danijar/daydreamer
A world-model-based reinforcement learning system that trains physical robots from real-world interaction using TensorFlow 2.
Not currently ranked — collecting fresh signals.
star history
DayDreamer learns compact discrete representations of the environment using recurrent neural networks that predict sequences given actions, reconstructing inputs and predicting rewards from recurrent states. The system trains robots using on-policy reinforcement learning purely inside the learned world model representation space, enabling farsighted behavior without simulators.
Frequently asked
- What is danijar/daydreamer?
- A world-model-based reinforcement learning system that trains physical robots from real-world interaction using TensorFlow 2.
- Is daydreamer open source?
- Yes — danijar/daydreamer is an open-source project tracked on heatdrop.
- What language is daydreamer written in?
- danijar/daydreamer is primarily written in Jupyter Notebook.
- How popular is daydreamer?
- danijar/daydreamer has 447 stars on GitHub.
- Where can I find daydreamer?
- danijar/daydreamer is on GitHub at https://github.com/danijar/daydreamer.