# X-EraAI/ActPhysCause-Challenge

Repository: https://github.com/X-EraAI/ActPhysCause-Challenge
Canonical: https://ross.abutalabs.com/products/actphyscause-challenge
License Family: other
Last push: 2026-07-16T09:23:58+00:00

## Health v2 (maintenance only)
Score: 54/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 92, release rhythm 35, longevity 3
- inputs: {"age_days": 48, "days_push": 48, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, young, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1004, forks 2 (observed 2026-09-03T02:15:08.598092+00:00)

## What it is
ActPhysCause Challenge is a benchmark and dataset for action-conditioned physical and causal world modeling, hosted as Track 0 of the LoViF 2026 @ ECCV Workshop challenge. It provides 2,000 training episodes and 500 evaluation samples of dual-arm tabletop manipulation, where models must generate physically plausible future videos from a first frame and a natural-language task instruction.

## Use cases
- generate future videos of robot manipulation from a first frame and instruction
- benchmark world models on physical plausibility and goal alignment
- train models on bimanual tabletop manipulation episodes
- evaluate action-conditioned video prediction for embodied AI
- compare submissions on a ranked leaderboard for future video generation
- study causal and physical understanding in video generation models

## When to choose
- you need a standardized benchmark for action-conditioned future video generation
- you are entering the LoViF 2026 @ ECCV Workshop challenge
- you want training data for dual-arm tabletop manipulation video prediction
- you need to evaluate world models on physical plausibility and temporal coherence

## When to avoid
- you need real-time robot control or policy execution rather than video prediction
- you need single-arm or non-tabletop manipulation scenarios
- you need evaluation annotations, which are hidden for the anonymized test set
- you need tracks beyond Track 0, which are reserved for future releases

## Facets
- artifact type: dataset
- maturity: active
- function: machine-learning, video-processing, benchmarking, data-generation
- domain: machine-learning, computer-vision, robotics, artificial-intelligence, simulation
- platform: python, cross-platform
- tags: video-prediction, world-modeling, embodied-ai, benchmark, robot-manipulation, bimanual-manipulation, challenge, eccv-workshop, huggingface-dataset, causal-reasoning, gpu

## Member repositories
- X-EraAI/ActPhysCause-Challenge (main) score 54

## Provenance
- Observed fields: from GitHub, fetched 2026-09-03T02:15:08.598092+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T07:12:44.935154+00:00, confidence not recorded.
  - readme: https://github.com/X-EraAI/ActPhysCause-Challenge (fetched 2026-09-03T02:15:08.598092+00:00, sha a3b7ebd2b0ed)
- Data as of 2026-08-30T08:39:29.467469+00:00.