Recurrent world models
David Ha
Coauthor of World Models and PlaNet, early approaches to learning compact dynamics for control.
- Tracked papers
- 2
- Past year
- 0
- Editor picks
- 0
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.
World Models
Trains a VAE to compress each frame and a recurrent mixture-density network to predict the next latent, then fits a tiny linear controller on their features; the agent can also be trained entirely inside the model's own generated 'dream' an…