Journal article Open Access

World Models

Ha, David; Schmidhuber, Jürgen

We explore building generative neural network models of popular reinforcement learning environments. Our world model can be trained quickly in an unsupervised manner to learn a compressed spatial and temporal representation of the environment. By using features extracted from the world model as inputs to an agent, we can train a very compact and simple policy that can solve the required task. We can even train our agent entirely inside of its own hallucinated dream generated by its world model, and transfer this policy back into the actual environment.

An interactive version of this article is available at

Interactive version of the article at
Files (3.0 MB)
Name Size
3.0 MB Download
All versions This version
Views 905899
Downloads 472470
Data volume 1.5 GB1.4 GB
Unique views 855850
Unique downloads 433431


Cite as