# danijar/dreamerv2

Mastering Atari with Discrete World Models

Repository: https://github.com/danijar/dreamerv2
Canonical: https://ross.abutalabs.com/products/dreamerv2
Homepage: https://danijar.com/dreamerv2
Language: Python
License: MIT
License Family: permissive
Topics: reinforcement-learning, world-models, artificial-intelligence, robotics, deep-learning, video-prediction, atari, research, machine-learning
Last push: 2023-01-21T07:43:42+00:00

## Health v2 (maintenance only)
Score: 32/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 35, longevity 100
- inputs: {"age_days": 2098, "days_push": 1320, "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 1056, forks 212 (observed 2026-08-28T04:03:24.450372+00:00)

## What it is
A TensorFlow 2 implementation of the DreamerV2 model-based reinforcement learning agent that learns world models from high-dimensional images and achieves human-level Atari performance. It is installable via pip and trains on a single GPU across discrete and continuous action environments.

## Use cases
- train a reinforcement learning agent on Atari games
- learn a world model from image observations
- run model-based RL on Gym environments like MiniGrid
- reproduce DreamerV2 research results
- compare model-based agents against Rainbow and IQN
- train RL agents with continuous action spaces

## When to choose
- you need a proven model-based RL agent with human-level Atari performance
- you want to train from high-dimensional image inputs on a single GPU
- you need a pip-installable RL library supporting discrete and continuous actions
- you are doing research on world models or video prediction

## When to avoid
- you need a simple model-free baseline like DQN or PPO
- you require PyTorch instead of TensorFlow 2
- you need production RL deployment rather than research experimentation
- you need actively maintained code with recent updates

## Facets
- artifact type: library
- maturity: stable
- function: reinforcement-learning, machine-learning, simulation
- domain: reinforcement-learning, artificial-intelligence, deep-learning, robotics
- platform: python
- tags: world-models, atari, model-based-rl, tensorflow, research-code, video-prediction, gpu, linux

## Member repositories
- danijar/dreamerv2 (main) score 32

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:24.450372+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 2026-08-30T06:58:22.535348+00:00, confidence not recorded.
  - readme: https://github.com/danijar/dreamerv2 (fetched 2026-08-28T04:03:24.450372+00:00, sha 94d6c2474251)
  - homepage: https://danijar.com/dreamerv2 (fetched 2026-08-29T13:00:10.742044+00:00, sha 64ca891d7905)
  - registry_pypi: https://pypi.org/pypi/dreamerv2/json (fetched 2026-08-29T13:00:10.751214+00:00, sha c2794f5d6f25)
- Data as of 2026-08-30T08:39:29.467469+00:00.