"Code as Worlds: Agentic Discovery of Executable World Representations for Physical Reasoning" by Hanyang Wang , Yimo Cai , Weiliang Chen , Jiawei Chi , Haowen Sun , Qiyu Dai , Yi-Hsin Hung , Xingzhuo Guo , Jinshan Ren , Runmao Yao , Ziwei Liu , Mingsheng Long , Yueqi Duan , Jun Gao , Jiangran Lyu , Fangfu Liu , Jialong Wu
TLDR:
The text discusses Code-as-World, a paradigm that represents physical environments as executable code to facilitate quantitative reasoning and scalable supervision for vision-language models. Modern vision-language models excel at recognizing and explaining physical events but often lack explicit representations of underlying mechanisms essential for understanding how the world evolves and responds to interventions. Code-as-World introduces an approach that expresses physical composition, dynamics, and visual appearance as executable code to provide a compact and grounded abstraction of the physical world. The method involves constructing representations through an agentic discovery loop inspired by abductive reasoning, allowing for the generation of executable world hypotheses from multimodal observations. By leveraging these verified executable worlds, vision-language models can be trained effectively on quantitative physical reasoning, leading to state-of-the-art performance and highlighting the potential of executable world representations for advancing physical intelligence.
Read Paper / Blog
huggingface.co
Paper page - Code as Worlds: Agentic Discovery of Executable World Representations for Physical Reasoning
Join the discussion on this paper page
1September 1, 2026 48
