/inpainting pass before background reconstruction, or a robot-free replacement for droid_v1.usdz, should resolve the issue.

Environment

• SimFoundry commit: 9e34ebefcd02
• Ubuntu 24.04.4
• NVIDIA GeForce RTX 5090
• NVIDIA driver 580.173.02
• Smoke test completed successfully (20 steps, H.264 output)

I can provide the generated comparison frames or video if useful.

Start with assets/scenes/DROID/droid_desk_serve_fruits/droid_desk_serve_fruits_scene_state_latest.json and inspect the referenced backgrounds/mesh_backgrounds/droid_v1.usd and droid_v1.usdz assets. Reproduce the issue with scripts/pipeline/C_application/stages/0_smoke_random_actions.py and compare the external-camera frames. Done means the reconstructed background contains no static robot arm, leaving only controllable robot0 visible.