Articulated Object Reconstruction from Rest-State Observation

2026-07-30

AI Summary

Reconstructing the 3D structure and motion of articulated objects from a single static observation is highly ill-posed due to the absence of explicit motion cues. This work proposes a novel approach that operates solely from a closed, static configuration by jointly leveraging geometric, semantic, and motion priors. It achieves precise part decomposition through a vision-language model and a segmentation network, generates and validates physically plausible motion hypotheses using a video diffusion model, and employs explicit mesh representations as an intermediate representation to optimize joint parameters via geometric consistency. To the best of our knowledge, this is the first method to enable articulated object reconstruction without any dynamic observations, demonstrating competitive performance across multiple benchmarks and significantly improving the accuracy of both structural and motion recovery.

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 reconstruction-based, generation-based, and modular pretrained-model baselines.

Problem

• Articulated Object Reconstruction
• Rest-State Observation
• Kinematic Structure
• 3D Geometry
• Motion Priors

Innovation

• articulated object reconstruction
• rest-state observation
• video diffusion model
• explicit mesh representation
• geometric consistency