/hackathon_RL_XHEC_2020

Primary LanguagePython

This repository contains the materials necessary for the ending Hackathon for the 2020 RL@XHEC course.

The goal is for the students to define the rewards and train&tweak the agent to achieve the best score possible.

How-to use it:

  1. Clone this repository : git clone https://gitlab.com/the_insighters/x-hec/hackathon-rlxhec.git
  2. Enter the repo folder cd hackathon-rlxhec
  3. (Optional if already done) : install poetry curl -sSL https://raw.githubusercontent.com/python-poetry/poetry/master/get-poetry.py | python -
  4. Install the libraries : poetry install
  5. You can now run main.py using poetry: poetry run python main.py
  6. Modify main.py to tweak the agent and reward.
  7. Good luck !

To-do list

  • Transform the Rocker lander from https://github.com/EmbersArc/gym_rocketLander/blob/master/gym/envs/box2d/rocket_lander.py into an exploitable environment.
  • Customize the environment for the reward function to be definable from main.py
  • Create a nice training&testing interface
  • Add a save Agent functionnality
  • Solve the task with PPO)
  • Add video recording (instead of just being able to watch the episode at the time it is done)
  • Add a cue when the task is "solved"
  • Add landed and landed full to available variables
  • Add difficulty levels (done but not very useful, maybe a score system will be beter)
  • Add score
  • Add variables description
  • Tweak the difficulty for a nice experience
  • Test the exercise with different people to measure difficulty and time to beat

Architecture:

  • main.py : This is where the student will define reward function and agent and will be able to train and see the results
  • utils.py : Scripts to help the students. non directly related to RL.
  • gym_rocketLander (submodule) : Custom gym environment for SpaceX Rocket Lander (original work from EmbersArc modified to tweak difficulty)
  • requirements.txt : Specifies the needed libraries
  • .gitignore : Specifies which files to ignore when commiting.
  • data (folder): stores the data needed for the plots and agent. It is of no use for the user
  • figs (folder): stores the graphs (rewards over the episodes)
  • vids (folder): stores the vids of the agent's execution for each batch

Variable explanation

Disclaimer: The environment is based on https://github.com/EmbersArc/gym_rocketLander