claritycoders / fall-guys-ai

Fall Guys

Python / FastAI CNN - Playing Fall Guys

This code was used to gather and process data while playing the game Fall Guys. The network was then trained using the FastAI Libary.

Setup

• I used dual monitors with a game playing at a 800 X 600 resolution on one screen.
• You can start data collection by running CreateData.py
• Pressing B will start the screen / key grab. These will be stored in lists until the episode is done.
• Once the episode ( Round ) ends pressing h will stop the screen / key grab process and all data will be moved to a numpy array.
• Then I used a script in util folder called CreateImages.py to put then onto a disk drive in folders corresponding to their actions.

Train

• Use the file called training.py
• Point it at your image directory

Run Agents

• Fully random agent is RandomAgent.py
• Trained Agent is TrainedAgent.py
• You will have to load in the pkl created from training.

Inputs (Observations)

• Uses inputs to the nural network (Observations) of pixes in the game.
• 224 X 224
• Line detection

Contact!

• YouTube Clarity Coders
• Chat with me! Discord

fall-guys-ai's Issues

TrainedAgent.py Actions Don't Return An Action But Rather The File Name?

1. TrainedAgent.py - Line 43: Action being declared from result[0] but when printed, returns the name of the folder it got the data from. Therefore the if statements that activate the action are never activated with exception of Jump because of the result[2][0] > .1 referencing the tensorboard which makes no sense to me. How exactly does it return a string for the action and if I were to add more actions, how would I be able to label it like "Left Click". Please enlighten my idiotic self.

can you provide a colab version

Doesn't work

when I run anything at all, it just instantly closes. How to fix it?

Ai