Researchers have introduced FAMOS, a feed-forward framework designed to model 3D articulated objects from sparse, unordered sets of partial point clouds. Unlike traditional methods that depend on category-level shape priors from single views, FAMOS jointly reasons over multiple observations to predict movable-part segmentation and joint parameters. To effectively aggregate articulation cues across varying numbers of input views, the authors developed a Multi-state Articulation Transformer that utilizes alternating state-wise and global attention mechanisms. A key innovation is the "observed articulation span objective," a loss function that supervises the motion range of each part across the provided observations, forcing the model to utilize the full temporal or spatial context of the input set. To address data scarcity, the team implemented a procedural data generator to synthesize self-annotated assets for training. Empirical evaluations conducted on the PartNet-Mobility, ACD, and ArtiCraft-10K datasets show that FAMOS achieves consistent performance improvements over existing optimization-based and feed-forward baselines, demonstrating its ability to reconstruct complex articulation from limited monocular data.