AI Agent for Light Rider -- CS221 Final Project at Stanford

Authors: Yipeng He, Wantong Jiang, Di Bai

This is our final project for CS221 AI Fall 2018 at Stanford.

In this project, we implemented an AI agent for game Light Rider. The game is originally provided by Riddles.io.

The project contains:

• A simulator written by us to run 'Light Rider' game.
• Minimax agents with different settings and a linear Q-Learning agent trained against minimax agents.

Quick Start

cd Bot
python main.py

Enter the Bot directory, and run main.py. It will automatically start training a Q-Learning agent against a minimax agent.

Parameters

--mode game|rl # change mode fron Q-Learning to minimax competition
--grid 6 # change the grid size to 6
--rl_trial 10000 # change Q-Learning trial number to 10000
--test_rl # whether to test a trained Q-Learning agent
--rl_weights weights.txt # read from weights.txt to get learned parameters for Q-Learning agent
--rl_save file.txt # save learned paraments for Q-Learning agent

For more details, run

python main.py -h

To change the settings of two minimax agents in the game mode, please look into Bot/simulator.py.

To change the setting of the minimax agent against Q-Learning agent, please look into Bot/rl.py.

File Struture

• board.py implements board information, update, and print
• bot.py implements bot interface
• game.py implements interface with Riddles.io
• minimax.py implements minimax agent, minimax with alphabeta agent and different evaluation functions
• player.py implements player class, containing player's position
• rl.py implements Q-Learning agent and training against minimax agent
• simulator.py implements two player game simulator
• state.py implements game state
• util.py implements parameter parsing

License

MIT License