PhyScensis: Physics-Augmented LLM Agents for Complex Physical Scene Arrangement
Yian Wang 1,2 Han Yang 1 Minghao Guo 3 Xiaowen Qiu 2 Tsun-Hsuan Wang 2 Wojciech Matusik 3 Joshua B. Tenenbaum 3 Chuang Gan 1,4 1 UMass Amherst 2 Genesis AI 3 MIT 4 MIT-IBM Watson AI Lab {yianwang,hanyang,chuanggan}@umass.edu [email protected] {xiaowen.qiu,johnson}@genesis-ai.company {wojciech,jbt} · 2026
Read PhyScensis: Physics-Augmented LLM Agents for Complex Physical Scene Arrangement by Yian Wang 1,2 Han Yang 1 Minghao Guo 3 Xiaowen Qiu 2 Tsun-Hsuan Wang 2 Wojciech Matusik 3 Joshua B. Tenenbaum 3 Chuang Gan 1,4 1 UMass Amherst 2 Genesis AI 3 MIT 4 MIT-IBM Watson AI Lab {yianwang,hanyang,chuanggan}@umass.edu [email protected] {xiaowen.qiu,johnson}@genesis-ai.company {wojciech,jbt} (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-2602-14968.txt.
Description
Wang and colleagues address a gap in automatic 3D scene building for robot training. Most methods place objects but ignore ties such as contact, support, balance, and containment. Their system, PhyScensis, pairs a language-model agent with a solver run by a physics engine. The agent puts forward assets with spatial and physical tags. The solver builds them, and solver feedback drives more rounds.
Chance-based code keeps scenes steady, while a second rule weighs steadiness against spatial ties. The system stays easy to steer through text and numbers such as relative spots. In tests it beats prior methods on scene richness, visual quality, and true-to-life accuracy, giving one pipeline for dense, lifelike scenes for robot hand work like shelves and packed boxes.
Catalog record
| CREATOR | Yian Wang 1,2 Han Yang 1 Minghao Guo 3 Xiaowen Qiu 2 Tsun-Hsuan Wang 2 Wojciech Matusik 3 Joshua B. Tenenbaum 3 Chuang Gan 1,4 1 UMass Amherst 2 Genesis AI 3 MIT 4 MIT-IBM Watson AI Lab {yianwang,hanyang,chuanggan}@umass.edu [email protected] {xiaowen.qiu,johnson}@genesis-ai.company {wojciech,jbt} |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 70 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-14968