Playing Slither.io with Reinforcement Learning

This is a fun project to showcase some fun RL learning on a simple to play webgame.

Components

SlitherPlayer

The player, runs on selenium. It interfaces with a web browser to "see" the game. It communicates to the trainer via gRPC stream.

SlitherTrainer

Runs a gRPC server and holds the deep learning models that make the moves.

Setup

• Run pip install on requirements.txt inside SlitherTrainer
• Inside ./SlitherTrainer run python -m src --export to export image featurizer to correct placement (should be in repo root)

Running

• Run python -m SlitherTrainer --server
• Navigate to the SlitherPlayer directory and run dotnet run
• Watch it play Note: Because of how dotnet runs, C# is setup to run from dotnet run not from an executable at the moment

Training

• While the game plays, after a certain step size it will start a training loop. Currently adding error handling to ensure that it reaches that step.
• Backup log files, data files and image are being taken in data

More documentation to come.