A clone of the famous multiplayer game https://agar.io/ in python 3.6 that supports third-party AI controllers and can be easily used for reinforcement learning experiments.
To play a game with your mouse, start the program with tag --play and then list the controllers used by opponents, for example: python main.py --play 3,coward_hungry_ai_controller 8,stupid_hungry_ai_controller 1,middle_ai_controller
A left click on your mouse splits the blobs, a right click shoots a pellet from the main blob.
Your new controller should be a class with the same name as its file and should be placed inside the 'controllers' folder. The class should implement abstract_controller and should have parameterless constructor and parameterless method update(). When correctly mentioned in command-line arguments, the class will be automatically registered and provided with an instance of manipulator class in self.manipulator. The update() function is automatically called each frame, inside you should use (only) self.manipulator to gather information about the environment and control your blobs.
main.py [-h] [--display] [--play] N,C [N,C ...]
N,Clist of pairs where C is controller class name and N is the number of these controllers to instantiate
-h,--helpshow this help message and exit--displaydisplay the first controller from the first person view--playplay for one blob manually
python main.py 10,coward_hungry_ai_controller
python main.py --display 3,coward_hungry_ai_controller 8,stupid_hungry_ai_controller 1,middle_ai_controller
python main.py --play 3,coward_hungry_ai_controller 8,stupid_hungry_ai_controller 1,middle_ai_controller
- pygame
TODO
https://gamefaqs.gamespot.com/webonly/163063-agario/faqs/73510
- support fast play without graphical interface
- play a batch of games and output the statistics
- make a reasonable game mode (2 minute play with resurrection?) that can be better used for reinforcement learning
- team play
- add comments
- create an interface for reinforcement learning
- add leaderboard to GameboardView