Neural Implicit Representation for Building Digital Twins of Unknown Articulated Objects

TL;DR

This paper presents a novel neural implicit approach for creating digital twins of unknown articulated objects from two RGBD scans, accurately modeling shape, parts, and joint articulations without prior shape assumptions.

Contribution

It introduces a two-stage method that reconstructs object shape and recovers articulation models, handling multiple movable parts without relying on shape priors.

Findings

More accurate and stable reconstructions than previous methods

Handles multiple movable parts without shape priors

Explicitly models point correspondences and kinematics

Abstract

We address the problem of building digital twins of unknown articulated objects from two RGBD scans of the object at different articulation states. We decompose the problem into two stages, each addressing distinct aspects. Our method first reconstructs object-level shape at each state, then recovers the underlying articulation model including part segmentation and joint articulations that associate the two states. By explicitly modeling point-level correspondences and exploiting cues from images, 3D reconstructions, and kinematics, our method yields more accurate and stable results compared to prior work. It also handles more than one movable part and does not rely on any object shape or structure priors. Project page: https://github.com/NVlabs/DigitalTwinArt