FAMOS: Feed-Forward 3D Articulation Modeling from Sparse Observations

FAMOS is a feed-forward model that predicts movable-part segmentation and joint parameters from a sparse, unordered set of partial point clouds, outperforming feed-forward and optimization-based baselines on PartNet-Mobility, ACD, and ArtiCraft-10K.

FAMOS predicts articulated-object segmentation and joint parameters from a sparse, unordered set of partial monocular point clouds, jointly reasoning across a variable number of observations, including a single view. It introduces a Multi-state Articulation Transformer with alternating state-wise and global attention plus an observed articulation span objective that supervises the motion ranges exhibited across inputs. To address the limited scale and diversity of articulation datasets, the authors built a procedural generator that synthesizes self-annotated training assets. Experiments on PartNet-Mobility, ACD, and ArtiCraft-10K show consistent improvements over both feed-forward and optimization-based baselines. Both reports (Hugging Face daily papers, 2026-09-16, and arXiv, 2026-09-17) are consistent, with the arXiv version adding the detail that inputs are partial monocular point clouds and clarifying the role of the articulation span objective.

• FAMOS is a feed-forward model predicting movable-part segmentation and joint parameters from a sparse, unordered set of partial monocular point clouds, supporting a variable number of inputs including a single view.
• It uses a Multi-state Articulation Transformer with alternating state-wise and global attention across observations.
• An observed articulation span objective supervises the motion ranges exhibited across inputs.
• A procedural generator synthesizes self-annotated training assets to address the limited scale and diversity of articulation datasets.
• Experiments on PartNet-Mobility, ACD, and ArtiCraft-10K show consistent improvements over both feed-forward and optimization-based baselines.
• Sources: Hugging Face daily papers (2026-09-16) and arXiv cs.AI/cs.LG/cs.CL (2026-09-17); no disagreements between reports.