- 2021.11.13 We are holding a challenge —— 2021 "AI BOB：Go-Bigger Multi-Agent Decision Intelligence Challenge". Come and make your agents in the game!
Multi-agent confrontation is an important part of decision intelligence AI, and it is also a very challenging problem. In order to enrich the multi-agent confrontation environment, OpenDILab has developed a multi-agent confrontation competitive game, named GoBigger. Based on GoBigger, the purpose of this challenge is to explore the research of multi-agent games and promote the training of technical talents in the fields of decision intelligence, so as to create a "global leading", "original" and "open" decision intelligence AI open source technology ecosystem.
This challenge needs competitors to submit their agents. We will return the score for agents to help competitors have a more correct understanding of the performance of the submitted agent. At the end of the challenge, we will fully test all submissions and the final ranking of the participating teams will be conducted.
This challenge uses Go-Bigger as the game environment. Go-Bigger is a multi-players competitive environment. For more details, please refer to the Go-Bigger documentation. In the match, each team participating in the challenge controls one team in the game (each team consists of multiple players). Contest participating teams need to submit an agent to control a certain team in the match and the players it contains, and obtain higher scores through teamwork, thereby achieving a higher ranking in the match.
Here in submit, we provide examples of submission for all teams in our challenge. We also provide BaseSubmission, and participants should implements their owned submission based on the code.
class BaseSubmission:
def __init__(self, team_name, player_names):
self.team_name = team_name
self.player_names = player_names
def get_actions(self, obs):
'''
Overview:
You must implement this function.
'''
raise NotImplementedErrorNote that all submission should extend with BaseSubmission. We will provide team_name and player_names for each submission as their basic parameters. team_names means the name of team which this submission controls. We also know that there are several players in a team, which is relative with the player_names in the parameters. We will call get_actions() when we try to get actions from this submission. So that participants should implements get_actions() in their submission. This function will receive obs as its parameters, which is similar with what we provide in tutorial. For example, submissions will get obs as following:
global_state, player_state = obsglobal_state in details:
{
'border': [map_width, map_height], # the map size
'total_time': match_time, # the duration of a game
'last_time': last_time, # the length of time a game has been played
'leaderboard': {
team_name: team_size
} # the team with its size in this game
}Participants can find their team_name in submission matched with the team_name in leaderboard.
player_state in details:
{
player_name: {
'feature_layers': list(numpy.ndarray), # features of player
'rectangle': [left_top_x, left_top_y, right_bottom_x, right_bottom_y], # the vision's position in the map
'overlap': {
'food': [[position.x, position.y, radius], ...],
'thorns': [[position.x, position.y, radius], ...],
'spore': [[position.x, position.y, radius], ...],
'clone': [[[position.x, position.y, radius, player_name, team_name], ...],
}, # all balls' info in vision
'team_name': team_name, # the team which this player belongs to
}
}
However, we will only provide the submission with the player_state matched with its players. That means, if player_a and player_b (both are player name) are in the team belongs to this submission, and player_c not belongs to this team, participants will only get player_a and player_b in the submission.
After getting the obs, submissions should return actions in get_actions(). actions should look like:
{
player_a: actions_a,
player_b: actions_b
}Remember that both player_a and player_b should be the name in your submission's player_names. And actions_a should be a list, which contains there items, which are the same with what we propose in action-space.
We provide RandomSubmission and BotSubmission. RandomSubmission provide actions randomly, and BotSubmission provide actions based on a script. Both of them could be an example of your submission. More details in code.
We also provide an example for the pipeline of the submission. Please refer to submission_example for more details. You can also develop your agent in this directory. Once you finish your my_submission.py, you can call python -u test.py to check your submission and finally get the .tar.gz file to upload.
If you want to add other things in your submission, such as model checkpoints or any other materials, please place them in ./supplements and tar them with submission.
You should place all your code and materials under my_submission/. Use tar zcf submission.tar.gz my_submission/ to get your final submission files. The final submission.tar.gz should be:
- my_submission
| - __init__.py
| - requirements.txt
| - my_submission.py
| - supplements/
| - checkpoints or other materials
Attention: __init__.py should be an empty file.
Maybe you are not very familiar with our competition, but don't worry, we provide the simplest case submission! Try the following code to quickly generate a my_submission.tar.gz for submission!
$ cd submit/submission_example
$ python -u test.py
The above test.py will check whether your submission is correct. If it is correct, you will get the following output:
Success!
###################################################################
# #
# Now you can upload my_submission.tar.gz as your submission. #
# #
###################################################################
Now you only need to submit your my_submission.tar.gz!
- Note: This submission is made of a random policy. You can check the code and change the policy to get better performance.
We also develop submission_example_di based on DI-engine. You can place your ckpt in supplements to get a completed submission.