Todo

A unified hub for RL.

Starting point

We should bring in the baseline data from:

• MARLLib
• BenchMARL
• https://github.com/DLR-RM/rl-baselines3-zoo
• Gymnax
• JaxMARL

This provides a strong set of baselines. It's important to ensure that our implementations are correct, after all!

Next:

• Add theme options to the reusable visualisation helpers.
  • See the following website for examples, e.g.: https://plottie.art/plots/94907
• CleanRL and UoE Edinburgh Book?

Based on

• CleanRL
• CleanMARL https://github.com/AmineAndam04/cleanmarl
• https://github.com/FareedKhan-dev/all-rl-algorithms
• AgileRL
• Counting Reward Machines, has something on Counterfactual Q-Learning too
• PufferLib, some nice words:
  • https://puffer.ai/docs.html
  • https://puffer.ai/blog.html
• https://github.com/rail-berkeley/rlkit/tree/master/rlkit/exploration_strategies

Theoretical

• Spinning Up RL
• https://github.com/MathFoundationRL/Book-Mathematical-Foundation-of-Reinforcement-Learning

Goals

• Portable outside the current project for other projects.
• Model based or non-model based.
• Weights & Biases integration
• Logging engine.
• Visualisation/Graphing engine.
• Hyperparameter tuning (for a set of environments, too!).
• Multiagent.

Documentation

• Look into how icons are handled
• When I click on the page title in the navigation bar, I'd like to go to index or open the menu, please.

End