danijar/dreamerv3

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

3704 stars · 598 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

A Python/JAX reimplementation of DreamerV3, a model-based reinforcement learning algorithm that learns a world model and trains an actor-critic policy from imagined trajectories. It masters diverse control tasks (Atari, Minecraft, Crafter, robotics) with fixed hyperparameters and favorable scaling properties.

Use cases

• train a reinforcement learning agent that works across many tasks without hyperparameter tuning
• reproduce DreamerV3 benchmark results on Atari or Crafter
• learn a world model from sensory inputs and train policies from imagined trajectories
• apply model-based RL to Minecraft or other diverse control domains
• experiment with scalable RL algorithms in JAX
• study data-efficient reinforcement learning with larger world models

When to choose

• you need a general RL algorithm with fixed hyperparameters across domains
• you want a well-cited, actively maintained reference implementation of DreamerV3
• you prefer JAX for GPU-accelerated RL research

When to avoid

• you need a simple tuned RL baseline for a single specific environment
• you lack GPU resources, since world model training is compute-intensive
• you need a production RL service rather than research code

Facets

library · maturity active

machine-learning reinforcement-learning simulation artificial-intelligence reinforcement-learning machine-learning gaming-tools python world-models jax model-based-rl actor-critic research-code linux macos gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="danijar/dreamerv3")