2019 · Google Research · Advanced
Dreamer: Reinforcement Learning with Latent Imagination
Dreamer learns long-horizon behavior by propagating value gradients through imagined latent trajectories.
Direct answer
What does Dreamer: Reinforcement Learning with Latent Imagination contribute?
Dreamer learns long-horizon behavior by propagating value gradients through imagined latent trajectories.
Background
Dreamer learns a compact dynamics model from pixels and uses it to imagine possible futures. Actor and value networks learn from these latent rollouts without reconstructing every future image during policy optimization.
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
Dreamer learns long-horizon behavior by propagating value gradients through imagined latent trajectories.
Architecture and method
Dreamer learns a compact dynamics model from pixels and uses it to imagine possible futures. Actor and value networks learn from these latent rollouts without reconstructing every future image during policy optimization.
- Latent imagination
- Actor-critic learning through dynamics
- Pixel-based continuous control
Results and impact
It demonstrated that latent imagination could support strong continuous-control learning from visual input.
Prerequisites
- World Models
- Actor-critic methods
- Latent dynamics
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.
DreamerV3: Mastering Diverse Domains through World Models
DreamerV3 uses robust normalization and objectives to learn across more than 150 tasks with one configuration.
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 Dreamer: Reinforcement Learning with Latent Imagination?
Dreamer learns long-horizon behavior by propagating value gradients through imagined latent trajectories.
Why is Dreamer: Reinforcement Learning with Latent Imagination important?
It demonstrated that latent imagination could support strong continuous-control learning from visual input.
What should I learn before reading Dreamer: Reinforcement Learning with Latent Imagination?
Recommended prerequisites are World Models, Actor-critic methods, Latent dynamics.