# NVIDIA/DreamDojo

Official Codebase for "DreamDojo: A Generalist Robot World Model from Large-Scale Human Videos" (ICML 2026)

Repository: https://github.com/NVIDIA/DreamDojo
Canonical: https://ross.abutalabs.com/products/dreamdojo
Homepage: https://dreamdojo-world.github.io
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2026-03-21T19:03:33+00:00

## Health v2 (maintenance only)
Score: 48/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 73, release rhythm 35, longevity 14
- inputs: {"age_days": 205, "days_push": 165, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1059, forks 75 (observed 2026-08-28T04:03:25.073515+00:00)

## What it is
NVIDIA's official PyTorch codebase for DreamDojo, a generalist robot world model pretrained on 44k hours of human egocentric video and post-trained on robot embodiments for action-conditioned video generation. It includes training, post-training, distillation, and evaluation pipelines plus pretrained 2B/14B checkpoints.

## Use cases
- train a robot world model from human videos
- generate action-conditioned video rollouts for robot policies
- post-train a world model on a new robot embodiment
- distill a video world model for real-time interactive generation
- pretrain on large-scale egocentric video datasets
- evaluate world model generalization across robots and environments

## When to choose
- you need a generalist robot world model with pretrained checkpoints
- you want to post-train a video world model on your own robot data
- you need real-time interactive world model rollouts via distillation
- you are doing research on video-based robot learning

## When to avoid
- you need a lightweight CPU-only tool
- you want a production robot control stack rather than research code
- you lack multi-GPU resources for training or inference
- you need a non-research-licensed or turnkey commercial product

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning, video-processing, simulation, robotics, llm-training, data-science
- domain: robotics, machine-learning, deep-learning, artificial-intelligence, simulation, computer-vision
- platform: python
- tags: world-model, robot-learning, video-generation, latent-action-model, model-distillation, egocentric-video, research-code, nvidia, icml-2026, foundation-model, linux, gpu, docker

## Member repositories
- NVIDIA/DreamDojo (main) score 48

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:25.073515+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 2026-08-30T06:57:05.501629+00:00, confidence not recorded.
  - readme: https://github.com/NVIDIA/DreamDojo (fetched 2026-08-28T04:03:25.073515+00:00, sha 8b57032b30b4)
  - homepage: https://dreamdojo-world.github.io (fetched 2026-08-29T12:59:11.723110+00:00, sha bcee6b81ee3a)
- Data as of 2026-08-30T08:39:29.467469+00:00.