# What-If World: A Causal Benchmark for General World Models in Embodied Scenarios

arXiv: https://arxiv.org/abs/2605.27589 · HTML: https://mlsanity.com/paper/2605.27589/

## Key facts

- Cited by 8 papers in the map (data through 14 Oct 2026).
- Placed in Latent predictive world models, one of 11 directions in World models.
- Builds on 69 of its 83 references that are in the map.
- First posted on arXiv 26 May 2026.

## Placed in

- World models / [Latent predictive world models](https://mlsanity.com/wm/latent-predictive-world-models/)

## Builds on (most cited first, 30 of 69)

- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [nuScenes: A multimodal dataset for autonomous driving](https://mlsanity.com/paper/1903.11027/) (2019)
- [OpenVLA: An Open-Source Vision-Language-Action Model](https://mlsanity.com/paper/2406.09246/) (2024)
- [Wan: Open and Advanced Large-Scale Video Generative Models](https://mlsanity.com/paper/2503.20314/) (2025)
- [InstructPix2Pix: Learning to Follow Image Editing Instructions](https://mlsanity.com/paper/2211.09800/) (2022)
- [G-Eval: NLG Evaluation using GPT-4 with Better Human Alignment](https://mlsanity.com/paper/2303.16634/) (2023)
- [CogVideoX: Text-to-Video Diffusion Models with An Expert Transformer](https://mlsanity.com/paper/2408.06072/) (2024)
- [Shortcut Learning in Deep Neural Networks](https://mlsanity.com/paper/2004.07780/) (2020)
- [VBench: Comprehensive Benchmark Suite for Video Generative Models](https://mlsanity.com/paper/2311.17982/) (2023)
- [Mastering Diverse Domains through World Models](https://mlsanity.com/paper/2301.04104/) (2023)
- [Large Language Models are not Fair Evaluators](https://mlsanity.com/paper/2305.17926/) (2023)
- [WorldSimBench: Towards Video Generation Models as World Simulators](https://mlsanity.com/paper/2410.18072/) (2024)
- [Open X-Embodiment: Robotic Learning Datasets and RT-X Models](https://mlsanity.com/paper/2310.08864/) (2023)
- [DROID: A Large-Scale In-The-Wild Robot Manipulation Dataset](https://mlsanity.com/paper/2403.12945/) (2024)
- [Genie: Generative Interactive Environments](https://mlsanity.com/paper/2402.15391/) (2024)
- [NExT-QA:Next Phase of Question-Answering to Explaining Temporal Actions](https://mlsanity.com/paper/2105.08276/) (2021)
- [GAIA-1: A Generative World Model for Autonomous Driving](https://mlsanity.com/paper/2309.17080/) (2023)
- [MagicBrush: A Manually Annotated Dataset for Instruction-Guided Image Editing](https://mlsanity.com/paper/2306.10012/) (2023)
- [Revisiting Feature Prediction for Learning Visual Representations from Video](https://mlsanity.com/paper/2404.08471/) (2024)
- [Winoground: Probing Vision and Language Models for Visio-Linguistic Compositionality](https://mlsanity.com/paper/2204.03162/) (2022)
- [CLEVRER: CoLlision Events for Video REpresentation and Reasoning](https://mlsanity.com/paper/1910.01442/) (2019)
- [Vista: A Generalizable Driving World Model with High Fidelity and Versatile Controllability](https://mlsanity.com/paper/2405.17398/) (2024)
- [Video PreTraining (VPT): Learning to Act by Watching Unlabeled Online Videos](https://mlsanity.com/paper/2206.11795/) (2022)
- [EvalCrafter: Benchmarking and Evaluating Large Video Generation Models](https://mlsanity.com/paper/2310.11440/) (2023)
- [Don't Just Assume; Look and Answer: Overcoming Priors for Visual Question Answering](https://mlsanity.com/paper/1712.00377/) (2017)
- [VBench-2.0: Advancing Video Generation Benchmark Suite for Intrinsic Faithfulness](https://mlsanity.com/paper/2503.21755/) (2025)
- [Diffusion for World Modeling: Visual Details Matter in Atari](https://mlsanity.com/paper/2405.12399/) (2024)
- [UniSim: A Neural Closed-Loop Sensor Simulator](https://mlsanity.com/paper/2308.01898/) (2023)
- [Counterfactual VQA: A Cause-Effect Look at Language Bias](https://mlsanity.com/paper/2006.04315/) (2020)
- [VideoPhy: Evaluating Physical Commonsense for Video Generation](https://mlsanity.com/paper/2406.03520/) (2024)

## Built on by (most cited first, 8 of 8)

- [Twin Rollouts: Noise-Coupled Counterfactual Branching in Interactive Video World Models](https://mlsanity.com/paper/2608.08982/) (2026)
- [YoCausal: How Far is Video Generation from World Model? A Causality Perspective](https://mlsanity.com/paper/2605.30346/) (2026)
- [WorldSimProbe: Diagnosing Simulator Faithfulness in Action-Conditioned World Models for Embodied Manipulation](https://mlsanity.com/paper/2608.09298/) (2026)
- [Why Does the Future Branch? Identifiable Closure Tests for Stochastic Physical World Models](https://mlsanity.com/paper/2608.00591/) (2026)
- [From Generation to Simulation: How Far Are World Models from Being True Simulators?](https://mlsanity.com/paper/2608.23070/) (2026)
- [VehDyn: A Driving World Model Benchmark for Vehicle Dynamics](https://mlsanity.com/paper/2609.33264/) (2026)
- [When Generative Models Act as Simulators](https://mlsanity.com/paper/2610.12847/) (2026)
- [PhysCodeGym: Benchmarking Coding Agents for Physically Verifiable 4D Reconstruction](https://mlsanity.com/paper/2610.17288/) (2026)

## Abstract (arXiv)

Video generation models are increasingly used as world simulators for tasks like driving and robotic manipulation. What matters in these settings is not whether a single video looks right, but whether the model's output changes when its input changes. We test this by giving a model two prompts describing the same scene with one physical detail varied, and checking whether the two videos diverge the way physics predicts. The wording difference between the prompts is small by design, since only one variable is changed, but the correct physical difference is not. A model that misses this can still produce two videos that each look plausible individually, and existing benchmarks score videos one at a time and cannot detect this failure. We introduce What-If World, 319 such prompt pairs built on real frames from nuScenes and DROID, organized by a taxonomy of six physical variables shared across driving and manipulation. Each pair is scored with APEO, a four-part rubric checking whether each video follows its prompt (Adherence), is physically consistent (Physics), preserves the shared scene (Environment), and ends in the correct difference (Outcome). Across nine state-of-the-art models, no system exceeds 52% on the paired score, and open-source models cluster near 28%. Every model tested fails on a large fraction of causal interventions, indicating substantial room before these models can reliably support action-conditioned simulation or model-based planning. Where models do score well, performance appears to track the visual prominence of the intervention rather than the tractability of its underlying physics. Some visually subtle interventions score as low as 14.2%, while visually pronounced ones reach 40.4%.