Research map / Multi-agent systems and games / LLM agents
A Physics-Grounded Benchmark for Multi-Agent Dynamics in World Models
Nuo Chen, Lulin Liu, Zihao Li, Ziyao Zeng, Zihao Zhu, Wenyan Cong, Junyuan Hong, Yunhao Yang, Zhengzhong Tu, Yan Wang, Boris Ivanovic, Marco Pavone and 3 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| post-train on CrashTwin with physical consistency objectives | Cosmos-Predict2-2B · CrashTwin-Eval | momentum residual J_p: 0.8890 -> 0.6479 | Improved |
| post-train on CrashTwin with physical consistency objectives | Cosmos-Predict2-2B · CrashTwin-Eval | energy gain J_E: 0.8590 -> 0.5534 | Improved |
| post-train on CrashTwin with physical consistency objectives | Cosmos-Predict2-2B · CrashTwin-Eval | instance stability: 0.6129 -> 0.7746 | Improved |
| post-train on CrashTwin with physical consistency objectives | Cosmos-Predict2-2B · CrashTwin-Eval | warp error: 0.0240 -> 0.0085 | Improved |
| scale Cosmos-Predict2 from 2B to 14B | Cosmos-Predict2 · CrashTwin-Eval | momentum residual J_p: 0.8890 -> 0.6828 | Improved |
| scale Cosmos-Predict2 from 2B to 14B | Cosmos-Predict2 · CrashTwin-Eval | flow divergence: 0.6748 -> 0.7180 | Hurt |
| add instance relinking to basic 3D tracking (ablation) | reconstruction pipeline · simulated crashes scene-level SE(3) ATE | ATE (m): 11.71 -> 5.96 | Improved |
| add Kalman filtering (ablation) | reconstruction pipeline · simulated crashes instance-level SE(3) ATE | ATE (m): 3.73 -> 3.29 | Improved |
| add metric depth correction (ablation) | reconstruction pipeline · simulated crashes instance-level Sim(3) ATE | ATE (m): 1.47 -> 0.91 | Improved |
| add metric depth correction (ablation) | reconstruction pipeline · simulated crashes scene-level Sim(3) ATE | ATE (m): 4.68 -> 3.70 | Improved |
| subtract friction impulse from momentum residual (ablation) | evaluator · CrashTwin-Eval | J_p: 0.6828 -> 0.5661 | No clear effect |
| lengthen impact window from 3 to 7 frames (ablation) | evaluator · CrashTwin-Eval | J_p (Cosmos-Predict2-14B): 0.6605 -> 0.7089 | No clear effect |
Closest papers by meaning
- stable-worldmodel: A Platform for Reproducible World Modeling Research and Evaluation (NeurIPS 2026 · similarity 0.87)
- WorldModelBench: Judging Video Generation Models As World Models (NeurIPS 2025 · similarity 0.85)
- Agent2World: Learning to Generate Symbolic World Models via Adaptive Multi-Agent Feedback (NeurIPS 2026 · similarity 0.85)
- Masked Diffusion Language Models are Strong and Steerable Text-Based World Models for Agentic RL (NeurIPS 2026 · similarity 0.85)
- WorldCoder-Bench: Benchmarking Physically Grounded 3D World Synthesis (NeurIPS 2026 · similarity 0.84)
- Dyn-O: Building Structured World Models with Object-Centric Representations (NeurIPS 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
