Skip to content

Reinforcement learning agent using dqqn, dueling network, per to play the google chrome trex browser game.

Notifications You must be signed in to change notification settings

rybread1/deep-rl-trex

Repository files navigation

DeepRlTrex

Reinforcement learning implementation of double-deep-q-learning, dueling network architure and PER to play the Google Chrome Trex Game directly from the browser.

To run this on your local machine you'll probably have to fine-tune a few parameters:

  • Because the env runs by grabbing screenshots directly from your monitor you will have to ensure the screen capture dimensions are actually working correctly
    • Check the bbox and terminal_bbox: bbox should capture the play area for the actualy game, terminal_bbox should capture some unique identifier for when the game is over
    • Make sure the render dimensions are large enough to fully expand the entire "runway" for the t-rex. If the dimensions of the window rendered are too small, the agent has more difficulty (since it can't see as far ahead)
    • Make sure that the _update_state() function properly reshaping the frames
    • Finally, in agent.py, check the input dimensions for the CNN

References:

About

Reinforcement learning agent using dqqn, dueling network, per to play the google chrome trex browser game.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published