RoboSnap: One-Shot Real-to-Sim Scene Generation for Generalizable Robot Learning and Evaluation
RoboSnap: One-Shot Real-to-Sim Scene Generation for Generalizable Robot Learning and Evaluation
This paper introduces RoboSnap, a real-to-sim framework that generates simulation-ready scenes from a single RGB image by separating physics-critical foreground assets for stable interaction from a 3D Gaussian splatting visual layer for faithful background appearance. The authors also release DROID-Sim, a companion dataset comprising 564 real-world scenes from the DROID benchmark. Experiments demonstrate that RoboSnap enables reliable trajectory replay, supports task-specific synthetic data generation for policy training, and provides meaningful sim-real correlation for evaluation, positioning real-to-sim methods as reusable infrastructure for robot learning.
The authors present RoboSnap, a framework converting single RGB images into physically stable simulation scenes using layered physics and visual reconstruction, validated on DROID datasets with reported sim-real correlation for policy evaluation.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
recovered scenes, supports task-specific synthetic data generation for policy training, and yields meaningful sim-real correlation for policy evaluation. To further support real-to-sim research, we introduce DROID-Sim, a real-to-sim companion dataset constructed from 564 real-world scenes in DROID. Extensive experiments suggest that the value of real-to-sim methods lies not only in high-fidelity visual reconstruction
Open source S4
, but in turning real environments into reusable infrastructure for robot learning and evaluation.
Open source S5
RoboSnap: One-Shot Real-to-Sim Scene Generation for Generalizable Robot Learning and Evaluation
Open source S1
arXiv:2607.06699v2 Announce Type: replace Abstract: Recovering real-world scenes as interactive simulation environments can enable generalizable robot learning and reproducible policy evaluation. However, constructing scenes that are both physically stable and visually faithful remains slow and expensive. In this work, we present RoboSnap, a real-to-sim framework that turns a single RGB image into a simulation-ready
Open source S2
scene. The key idea is a layered design that separates the physics-critical interaction area from the surrounding visual context: collision-aware foreground assets are refined for stable robot interaction, while a 3D Gaussian splatting visual layer preserves faithful background appearance under novel views. Experiments on DROID scenes and real-robot tasks show that RoboSnap achieves reliable trajectory replay in the
Open source S3
Source：arXiv Robotics — research abstracts · arxiv.org