[Feature Request] Implement MBPO algorithm

[dtch1997]

Important Note: We do not do technical support, nor consulting and don't answer personal questions per email. Please post your question on the RL Discord, Reddit or Stack Overflow in that case.

🚀 Feature

I would like to implement a model-based RL algorithm, MBPO proposed here.

Motivation

The proposed algorithm claims to be simpler and up to 10x as sample efficient as some other baselines like SAC. This would be helpful in my own work too.

Checklist

• [ x] I have checked that there is no similar issue in the repo (required)

[dtch1997]

Original implementation is available here: https://github.com/JannerM/mbpo

[araffin]

Hello, I've got mixed feeling about that one. It would be interesting to have it, at least in a separate repo. However, the current focus of SB3 is model free RL... and I hope this algorithm won't add too much complexity. So, please give it a try and open a PR once you managed to reproduce results (please read and follow SB3 contrib CONTRIBUTING.md guide carefuly). If it's too complex, we would still add a link to the implementation in our doc.

[araffin]

closing as out of scope of SB3 (we focus on model-free RL), however, if you want a model-free RL algorithm that is as sample efficient as MBPO, you can take a look at DroQ:

• in SB3 / SB3 contrib, there are branches: https://github.com/DLR-RM/stable-baselines3/pull/1036
• in SBX (SB3 + Jax): https://github.com/araffin/sbx (much faster implementation but limited features)