SceneRig builds simulation-ready 3D scenes from single images
Original title: SceneRig: An Agentic System for Simulation-Ready 3D Scene Reconstruction from Single Images
An agent edits and validates scenes in a physics simulator so reconstructed objects stay physically stable.
SceneRig is an agentic system that turns one image into an editable, simulation-ready 3D scene using geometric, visual and physical tools. An agent then runs edit-and-validate loops inside a physics simulator to reach physical stability, addressing misplacement from post-hoc physics settling. The authors say it improves segmentation, appearance and placement and better supports downstream robotics than a baseline.
Unconfirmed: Downstream robotics gains are reported by the authors; the baseline method is not named in the abstract.
- Scenes are meant to predict real-robot policy outcomes more accurately than the baseline method.
- Reconstructed scenes more reliably support replay of recorded manipulations without trajectory adaptation.
- The paper says visual similarity alone does not make a scene usable under manipulation.
From the original article
1 min read at arxiv.orgSimulation is a powerful tool for robotics, but building diverse, simulation-ready scenes at scale remains difficult. Reconstructing scenes from single images is a promising alternative, but visual similarity alone is insufficient: objects must also be recovered, accurately placed, and physically grounded.
By Rundong Luo · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.