MoRE: Motion-Regularized Encoding for Articulated 3D Generation

Guantian Zheng; Shan Jin; Tim Tso; Hongfei Cai

cs.CVOct 21, 2026

MoRE: Motion-Regularized Encoding for Articulated 3D Generation

Authors: Guantian Zheng, Shan Jin, Tim Tso, Hongfei Cai

Organizations: Manifest Limited · School of Mechanical and Aerospace Engineering, Nanyang Technological University · Department of Engineering, University of Cambridge

Abstract

Generating an articulated object from a single image requires recovering both its appearance and the mechanism governing its parts. A compact encoding of joint attributes does not by itself ensure that reconstruction preserves the resulting motion. We present MoRE, Motion-Regularized Encoding for articulated 3D generation. MoRE represents coordinate-aligned joint lines with axis classification and two axis-orthogonal Plücker moment coefficients, removing the arbitrary choice of an origin along the axis. During representation learning, decoded joint hypotheses are used to move corresponding three-dimensional anchors, and their motions are compared with those of annotated joints. A differentiable expectation over joint types and axis directions couples attribute reconstruction to the motion it induces. An image-conditioned flow learns the resulting functional latents, while a multi-state-adapted Gaussian model supplies appearance. At inference, recovered part assignments and joints articulate a single set of rest-pose Gaussians without regenerating the object for each state. Experiments on PartNet-Mobility and ACD show stronger part-box agreement, articulated-state geometry and visual similarity than the compared baselines, while joint prediction reduces axis, origin and range errors relative to ArtiLatent. Controlled ablations support the joint-line encoding and show that motion supervision reduces anchor-motion error. Code and pretrained checkpoints will be released in the future.

Figures & tables

Fig. 1: From a single image to functional voxels and articulated configurations. Earlier attribute-only checkpoint.

Fig. 2: MoRE overview. A shared DINOv3 conditions sparse support, functional and appearance generation; an adapter conditions the appearance branch. Training adds expected anchor-motion error to attribute reconstruction; appearance is adapted across states. Cabinet photo: illustrative input, not a tested generation result; PNGimg [ 24 ] , CC BY-NC 4.0.

PartNet-Mobility
Method	RS- $d_{\mathrm{gIoU}}\downarrow$	AS- $d_{\mathrm{gIoU}}\downarrow$	RS-CD $\downarrow$	AS-CD $\downarrow$	AOR $\downarrow$	CLIP $\uparrow$
SINGAPO [ 17 ]	0.5119	0.5166	0.0040	0.0202	0.0097	0.8779
ArtFormer [ 26 ]	1.3165	1.3212	0.0504	0.1849	0.0191	0.7877
PhysX-3D [ 1 ]	0.7610	0.7637	0.0115	0.1204	0.2125	0.7972
PhysX-Anything [ 2 ]	0.9596	0.9661	0.0097	0.1106	0.0056	0.8859
Articulate-Anything [ 13 ]	0.6865	0.6969	0.0107	0.1230	0.0056	0.8795

TABLE I: Single-image articulated generation. Green marks the best values on each dataset, including ties.

Method	Type (%) $\uparrow$	Axis err. ( ∘ ) $\downarrow$	Origin $\downarrow$	Range (rot.) ( ∘ ) $\downarrow$	Range (trans.) $\downarrow$
SINGAPO [ 17 ]	97.37	1.29	0.39	6.73	0.0265
ArtiLatent [ 3 ]	99.17	1.14	0.10	7.37	0.0159
MoRE	99.16	1.03	0.04	6.29	0.0107

TABLE II: Joint prediction on PartNet-Mobility. Green marks the best values; baseline scores are from ArtiLatent Table 3.

Encoding	$\mathcal{L}_{\mathrm{motion}}$	Axis err. ( ∘ ) $\downarrow$	Line dist. $\downarrow$	RMSE-R $\downarrow$	RMSE-P $\downarrow$	Range-R ( ∘ ) $\downarrow$	Range-P $\downarrow$
Axis + origin	off	1.24	0.0253	0.0477	0.0214	7.28	0.0132
Axis + moment	off	1.17	0.0229	0.0431	0.0207	7.20	0.0123
Axis + origin	on	1.15	0.0206	0.0383	0.0185	6.31	0.0117
Axis + moment (MoRE)	on	1.03	0.0191	0.0362	0.0168	6.29	0.0107

TABLE III: Joint-line encoding and motion regularization on PartNet-Mobility. Green marks the best values.

Fig. 3: Three objects across four configurations, with reference above reconstruction in each pair. Known joints articulate Gaussians decoded from encoder latents.

Study	Setting	PSNR (dB) $\uparrow$	IoU $\uparrow$
Alignment	Before correction	15.8	0.781
	After correction	20.5	0.885
Color source	Decoded	19.72	0.872
	GT back-projected	22.51	0.879

TABLE IV: Known-joint appearance reconstruction and diagnostics.

Articulated Object Reconstruction

Articulated object reconstruction recovers 3D part geometry, joints, and motion from images, video, or depth for digital twins and robot interaction.

See all 60 papers

Artic-O: End-to-End Articulated Object Reconstruction via Latent Geometry Learning
Jun 20, 2026Xuyang Wang, Zhenyu Li, Jian Ding +4Articulated Object Reconstruction 3D Reconstruction
Australian National University, Canberra, Australia · King Abdullah University of Science and Technology, Thuwal, Saudi Arabia
Instruct-Particulate: Scaling Feed-Forward 3D Object Articulation with Kinematic Control
Jun 12, 2026Ruining Li, Yuxin Yao, Matt Zhou +5Articulated Object Reconstruction 3D Part Segmentation
University of Oxford · University of Cambridge · 3Nanyang Technological University
PAct: Part-Decomposed Single-View Articulated Object Generation
Feb 16, 2026Qingming Liu, Xinyue Yao, Shuyuan Zhang +4Articulated 3D Asset Generation 3D Asset Generation
The Chinese University of Hong Kong, Shenzhen, China · DexForce Technology, China

Jun 20, 2026J/K move · Enter open · S save

Artic-O: End-to-End Articulated Object Reconstruction via Latent Geometry Learning

Xuyang Wang, Zhenyu Li, Jian Ding, Habib Slim+3

Australian National University, Canberra, Australia · King Abdullah University of Science and Technology, Thuwal, Saudi Arabia

Articulated Object Reconstruction 3D Reconstruction Sparse-View 3D Reconstruction Feed-Forward 3D Reconstruction

Open paper AskPDF arXiv

Reconstructing articulated objects from sparse images requires recovering complete geometry, movable parts, and motion parameters. Recent methods typically separate geometry reconstruction, part reasoning, and articulation estimation into different stages. This separation can weaken consistency between shape, active parts, and motion, while also incurring substantial inference cost. We introduce Artic-O, an end-to-end, feed-forward framework for articulated object reconstruction via latent geometry learning. Instead of fitting geometry in image or view space, Artic-O maps sparse multi-state observations into a pretrained latent geometry space, where a frozen flow-matching decoder provides a complete-shape prior for recovering visible and occluded structures. To connect geometry with articulation, Artic-O fuses visual tokens, geometry latents, and point-wise decoder features in an image-grounded part-reasoning module for active-part segmentation and articulation prediction. We further train the model with a geometry-to-articulation curriculum and a decoupled two-pass strategy to balance reconstruction and part-level supervision. On PartNet-Mobility, Artic-O achieves strong reconstruction quality while being substantially more efficient than LARM, a strong prior method. It reduces Chamfer Distance, improves F-score, and achieves comparable or better articulation accuracy across most joint metrics, while reducing inference time from 9 minutes to about 0.3 seconds per object.

Abstract

Articulated Object Reconstruction

Explore similar work

Artic-O: End-to-End Articulated Object Reconstruction via Latent Geometry Learning

Instruct-Particulate: Scaling Feed-Forward 3D Object Articulation with Kinematic Control

PAct: Part-Decomposed Single-View Articulated Object Generation