SAGE: Scalable Agentic 3D Scene Generation for Embodied AI
Hongchi Xia Xuan Li Zhaoshuo Li Qianli Ma Jiashu Xu Ming-Yu Liu Yin Cui, Tsung-Yi Lin Wei-Chiu Ma Shenlong Wang Shuran Song Fangyin Wei · 2026
Read SAGE: Scalable Agentic 3D Scene Generation for Embodied AI by Hongchi Xia Xuan Li Zhaoshuo Li Qianli Ma Jiashu Xu Ming-Yu Liu Yin Cui, Tsung-Yi Lin Wei-Chiu Ma Shenlong Wang Shuran Song Fangyin Wei (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-2602-10116.txt.
Description
Collecting real-world data for embodied agents is costly and sometimes unsafe, yet generated scenes often break physics or look artificial. A 2026 paper led by Hongchi Xia, with eleven co-authors, presents SAGE, an agentic framework that builds simulator-ready 3D environments from a user's task description. One example task is picking up a bowl and placing it on a table. Generators for layout and objects pair with critics judging meaning, looks, and stability.
Through iterative reasoning and adaptive tool choice, the agent refines scenes until they match intent and obey physics. The resulting environments are lifelike, varied, and directly usable in modern simulators. Policies trained purely on this data scale cleanly and generalize to unseen objects and layouts.
Catalog record
| CREATOR | Hongchi Xia Xuan Li Zhaoshuo Li Qianli Ma Jiashu Xu Ming-Yu Liu Yin Cui, Tsung-Yi Lin Wei-Chiu Ma Shenlong Wang Shuran Song Fangyin Wei |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 55 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-10116