RoboStressBench: Benchmarking VLM Robustness to Physical Visual Stress in Embodied Scenes
Leyi Wu, Yifan Zhao, Jinjie Zhang, Suzeyu Chen, Wosong Chen, Zhifei Chen, Tianshuo Xu, Qingchun He, Hongxin Hu, Haojian Huang, Yangkai Wei, Wenqian Li, Yinchuan Li, Ying-Cong Chen · 2026
Read RoboStressBench: Benchmarking VLM Robustness to Physical Visual Stress in Embodied Scenes by Leyi Wu, Yifan Zhao, Jinjie Zhang, Suzeyu Chen, Wosong Chen, Zhifei Chen, Tianshuo Xu, Qingchun He, Hongxin Hu, Haojian Huang, Yangkai Wei, Wenqian Li, Yinchuan Li, Ying-Cong Chen (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-2606-00828.txt.
Description
Wu and co-authors build a stress test for vision-language models in this 2026 paper. Such models now guide robots and other embodied systems. Past tests use clean images or one tweak at a time. Real scenes, though, bend under physical forces as they form. The team asks how visual stress should be defined in a principled way.
Their benchmark, RoboStressBench, grows from the physics of rendering. It splits stress into four axes: material, viewpoint, lighting, and geometry. That split covers a broad range of real-world strain in embodied scenes. It also lets each factor be varied with control. The team runs top vision-language models through the set. They find stress-specific failure modes: unlike factors break unlike skills. Broad scores had hidden those splits. The work gives builders a finer map of where a model's sight fails under physical strain.
Catalog record
| CREATOR | Leyi Wu, Yifan Zhao, Jinjie Zhang, Suzeyu Chen, Wosong Chen, Zhifei Chen, Tianshuo Xu, Qingchun He, Hongxin Hu, Haojian Huang, Yangkai Wei, Wenqian Li, Yinchuan Li, Ying-Cong Chen |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 55 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.00828