Towards structured world models for control

In recent years model-based RL has broadly lived up to
its promise and has kept bringing outstanding benchmark performance,
especially in the limited data regime.  Secondly, there is a growing
interest in using structural priors to improve world models, making
them less data-hungry, more interpretable and generalizable.

We are going to give an overview of the recent literature and if time
allows, take a deeper dive into Dreamer - a recent model-based RL
algorithm.

Bibliography:

Model-based rl / control
Mastering Atari with Discrete World Models
https://arxiv.org/abs/2010.02193

Dream to Control: Learning Behaviors by Latent Imagination
https://arxiv.org/abs/1912.01603

Structural models
SPACE: Unsupervised Object-Oriented Scene Representation via Spatial
Attention and Decomposition
https://arxiv.org/abs/2001.02407

Improving Generative Imagination in Object-Centric World Models
https://arxiv.org/abs/2010.02054