Implementations of Curious Replay, a method for prioritizing experience replay that is tailored to model-based reinforcement learning agents.
Experiences are prioritized based on how interesting they are, as measured by a curiosity signal. In combination with DreamerV3, this method achieves a new state-of-the-art on the Crafter benchmark.
If you find this code useful, please reference in your paper:
@article{kauvar2023curious,
title={Curious Replay for Model-based Adaptation},
author={Kauvar, Isaac and Doyle, Chris and Zhou, Linqi and Haber, Nick},
journal={International Conference on Machine Learning},
year={2023}
}
Here we provide links to implementations with different model-based agents.