HCLSM : Hierarchical Causal Latent State Machines for Object-Centric World Modeling

Jaber Jaber, Osama Jaber · 2026

Read HCLSM : Hierarchical Causal Latent State Machines for Object-Centric World Modeling by Jaber Jaber, Osama Jaber (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-29090.txt.

Description

Jaber and Jaber present HCLSM, a world model that predicts future states from video. The model splits scenes into objects with slot attention. Three stacked engines handle smooth physics, discrete events, and abstract goals. A graph network learns how objects affect each other. Training first forces slots to specialize, then teaches prediction. Trained on a robot pushing task, the 68-million-parameter model reports low prediction error. A custom kernel speeds one scan step up 38-fold.

Catalog record

CREATORJaber Jaber, Osama Jaber
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE26035
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.29090

Browse the full library · About the Tower of Babel