Skip to content

Latest commit

 

History

History
28 lines (23 loc) · 941 Bytes

README.md

File metadata and controls

28 lines (23 loc) · 941 Bytes
Raw

Orbital

An Astrophysics game embedded with Reinforcement Learning.

Objective:

  • Maximize orbit laps

Control:

  • Trust toward sun
  • Thrust away from sun

Constraints:

  • Stay within bounds of game window
  • Do not hit the sun
  • Do not hit the astroids
  • Do not run out of fuel

Under the Hood:

  • Game Environment coded from scratch using Pygame
  • Q-Learning RL Algorithm coded from scratch using Numpy
  • Functions to:
    • Reduce continuous statespace to discrete
    • Handle hyperparameter selection from terminal
    • Train independently in AWS EC2 environments
    • Log relevant learning metrics

Current Project Status:

The agent has trained against itself for 50 million episodes (games) using one set of Q-Learning hyperparameters. Currently, I am refining the algorithm hyperparameters through programatic training, evaluation, and selection.