Hosted on github pages

RL agent for gridworld problem

JavaScript implementation for a TD RL agent learning optimal paths on a gridworld. Inspired by Reinforcement learning specialization

To regenerate a new random gridworld - click "apply".

Amount of bombs is scaled bases on size of grid world.

To learn an agent - click on "run RL". Might take some time on slower devices or bigger grid sizes.

Current state

• implemented a gridworld problem with some obstacles(bombs are bad for the agent)
• implemented SARSA agent with these parameters:
  • ε-greedy policy (starting with 0.5 and decaying over time)
  • 1000 episodes
  • no discounted reward
  • step size of 0.1

Frontend

Open Web Components library is used for frontend. No specific reason for it, just wanted to give it a try :)

Scripts

• start runs your app for development, reloading on file changes
• start:build runs your app after it has been built using the build command
• build builds your app and outputs it in your dist directory
• test runs your test suite with Web Test Runner
• lint runs the linter for your project