arXiv：cs.AI· Shuai Shao, Jinchen He, Cunxin Fan, Zihan Guo, Shijian Wang, Qihan Ren, Kangning Zhang, Zunnan Xu, Haoyuan Li, Hao Wang, Jianghao Lin, Weiwen Liu, Weinan Zhang·· 6 小时前
PhysCodeGym：面向物理可验证 4D 重建的编程智能体基准
PhysCodeGym: Benchmarking Coding Agents for Physically Verifiable 4D Reconstruction
AI 导读
PhysCodeGym 是一个面向物理可验证 4D 重建的编程智能体基准，包含 156 个任务，覆盖刚体、液体、可变形体和风力驱动运动。智能体需自主查看视频、编写并运行 Blender 程序，并依据模拟反馈修改场景，评估采用 6 项结合规则测量与评分量表判断的指标。实验显示前沿编程智能体在推理投入设置下存在产物质量与完成度之间的权衡，反事实测试还暴露出通过常规物理检查的重建中潜藏的接触与支撑失效。
正文
Authors:Shuai Shao, Jinchen He, Cunxin Fan, Zihan Guo, Shijian Wang, Qihan Ren, Kangning Zhang, Zunnan Xu, Haoyuan Li, Hao Wang, Jianghao Lin, Weiwen Liu, Weinan Zhang
Abstract:Building physical simulations from video requires coding agents to translate observations into executable models and refine them through tool use. We introduce PhysCodeGym, a benchmark for physically verifiable 4D reconstruction with $156$ tasks covering rigid-body, liquid, deformable, and wind-driven motion. Agents autonomously inspect videos, write and run Blender programs, and revise their scenes using simulation feedback. Six metrics combine rule-based measurements with rubric-based judgments to assess visual and physical reconstruction quality. Program inspection, scene measurements, and counterfactual interventions test physical consistency under original and modified conditions. Experiments with frontier coding agents reveal a tradeoff between artifact quality and completion across reasoning-effort settings, while harness comparisons and agent traces link verification behavior with reconstruction performance. Case studies show that matching visible objects and events can coexist with physical defects. Counterfactual tests further expose latent contact and support failures in reconstructions that pass ordinary physical checks, highlighting the gap between reproducing observations and building reliable physical models. Code is available at this https URL.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.17288 [cs.AI] |
| (or arXiv:2610.17288v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.17288 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Shuai Shao [view email]
[v1]
Mon, 12 Oct 2026 18:44:55 UTC (33,237 KB)
来源：arXiv：cs.AI · arxiv.org