SAMoR: Motion Modelling for Articulated Objects of Any Skeleton and Topology

2026-07-02 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Computer Vision · Depth: Expert, quick

Summary

SAMoR (Skeleton-Aware Motion Representation for Articulated Objects) is a novel cross-topology motion representation addressing the challenge of modeling motion for articulated objects with arbitrary skeleton topologies, a limitation of existing fixed-skeleton generators. It leverages the observation that functional joint groups share motion structure across species despite differing joint counts. SAMoR encodes each motion segment as a fixed number ($K=8$) of part tokens, shared across diverse skeletons. Its architecture uses a graph-transformer encoder, processing per-joint motion features, kinematic graph structure, and joint-name embeddings, compressing them into part-level tokens via cross-attention pooling and residual vector quantization. A topology-agnostic attention supervision loss and joint-name dropout prevent token collapse. Evaluated on HumanML3D, Truebones Zoo, and animated Objaverse-XL, SAMoR achieves accurate reconstruction and cross-topology transfer, reaching \$2.75 \times 10^{-2}$ normalized MPJPE on cross-topology reconstruction, \$5.8\times$ below the strongest adapted variable-$J$ tokenizer baseline. It also enables text-conditioned generation and part-wise editing.

Key takeaway

For animation developers or researchers working with diverse articulated models, SAMoR offers a robust solution for motion generation and transfer across arbitrary skeleton topologies. You can now achieve high-fidelity motion reconstruction and cross-topology transfer, significantly outperforming variable-J tokenizers. Consider integrating SAMoR's approach for text-conditioned generation and part-wise editing, especially when dealing with heterogeneous character rigs, to streamline your animation workflows and expand creative possibilities.

Key insights

SAMoR enables universal motion modeling for articulated objects by representing motion segments as shared, fixed-size part tokens.

Principles

• Functional joint groups share motion structure across species.
• Joint names partially expose motion correspondence.
• Topology-agnostic attention prevents token collapse.

Method

A graph-transformer encoder compresses per-joint features, kinematic structure, and joint-name embeddings into $K=8$ part tokens via cross-attention pooling and residual vector quantization.

In practice

• Reconstruct motion for unseen skeletons.
• Transfer motion across different topologies.
• Edit motion part-wise using MaskGIT.

Topics

• Motion Modelling
• Articulated Objects
• Graph Transformers
• Cross-Topology Transfer
• Motion Generation
• Residual Vector Quantization

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Computer Vision Engineer

Related on AIssential

• Generative Data Augmentation for Skeleton Action Recognition — A conditional generative pipeline synthesizes diverse skeleton data to augment action recognition, improving performance in low-data settings.
• Segment Anything with Motion, Geometry, and Semantic Adaptation for Complex Nonlinear Visual Object Tracking — SAMOSA adapts SAM 2 for robust visual object tracking by integrating explicit motion, geometry, and semantic cues.
• MoVerse: Real-Time Video World Modeling with Panoramic Gaussian Scaffold — MoVerse creates interactive 3D worlds from single images by separating world construction from observation rendering using panoramic diffusion and Gaussian scaffolds.
• 🍧10,000× faster SAM-3D🍧 👉Fast SAM 3D Body achieves up to 10.9× speedup, over 10,000×... — A novel method, "Fast SAM 3D Body," significantly accelerates 3D body reconstruction, achieving an impressive 10.9× speedup and over 10,000× faster MHR-to-SMPL conversion.
• Sketch2Arti: Sketch-based Articulation Modeling of CAD Objects — Sketch2Arti enables intuitive, precise articulation modeling and internal geometry completion for CAD objects using simple 2D user sketches.
• SAM 3D - AI at Meta — Meta's SAM 3D models reconstruct detailed 3D objects and human bodies from single 2D images.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.