Skip to main content
This page is under construction. Content coming soon.
Model-based reinforcement learning algorithms learn a model of the environment dynamics and use it for planning, often achieving better sample efficiency than model-free methods.

​
Key Concepts

  • World Models: Neural networks that predict future states
  • Model Predictive Control (MPC): Planning with learned dynamics
  • Dyna Architecture: Combining real and simulated experience
  • Latent Space Models: Learning compressed state representations
Key references: (Schmidhuber, 2015; Lillicrap et al., 2015; Mnih et al., 2013)

​
References

  • Lillicrap, T., Hunt, J., Pritzel, A., Heess, N., Erez, T., et al. (2015). Continuous control with deep reinforcement learning.
  • Mnih, V., Kavukcuoglu, K., Silver, D., Graves, A., Antonoglou, I., et al. (2013). Playing Atari with Deep Reinforcement Learning.
  • Schmidhuber, J. (2015). On Learning to Think: Algorithmic Information Theory for Novel Combinations of Reinforcement Learning Controllers and Recurrent Neural World Models.
Edit this page on GitHub or file an issue.