Research map / Multi-agent systems and games / Human-AI and goal-conditioned
WorldLens: Full-Spectrum Evaluations of Driving World Models in Real World
Ao Liang, Lingdong Kong, Tianyi Yan, Hongsi Liu, Yu Yang, Ziqi Huang, Wei Yin, Jialong Zuo, Yixuan Hu, Dekai Zhu, Dongyue Lu, Youquan Liu and 10 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 |
|---|---|---|---|
| multi-dataset training (OpenDWM) vs single-dataset models | driving world models · WorldLens Generation | Subject Fidelity: 36.30 vs 28.49 (MagicDrive) | Improved |
| multi-domain large-scale training | OpenDWM · WorldLens Downstream | 3D Detection NDS: 21.96% vs 33.22% (DiST-4D) | Hurt |
| RGB-D generation / geometry-aware supervision | DiST-4D · WorldLens Reconstruction | Photometric Error: 0.140 (MagicDrive) -> 0.066 | Improved |
| temporal conditioning with ground-truth frames | DiST-4D, DriveDreamer-2 · WorldLens Downstream | Map Seg / 3D Det: 35.55% and 33.62% vs 18.34% (MagicDrive) map seg | Improved |
| closed-loop vs open-loop evaluation of the same world models | all world models · WorldLens Action-Following | Route Completion / PDMS: e.g. RLGF 78.45% open-loop PDMS vs 13.51% route completion | Hurt |
| lowest perceptual discrepancy does not imply best fidelity | DiST-4D · WorldLens Generation | FVD vs Subject Fidelity: FVD 58.08 best; Subject Fidelity 30.32 vs OpenDWM 36.30 | Mixed |
| strong perceptual quality vs downstream usability | OpenDWM · WorldLens Downstream | 3D Detection NDS: 21.96% vs 33.22% | Hurt |
Closest papers by meaning
- 4DWorldBench: A Comprehensive Evaluation Framework for 3D/4D World Generation Models (CVPR 2026 · similarity 0.90)
- WorldSimBench: Towards Video Generation Models as World Simulators (ICML 2025 · similarity 0.90)
- WorldModelBench: Judging Video Generation Models As World Models (NeurIPS 2025 · similarity 0.89)
- World Models as Group Actions (NeurIPS 2026 · similarity 0.88)
- Learning 3D Persistent Embodied World Models (NeurIPS 2025 · similarity 0.88)
- SyncWorld: Visual Calibration Enables World Models as Zero-Shot Simulators (NeurIPS 2026 · similarity 0.88)
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.
