Is model-based RL always better?

Not always. Model-based RL excels when data is expensive (robotics) or environment dynamics are learnable. Model-free RL can be simpler and more robust when abundant data is available and dynamics are highly chaotic.

What is the Dreamer family?

The Dreamer family is a lineage of world model agents: PlaNet (2019), DreamerV1 (2020), DreamerV2 (2021), DreamerV3 (2023), each building on the RSSM architecture with improving representations and training methods.

New: the Timeline is live. Track world model releases, papers, and benchmark updates in real time.

world-models.io

The Knowledge Hub for AI World Models

World Models & Reinforcement Learning

The intersection of world models and reinforcement learning, where learned environment models enable agents to plan and learn from imagined experience.

robotics model-based-rl simulation embodied-ai

External Sources

Primary references and official sources surfaced directly in static HTML for crawlers and no-JS readers.

Dyna (1991)
Historical starting point
MuZero paper
Planning with learned models
DreamerV3 paper
Modern model-based RL

References

[1] Sutton, 1991. Dyna.
[2] Hafner et al., 2023. DreamerV3.
[3] Schrittwieser et al., 2020. MuZero.

This site requires JavaScript for interactive features. Crawlable, page-specific HTML is served for every canonical URL in the sitemap.