Skip to content

Dream to Control: Learning Behaviors by Latent Imagination

License

Notifications You must be signed in to change notification settings

tejshahresearch/dreamer

 
 

Repository files navigation

Dream to Control

NOTE: Check out the code for DreamerV2, which supports both Atari and DMControl environments.

Fast and simple implementation of the Dreamer agent in TensorFlow 2.

If you find this code useful, please reference in your paper:

@article{hafner2019dreamer,
  title={Dream to Control: Learning Behaviors by Latent Imagination},
  author={Hafner, Danijar and Lillicrap, Timothy and Ba, Jimmy and Norouzi, Mohammad},
  journal={arXiv preprint arXiv:1912.01603},
  year={2019}
}

Method

Dreamer

Dreamer learns a world model that predicts ahead in a compact feature space. From imagined feature sequences, it learns a policy and state-value function. The value gradients are backpropagated through the multi-step predictions to efficiently learn a long-horizon policy.

Instructions

Get dependencies:

pip3 install --user tensorflow-gpu==2.2.0
pip3 install --user tensorflow_probability
pip3 install --user git+git://github.com/deepmind/dm_control.git
pip3 install --user pandas
pip3 install --user matplotlib

Train the agent:

python3 dreamer.py --logdir ./logdir/dmc_walker_walk/dreamer/1 --task dmc_walker_walk

Generate plots:

python3 plotting.py --indir ./logdir --outdir ./plots --xaxis step --yaxis test/return --bins 3e4

Graphs and GIFs:

tensorboard --logdir ./logdir

About

Dream to Control: Learning Behaviors by Latent Imagination

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%