Learning and acting in imagination
Danijar Hafner
Coauthor of PlaNet, the Dreamer series and DayDreamer, spanning latent dynamics and physical robot learning.
- Tracked papers
- 5
- Past year
- 0
- Editor picks
- 0
Mastering Diverse Domains through World Models
DreamerV3 is a general model-based RL algorithm that learns a world model and improves behavior by imagining future scenarios; with a single configuration it outperforms specialized methods across over 150 tasks, and is the first to collect…
DayDreamer: World Models for Physical Robot Learning
DayDreamer applies the Dreamer algorithm to four physical robots that learn online in the real world without simulators, using the same hyperparameters across tasks: a quadruped, two robot arms and a wheeled robot.
Mastering Atari with Discrete World Models
DreamerV2 learns behaviors purely from predictions in the latent space of a world model that uses discrete latent representations and is trained separately from the policy, and is the first such agent to reach human-level performance on the…
Dream to Control: Learning Behaviors by Latent Imagination
Dreamer learns behaviors purely by latent imagination: it trains an actor and a value function on trajectories imagined in a learned world model, backpropagating analytic value gradients through the learned dynamics.
Learning Latent Dynamics for Planning from Pixels
PlaNet learns a latent dynamics model from images and chooses actions by fast online planning in latent space, using a model with both deterministic and stochastic transition parts and a multi-step 'latent overshooting' training objective.