IntPhys 2

How to read this picture

Tests whether video models can tell physically possible from impossible events in Unreal Engine videos.1

Sources
Last checked 10 Oct 2026Basic entry15 of 21 facts checked at the sourceNext check 8 Apr 2027
Runs in
Recorded data1
Checked against real robots
Not checked
Skill
Reasoning
Robot
No body1
Used by
482
citations
Licence
Unclear3

Comparisons with real robots

No comparison found Unknown

Details

Not built to predict robot performance. Searched paper, Meta blog, repo.

Known problems 1

  1. Main set released with metadata; authors add a held-out set without labels to detect conta

    contamination: Main set released with metadata; authors add a held-out set without labels to detect contamination.1

Details

About

What it is
Benchmark Inferred4
More

Classified by the Atlas from how the authors describe and distribute it.

Built by
All six authors: FAIR at Meta1
Released
2025-06 (arXiv v1 2025-06-11; Meta blog 2025-06-11)4
More

HF dataset object created 2025-05-13 (API) but announced 2025-06-11.

Version
arXiv v1 only; no release tags4
Last update
Repo commits 2025-10-21: 'Update README.md with Held-Out set instructions', dataloader and transforms fixes.5
More

GitHub API.

Status
Maintained Inferred5
More

Last commit 2025-10; leaderboard currently broken.

Setup

Runs in
Recorded data Inferred1
More

Classification on rendered videos; no control loop.

Robot
No body1
Setting
Mixed1
Size
Debug 5 scenes / 60 videos; Main 253 scenes / 1,012 videos; Held-out 86 scenes / 344 videos.5
More

Same table in paper and HF card.

Scoring and access

Scored by
Accuracy1
Leaderboard
Official leaderboard6
Code licence
custom: CC-BY-NC-4.0 with added restriction (evaluation only; generative AI uses prohibited)3
More

LICENSE.md text; GitHub reports 'Other'.

Data licence
custom: CC-BY-NC-4.0 with the same restriction; Unreal Engine / Epic content subject to Epic's terms7
More

HF card metadata cc-by-nc-4.0; restriction in README and LICENSE.md; Epic note in paper footnote.

Published at
Not published at a venue found. OpenReview lists it as 'Submitted to NeurIPS 2025 Datasets and Benchmarks Track' with venueid ending 'Rejected_Submission'.8
More

OpenReview API search result; the forum page itself was behind a browser check.

Sources 8

  1. 1IntPhys 2: Benchmarking Intuitive Physics Understanding In Complex Synthetic Environments (full text)Paper · Jun 2025 · checked 10 Oct 2026
  2. 2Semantic Scholar API recordIndex · checked 10 Oct 2026
  3. 3facebookresearch/IntPhys2 on GitHub (file LICENSE.md)Repository · checked 10 Oct 2026
  4. 4IntPhys 2: Benchmarking Intuitive Physics Understanding In Complex Synthetic EnvironmentsPaper · Jun 2025 · checked 10 Oct 2026
  5. 5facebookresearch/IntPhys2 on GitHub (repository)Repository · checked 10 Oct 2026
  6. 6facebook/physical_reasoning_leaderboard on Hugging Face (space)Leaderboard · checked 10 Oct 2026
  7. 7facebook/IntPhys2 on Hugging Face (dataset)Repository · checked 10 Oct 2026
  8. 8https://api2.openreview.net/notes/search?term=IntPhys%202%20intuitive%20physics%20complex%20syntheticIndex · checked 10 Oct 2026

Change history

  1. Created as a basic entry: identity facts checked at primary sources (phase 1 re-verification).