# yuantianyuan01/FastWAM

Official codebase for Fast-WAM: Do World Action Models Need Test-time Future Imagination?

Repository: https://github.com/yuantianyuan01/FastWAM
Canonical: https://ross.abutalabs.com/products/fastwam
Homepage: https://yuantianyuan01.github.io/FastWAM/
Language: Python
License: NOASSERTION
License Family: other
Last push: 2026-08-20T06:49:41+00:00

## Health v2 (maintenance only)
Score: 59/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 98, release rhythm 35, longevity 12
- inputs: {"age_days": 169, "days_push": 13, "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 1362, forks 173 (observed 2026-08-28T04:04:29.925824+00:00)

## What it is
Official PyTorch codebase for Fast-WAM, a World Action Model for robot manipulation that skips test-time future video imagination and generates actions directly from latent world representations. It includes training and evaluation code for LIBERO and RoboTwin benchmarks with LeRobot 2.1/3.0 dataset support.

## Use cases
- train a world action model for robot manipulation
- evaluate VLA-style policies on LIBERO benchmarks
- run fast real-time robot policy inference without future video generation
- train on LeRobot 3.0 robot datasets
- reproduce Fast-WAM paper results on RoboTwin
- compare imagine-then-execute vs direct action generation WAMs

## When to choose
- you need a fast real-time world action model for embodied control
- you want to train or evaluate policies on LIBERO or RoboTwin
- you need LeRobot dataset support for robot learning experiments

## When to avoid
- you need a general-purpose video generation model
- you need a production robot control stack rather than research code
- you work outside GPU-equipped Linux environments

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning, llm-training, simulation
- domain: robotics, machine-learning, artificial-intelligence, autonomous-vehicles
- platform: python
- tags: world-action-model, vla, robot-learning, video-modeling, libero, robotwin, lerobot, research-code, gpu, linux

## Member repositories
- yuantianyuan01/FastWAM (main) score 59

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:29.925824+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 2026-08-30T04:41:32.669932+00:00, confidence not recorded.
  - readme: https://github.com/yuantianyuan01/FastWAM (fetched 2026-08-28T04:04:29.925824+00:00, sha 45b140c5e26a)
  - homepage: https://yuantianyuan01.github.io/FastWAM/ (fetched 2026-08-29T11:59:13.441097+00:00, sha d05438f75b65)
- Data as of 2026-08-30T08:39:29.467469+00:00.