![](data:image/svg+xml;utf8,<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24" fill="none" stroke="%23D95C41" stroke-width="1.8" stroke-linecap="round" stroke-linejoin="round"><line x1="12.1" y1="11.9" x2="18.9" y2="8.2" /><line x1="12.1" y1="12.1" x2="20.3" y2="12.9" /><line x1="12.2" y1="12.4" x2="16.6" y2="19.1" /><line x1="11.8" y1="12.4" x2="7.3" y2="19.2" /><line x1="11.9" y1="12.1" x2="3.7" y2="13.3" /><line x1="11.8" y1="11.7" x2="7.8" y2="4.4" /></svg>)
Curriculum Roadmap
Curriculum Roadmap
Each step has a corresponding code project. When you get stuck, returning to the relevant section is more effective than reading everything through before writing any code.
Next Lecture
L02 starts from a concrete problem: how do you compress a 64×64 pixel image into a compact latent vector z? This is the task of the Variational Autoencoder (VAE), and it is the first building block of the entire Dreamer pipeline.
Once the compression is in place, z is fed into the dynamics model, which learns to predict "what z will look like at the next timestep." That is the RSSM. By the end of L02, you will have written the two most critical modules of a world model by hand, and you will be able to see from real loss curves how they learn.
This lecture requires no mathematical or coding background. If you are interested in the original papers by Craik, Ha & Schmidhuber, or Dreamer, see the further reading in L05.
Further Reading
- Craik, K.J.W. The Nature of Explanation. Cambridge University Press, 1943.
- Ha & Schmidhuber (2018): World Models: the V/M/C three-module framework and the original paper on training in dreams
- Hafner et al. (2019): Dream to Control (Dreamer V1): the first end-to-end implementation of RSSM and latent actor-critic
- LeCun (2022): A Path Towards Autonomous Machine Intelligence: the JEPA framework and the argument for world models as a cognitive core