ArtMesh: Part-Aware Articulated Mesh Fields with Motion-Consistent Dynamics

Anonymous Authors

ArtMesh teaser
We present ArtMesh, a mesh-native method for reconstructing articulated objects explicitly as connected triangle meshes with per-part rigid motion from multi-view images in start and end states. Existing 3D Gaussian Splatting pipelines for articulated reconstruction inherit the unstructured point-based geometry of their splatting base, which provides no surface topology for reasoning about part boundaries or enforcing motion consistency along the object's connectivity. ArtMesh instead builds on a mesh-based differentiable rendering backbone, enabling part-aware dynamics to act directly on the structured topology. To make the topology compatible with articulation, we introduce part-aware restricted Delaunay remeshing, producing connected submeshes whose triangles do not cross semantic part boundaries. The dynamic mesh field then optimizes articulation using bidirectional Vertex-wise Motion Consistency on transported mesh vertices and Pixel-wise Motion Consistency on rendered RGB-D observations. We introduce Articulate-100, a new benchmark of 100 articulated objects spanning 16 PartNet-Mobility categories. On this benchmark, ArtMesh outperforms prior 3DGS-based pipelines in joint parameter estimation and part-level geometric reconstruction, with the largest gains on objects with many movable parts.

Geometry Representation and Mesh Quality Comparison

Left: native reconstructed surfaces. ArtGS and GaussianArt produce unstructured Gaussians with gaps and fragmented coverage; ArtMesh outputs a connected triangle mesh end-to-end, with no conversion step. Right: the same comparison after TSDF-based post-processing on every method's output — ArtMesh still recovers cleaner geometry.

Geometry Representation

ArtMesh (ours)
ArtGS
GaussianArt
ArtMesh — Oven_7187 ArtGS — Oven_7187 GaussianArt — Oven_7187 ArtMesh — StorageFurniture_45612 ArtGS — StorageFurniture_45612 GaussianArt — StorageFurniture_45612 ArtMesh — TrashCan_102218 ArtGS — TrashCan_102218 GaussianArt — TrashCan_102218

TSDF Post-processed

ArtMesh (ours)
ArtGS
GaussianArt
ArtMesh + TSDF — Oven_7187 ArtGS + TSDF — Oven_7187 GaussianArt + TSDF — Oven_7187 ArtMesh + TSDF — StorageFurniture_45612 ArtGS + TSDF — StorageFurniture_45612 GaussianArt + TSDF — StorageFurniture_45612 ArtMesh + TSDF — TrashCan_102218 ArtGS + TSDF — TrashCan_102218 GaussianArt + TSDF — TrashCan_102218

Results

Reconstructed articulated objects from the Articulate-100 benchmark; each video shows the start→end state interpolation followed by a 360° orbit.

Eyeglasses 101293
Oven 7187
Pliers 100180
StorageFurniture 45612
StorageFurniture 45940
Table 22301
Table 23372
Table 23724
Window 103044

Export into Simulator

Based on the part-based reconstruction, ArtMesh's reconstructed object can be directly transformed to URDF format and exported into existing simulators, enabling further embodied-AI applications, e.g. humanoid interactions. The viewers below load the exported URDFs in your browser — rotate (left-click drag), zoom (scroll), pan (right-click drag), and drag any movable part to actuate its joint.

Eyeglasses 101293
Oven 7187
Pliers 100180
StorageFurniture 45612
StorageFurniture 45940
Table 22301
Table 23372
Table 23724
Window 103044

Articulate-100 Benchmark

Articulate-100 is a new benchmark of 100 articulated objects spanning 16 PartNet-Mobility categories. Each card shows one object: a grid of train views at the start and end articulation states (left), and the canonical-pose render of the textured mesh (right). Use the arrows to scroll through. The full dataset is available on Hugging Face.