stable-baselines3 VS stable-baselines3-contrib

The latest release (v3.0.0) of Upkie's software brings a functional sim-to-real reinforcement learning pipeline based on Stable Baselines3, with standard sim-to-real tricks. The pipeline trains on the Gymnasium environments distributed in upkie.envs (setup: pip install upkie) and is implemented in the PPO balancer. Here is a policy running on an Upkie:

PettingZoo 1.24.0 is now live! This release includes Python 3.11 support, updated Chess and Hanabi environment versions, and many bugfixes, documentation updates and testing expansions. We are also very excited to announce 3 tutorials using Stable-Baselines3, and a full training script using CleanRL with TensorBoard and WandB.

I am wondering why there is so few algorithms in Stable Baselines 3 (SB3, https://github.com/DLR-RM/stable-baselines3/tree/master)? I was expecting some algorithms like ICM, HIRO, DIAYN, ... Why there is no model-based, skill-chaining, hierarchical-RL, ... algorithms implemented there?

Discord is more focused, and they have a page for people who wants to contribute https://github.com/DLR-RM/stable-baselines3/blob/master/CONTRIBUTING.md

Therefore, I debugged this error to the ReplayBuffer that was imported from `SB3`. This is the problem function -

Have you ever wanted to use dm-control with stable-baselines3? Within Reinforcement learning (RL), a number of APIs are used to implement environments, with limited ability to convert between them. This makes training agents across different APIs highly difficult, and has resulted in a fractured ecosystem.

Changelog: https://github.com/DLR-RM/stable-baselines3/releases/tag/v1.8.0

Great project! One question though, is there any reason why you are not using existing RL models instead of creating your own, such as stable baselines?

# https://github.com/Stable-Baselines-Team/stable-baselines3-contrib/blob/master/sb3_contrib/tqc/tqc.py :

Here's a theoretical overview and an implementation of action masking for PPO.

Simplified version of rollout collection (adapted from ppo_mask.py line 282):

Transformers (GPT-3) aren't quite _supervised_, but it does require valid samples.

Agree 100% with RL being the path forward. You probably have already seen ( https://venturebeat.com/2021/06/09/deepmind-says-reinforceme... ). Personally I'm really stoked for this https://github.com/Stable-Baselines-Team/stable-baselines3-c... , which will make it a lot easier for rubes like me to use RL.

But as we already have vanilla DQN and QR-DQN (in our contrib repo: https://github.com/Stable-Baselines-Team/stable-baselines3-contrib ) I think it is already a good start for off-policy discrete action algorithms. (QR-DQN is usually competitive vs DQN+extensions)