Rest2Art

Articulated Object Reconstruction from Rest-State Observation

Daeun Lee, Jaeah Lee*, Woosung Kim*, Haebeom Jung, Jaesik Park
Seoul National University
*Indicates Equal Contribution
ECCV 2026
arXiv Code 🤗 Dataset BibTeX

Rest2Art recovers part-level geometry and joint parameters from rest-state observations alone, turning a single closed-state capture of an everyday object into a simulation-ready articulated asset.

Abstract

Building interactive digital twins requires recovering both 3D geometry and the kinematic structures that govern how objects articulate. Yet existing methods for articulated object reconstruction require explicitly observable motion from multiple articulation states. We introduce a rest-state formulation that reconstructs articulated objects from a single closed configuration, an inherently ill-posed setting where geometry, semantics, and motion priors compensate for the absence of motion cues. Our framework adopts an explicit mesh as an intermediate representation for cross-model verification and fusion, reconciling noisy outputs from vision-language and segmentation models into spatially consistent part structures. To estimate joint parameters without observed motion, we use a video diffusion model to synthesize articulation hypotheses and validate them through geometric consistency. Our approach achieves accurate part decomposition and physically plausible articulation, performing competitively with motion-observing baselines across reconstruction-based, generation-based, and modular pretrained-model families.

Method

Rest2Art method overview

Rest2Art turns any rest-state mesh into part-wise volumetric meshes with joint parameters, without ever observing the object move, in three steps:

  • A vision-language model and a segmentation model co-refine the part hierarchy and per-view masks, treating their disagreement as the correction signal, and confidence-weighted evidence lifting carries the result onto mesh faces.
  • A video diffusion model supplies articulation hypotheses in place of observed motion, and joint parameters are resolved by geometric consistency with the mesh rather than taken from the video.
  • Each part shell is solidified into a closed volume, making the output usable in physical simulation.

Interactive Results

Reconstructed articulated objects across input modalities; hover a thumbnail to see its input type. Scroll to pick one, then play its articulation.

Loading 3D model…

Applications

Virtual Asset Integration

Because Rest2Art recovers part geometry and joint parameters together, any object it reconstructs can be inserted into any target scene and interacted with there.

Real-to-Sim-to-Real

With Rest2Art, a real object is reconstructed into an articulated asset, imported into simulation, and used to execute manipulation on a real robot.

Sim

Real

Policy Learning

Rest2Art exports each reconstruction as a URDF, so its outputs can be used directly to train manipulation policies. Following DexArt, we learn a policy for opening a laptop lid on a PartNet-Mobility laptop reconstructed from a single open configuration.

BibTeX

@inproceedings{lee2026rest2art,
  title={Articulated Object Reconstruction from Rest-State Observation},
  author={Lee, Daeun and Lee, Jaeah and Kim, Woosung and Jung, Haebeom and Park, Jaesik},
  booktitle={European Conference on Computer Vision (ECCV)},
  year={2026}
}