This repository contains the AI Soccer simulation environment to the AI World Cup 2020 competition. The old 2-dimensional version used in previous AI World Cups can be seen in this repository. Download the program to develop your own algorithm for AI Soccer, AI Commentator, or AI Reporter.
- AI Soccer 3D simulation program requires Webots Robot Simulator. Please refer to Webots official website's installation procedure to install Webots (Webots version should be R2020a revision 1).
How to download the simulation program
- Go to releases and download the latest version for windows (8.1 or 10) or linux (ubuntu > 16.04), respectively.
How to run the simulation program
- Install Webots
- Install Python and update the PYTHONPATH environment variable.
- Extract the simulator and open the desired world in Webots
Obs: To update the PYTHONPATH variable you can follow the instructions in this link based on your Python version.
Descriptions of the simulation environment
controllers: Contains programs for managing AI Soccer simulation game system (You cannot modify the default controllers)
examples: Contains sample programs players can refer to (Players may implement AI programs referring to the sample programs provided in this directory)
-
common: Contains a basic interface for C++ information handling and communication with the simulation program
-
Remaining directories contain samples participants can refer to.
-
general_check-variables: A program that prints game information variables sent from the simulation program to participants program
-
general_frame-skip: A program that implements framing skipping. Frame skipping is advised when your program takes more than 50 ms in each game frame in generating the output control signal
-
general_image-fetch: A program that shows the game image frames using OpenCV
-
player_deeplearning-single-dqn_py and player_deeplearning-single-ddpg_py: Programs that implement a base skeleton for AI Soccer deep learning using Deep-Q-Networks (DQN) and Deep Deterministic Policy Gradients (DDPG) for a single robot agent.
-
player_deeplearning-multi-iql-dqn_py and player_deeplearning-multi-qmix-qn_py: Programs that implement a base skeleton for AI Soccer deep learning using Independent Q Learning (IQL) and QMIX for multiple robot agents.
-
player_deeplearning-supervised-play: Program that implement a supervised training strategy based on the dataset located in Supervised Dataset
-
player_random-walk: An AI Soccer program that simply sets robot wheel speeds to random value in each game frame
-
player_rulebased-A, player_rulebased-B, player_rulebased-C: Programs that implement a rule-based control of a team in AI Soccer ('rulebased-A' is a simplified version of 'rulebased-B' and 'rulebased-C' implements a high level controller)
-
player_skeleton: A base skeleton for AI Soccer program
-
commentator_skeleton: A base skeleton for AI Commentator program
-
reporter_skeleton: A base skeleton for AI Reporter program
plugins: Contains a physics plugin used for ball-robot collision detection
protos: Contains AI World Cup object models (robot, ball, stadium, etc.)
worlds: Contains AI World Cup simulation world files (Files in this directory can be run using Webots Robot Simulator)
- aisoccer.wbt: Webots world file for senior robots
before two world files existed 'aisoccer_1.wbt' (junior robots) and 'aisoccer_2.wbt' (senior robots)
config.json: Configuration file for setting player executables, setting game duration, and setting some rules on/off for effective training. Please refer to the wiki or manual for parameter descriptions (Participants should modify the player information in this file to tell the simulation which program to execute)