2023 · Google DeepMind · Advanced
DreamerV3: Mastering Diverse Domains through World Models
DreamerV3 uses robust normalization and objectives to learn across more than 150 tasks with one configuration.
Direct answer
What does DreamerV3: Mastering Diverse Domains through World Models contribute?
DreamerV3 uses robust normalization and objectives to learn across more than 150 tasks with one configuration.
Background
DreamerV3 refines latent world-model training, value learning, and scaling so the same algorithm works across varied domains. It includes Minecraft diamond collection from pixels without human data or curricula.
Problem
The work addresses a central constraint in World Models: building systems that learn useful representations or actions while remaining general enough to transfer beyond a single demonstration or environment.
Core idea
DreamerV3 uses robust normalization and objectives to learn across more than 150 tasks with one configuration.
Architecture and method
DreamerV3 refines latent world-model training, value learning, and scaling so the same algorithm works across varied domains. It includes Minecraft diamond collection from pixels without human data or curricula.
- Single configuration across domains
- Robust scaling techniques
- Broad benchmark performance
Results and impact
It is a strong reference for general model-based reinforcement learning and robust training design.
Prerequisites
- DreamerV2
- Model-based RL
- Deep reinforcement learning
Recommended reading order
Read the explanation above, review the related topic pages, then use the primary-source links below to inspect the abstract, figures, experiments, and released implementation.
Primary sources
External links are provided after the context needed to evaluate the work.
Follow-up research
Related papers and concepts
World Models
A compact latent model can let an agent learn behavior inside its own predicted environment.
Dreamer: Reinforcement Learning with Latent Imagination
Dreamer learns long-horizon behavior by propagating value gradients through imagined latent trajectories.
Genie: Generative Interactive Environments
Genie learns controllable interactive environments from unlabeled internet video.
Genie 2: A Large-Scale Foundation World Model
Genie 2 generates action-controllable 3D environments from a single prompt image.
Common questions
Frequently asked questions
What is the main idea of DreamerV3: Mastering Diverse Domains through World Models?
DreamerV3 uses robust normalization and objectives to learn across more than 150 tasks with one configuration.
Why is DreamerV3: Mastering Diverse Domains through World Models important?
It is a strong reference for general model-based reinforcement learning and robust training design.
What should I learn before reading DreamerV3: Mastering Diverse Domains through World Models?
Recommended prerequisites are DreamerV2, Model-based RL, Deep reinforcement learning.