Brainstorm ideas for V1 design and assist with training RL agents

[omar-emara]

This issue has been opened following the team meeting this week and the request for help with the following:

• [ ] Brainstorming ideas to how the prototype of V1 will look like
• [ ] Assist with training the RL agent

[omar-emara]

@lucyfarnik I think we could start with something simple like behavior cloning policy as a baseline for comparing our more advanced RL methods to. There is a good implementation here https://imitation.readthedocs.io/en/latest/tutorials/1_train_bc.html

We could also include a user interface to develop our own custom policy and replace the expert variable in the example in the documentation above. @vishaljoshi24 would need to add this in the UI if we decided to go down this route.

[vishaljoshi24]

So, Jonathan and I will sit down in the meeting tomorrow to brainstorm ideas about how the app will look like. Sketch a few drafts and then we can start implementing it on streamlit asap.

[omar-emara]

@lucyfarnik https://towardsdatascience.com/xrl-explainable-reinforcement-learning-4cd065cdec9a

Above is a good article summarizing three papers for XRL. Thought would be useful to share.