# Muesli Implementation

**URL:** <https://discuss.ray.io/t/muesli-implementation/2030>\
**Category:** RLlib\
**Created:** [May 3, 2021, 6:56pm UTC](https://discuss.ray.io/t/muesli-implementation/2030 "2021-05-03T18:56:45Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![smorad](https://sea2.discourse-cdn.com/flex020/user_avatar/discuss.ray.io/smorad/32/272_2.png) [@smorad](https://discuss.ray.io/u/smorad)\
**Post date:** [May 3, 2021, 6:56pm UTC](https://discuss.ray.io/t/muesli-implementation/2030/1 "2021-05-03T18:56:45Z")

</div>

A policy-gradient algorithm named Muesli was just published by DeepMind, and it shows roughly 500% improvement over Impala across the atari benchmarks: [https://arxiv.org/pdf/2104.06159.pdf](https://arxiv.org/pdf/2104.06159.pdf)

I was wondering if this is something that would be implemented at some point. I suspect that if implemented, this would become the most popular algorithm in `rllib`.

---

<div class="post-metadata">

**Author:** ![sven1977](https://sea2.discourse-cdn.com/flex020/user_avatar/discuss.ray.io/sven1977/32/53_2.png) [@sven1977](https://discuss.ray.io/u/sven1977)\
**Post date:** [May 4, 2021, 1:46pm UTC](https://discuss.ray.io/t/muesli-implementation/2030/2 "2021-05-04T13:46:58Z")

</div>

Thanks @smorad for suggesting this! Seems like a highly complex algo to implement as it consists of so many components. But I like the different criteria they discuss in the paper aiming for a super-stable, robust, versatile algo that learns fast.  
I does look like we already have most of these components in RLlib, like MAML, model-based, PPO, etc… This just brings everything together, similar to what Rainbow did for the different “add-ons” of DQN.