mlsanityData to 7 Oct 2026

IntPhys 2: Benchmarking Intuitive Physics Understanding In Complex Synthetic Environments

arXiv 2506.09849 ↗First posted 11 Jun 2025cs.CVcited by 49

Not on the map. Why?

Abstract (arXiv)

We present IntPhys 2, a video benchmark designed to evaluate the intuitive physics understanding of deep learning models. Building on the original IntPhys benchmark, IntPhys 2 focuses on four core principles related to macroscopic objects: Permanence, Immutability, Spatio-Temporal Continuity, and Solidity. These conditions are inspired by research into intuitive physical understanding emerging during early childhood. IntPhys 2 offers a comprehensive suite of tests, based on the violation of expectation framework, that challenge models to differentiate between possible and impossible events within controlled and diverse virtual environments. Alongside the benchmark, we provide performance evaluations of several state-of-the-art models. Our findings indicate that while these models demonstrate basic visual understanding, they face significant challenges in grasping intuitive physics across the four principles in complex scenes, with most models performing at chance levels (50%), in stark contrast to human performance, which achieves near-perfect accuracy. This underscores the gap between current models and human-like intuitive physics understanding, highlighting the need for advancements in model architectures and training methodologies.

Key facts. Cited by 49 papers in the map (data through 7 Oct 2026). Builds on 30 of its 52 references that are in the map. First posted on arXiv 11 Jun 2025.

Builds on

30 of 52 references.

References · 30

  1. 2020Language Models are Few-Shot Learnerscited by 32,199
  2. 2025Qwen2.5-VL Technical Reportcited by 5,418
  3. 2021RoFormer: Enhanced Transformer with Rotary Position Embeddingcited by 4,581
  4. 2019PIQA: Reasoning about Physical Commonsense in Natural Languagecited by 2,818
  5. 2017CARLA: An Open Urban Driving Simulatorcited by 2,431
  6. 2019HuggingFace's Transformers: State-of-the-art Natural Language Processingcited by 2,201
  7. 2018Datasheets for Datasetscited by 1,371
  8. 2018Learning Latent Dynamics for Planning from Pixelscited by 1,221
  9. 2020Learning to Simulate Complex Physics with Graph Networkscited by 867
  10. 2017AI2-THOR: An Interactive 3D Environment for Visual AIcited by 850
  11. 2025Cosmos World Foundation Model Platform for Physical AIcited by 833
  12. 2023VideoMAE V2: Scaling Video Masked Autoencoders with Dual Maskingcited by 516
18 more
  1. 2024Revisiting Feature Prediction for Learning Visual Representations from Videocited by 490
  2. 2019CLEVRER: CoLlision Events for Video REpresentation and Reasoningcited by 475
  3. 2024VideoPhy: Evaluating Physical Commonsense for Video Generationcited by 215
  4. 2024An Introduction to Vision-Language Modelingcited by 175
  5. 2025Do generative video models understand physical principles?cited by 151
  6. 2021Physion: Evaluating Physical Prediction from Vision in Humans and Machinescited by 145
  7. 2019PHYRE: A New Benchmark for Physical Reasoningcited by 137
  8. 2019CATER: A diagnostic dataset for Compositional Actions and TEmporal Reasoningcited by 137
  9. 2019Oops! Predicting Unintentional Action in Videocited by 82
  10. 2025Intuitive physics understanding emerges from self-supervised pretraining on natural videoscited by 66
  11. 2020Visual Grounding of Learned Physical Modelscited by 54
  12. 2023Physion++: Evaluating Physical Scene Understanding that Requires Online Inference of Different Physical Propertiescited by 53
  13. 2023PUG: Photorealistic and Semantically Controllable Synthetic Data for Representation Learningcited by 31
  14. 2023GRASP: A novel benchmark for evaluating language GRounding And Situated Physics understanding in multimodal language modelscited by 28
  15. 2024LLMs Are Biased Towards Output Formats! Systematically Evaluating and Mitigating Output Format Bias of LLMscited by 24
  16. 2023WorldSense: A Synthetic Benchmark for Grounded Reasoning in Large Language Modelscited by 19
  17. 2024Compositional 4D Dynamic Scenes Understanding with Physics Priors for Video Question Answeringcited by 18
  18. 2020Occlusion resistant learning of intuitive physics from videoscited by 7

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 14; 2026: 35.

49 papers.

Sort:

By area:

  1. 2025World Simulation with Video Foundation Models for Physical AIcited by 183
  2. 2025Embodied AI Agents: Modeling the Worldcited by 57
  3. 2026Inference-time Physics Alignment of Video Generative Models with Latent World Modelscited by 32
  4. 2025PAI-Bench: A Comprehensive Benchmark For Physical AIcited by 30
  5. 2025Generative Universal Verifier as Multimodal Meta-Reasonercited by 17
  6. 2025LikePhys: Evaluating Intuitive Physics Understanding in Video Diffusion Models via Likelihood Preferencecited by 14
  7. 2026WorldBench: Benchmarking Physical Understanding of World Models by Isolating Physics Conceptscited by 12
  8. 2025Rethinking JEPA: Compute-Efficient Video SSL with Frozen Teacherscited by 11
  9. 2026PhysInOne: Visual Physics Learning and Reasoning in One Suitecited by 9
  10. 2025Mind Meets Space: Rethinking Agentic Spatial Intelligence from a Neuroscience-inspired Perspectivecited by 7
  11. 2026Code as Worlds: Agentic Discovery of Executable World Representations for Physical Reasoningcited by 6
  12. 2025Inferring Dynamic Physical Properties from Video Foundation Modelscited by 5
37 more
  1. 2025VKnowU: Evaluating Visual Knowledge Understanding in Multimodal LLMscited by 5
  2. 2026Learning Physics-Grounded 4D Dynamics with Neural Gaussian Force Fieldscited by 5
  3. 2026VisPhyWorld: Probing Physical Reasoning via Code-Driven Video Reconstructioncited by 3
  4. 2026CRONOS: Benchmarking Counterfactual Physical Consistency in Video Modelscited by 3
  5. 2026YoCausal: How Far is Video Generation from World Model? A Causality Perspectivecited by 3
  6. 2026TRAPSBench: Vision-Language Models Encode but Fail to Express Epistemic Restraintcited by 3
  7. 2025Pixels to Principles: Probing Intuitive Physics Understanding in Multimodal Language Modelscited by 2
  8. 2025InPhyRe Discovers: Large Multimodal Models Struggle in Inductive Physical Reasoningcited by 2
  9. 2026Next-Gen CAPTCHAs: Leveraging the Cognitive Gap for Scalable and Diverse GUI-Agent Defensecited by 2
  10. 2026GEOPHYS: The Geometry of Physical Plausibilitycited by 2
  11. 2026PhysEditWorld: A Large-Scale Dataset Toward Physics-Editable World Modelscited by 2
  12. 2026PhiZero: A World Model Built Around Physical Languagecited by 2
  13. 2026PhysMind: From Video to Executable Worlds for Training-Free Physical Reasoningcited by 2
  14. 2025PhysFire-WM: A Physics-Informed World Model for Emulating Fire Spread Dynamicscited by 1
  15. 2025Modeling the Mental World for Embodied AI: A Comprehensive Reviewcited by 1
  16. 2026A Machine Learning Framework for Turbofan Health Estimation via Inverse Problem Formulationcited by 1
  17. 2026Tracing the Arrow of Time: Diagnosing Temporal Information Flow in Video-LLMscited by 1
  18. 2026Apple-π: Benchmarking Thinking with Video Towards Law-Grounded Physical Intelligencecited by 1
  19. 2025SRNN: Spatiotemporal Relational Neural Network for Intuitive Physics Understandingcited by 0
  20. 2026SpaRRTa: A Synthetic Benchmark for Evaluating Spatial Intelligence in Visual Foundation Modelscited by 0
  21. 2026Discovering High Level Patterns from Simulation Tracescited by 0
  22. 20263DSPA: A 3D Semantic Point Autoencoder for Evaluating Video Realismcited by 0
  23. 2026Einstein World Modelscited by 0
  24. 2026Event-Conditioned Diagnostics of Kinematic, Contact, and Object-Permanence Structure in Passive Object-State World Modelscited by 0
  25. 2026PhysMRV: Physical Memory Retrieval and Verification for Physics Plausibility Reasoningcited by 0
  26. 2026SiPhy: Single-Image Physical Property Reasoningcited by 0
  27. 2026FADE: From Passive Verification to Active Discovery in Counterfactual Video Understandingcited by 0
  28. 2026What, Where, and How: Probing Spatiotemporal Representations in Video Foundation Modelscited by 0
  29. 2026Principia: Relational Physics Tests for Video Modelscited by 0
  30. 2026PhysAI-Bench: A Benchmark for LLM-Based Agentic Decision-Making in Autonomous UAV-Centric Physical AIcited by 0
  31. 2026Do World Models Make Better Robots? A Survey of Evaluation Benchmarks for Predictive Embodied Intelligencecited by 0
  32. 2026MultiEcho: An Experimental Science of Learned Worldscited by 0
  33. 2026PhyProbe: Rethinking Physical Consistency Evaluation in Generated Videoscited by 0
  34. 2026PhysVista: Benchmarking Physical Intelligence in VLMs via a Perception-Reasoning-Assessment Loopcited by 0
  35. 2026PhysicsLENS: Diagnosing Physical Property Blindness in Video Generation Modelscited by 0
  36. 2026RoboJEPA: Scaling Robotic Latent World Modelscited by 0
  37. 2026WOVEN: Weaving Visual World Modeling into Multimodal LLMscited by 0