Hello all,
Welcome to Dr Derk’s Mutant Battleground!
Are you ready to build, train and test your Derkling out in the Wild West?
Dr Derk’s Mutant Battleground is our newest multi-agent RL challenge built around Dr Derk’s Gym game. It provides a lite GPU accelerated environment and employs both discrete and continuous action space. Build your own derkling, train it using many parameters at your disposal and defeat the opposition to become the winner of Battleground!
What makes this challenge interesting is the tournament-style showdown. At the end, you’ll be able to see your RL-model fight another and see your Derkling win! As a bonus, we have a special treat for all the gamers, keep reading to find out!
Make your first submission in less than 15 minutes following these 5 easy steps.
- Create a Gitlab account, set up a personal access token, add an SSH key to your profile.
- Click here to fork the starter kit.
- Create tag push to your repository with the prefix submission-
- Now, you can see the details of your submission in the repository’s issue tracker!
Click here to see more details on making submissions. Happy gaming!
Evaluation Metric
The bot submitted by the participant will be evaluated against a combination of easy, medium and difficult bots. The difference in score between the participant bot and the evaluator bot will together decide the leaderboard score.
Read more about the calculation of the weighted sum over here: https://www.aicrowd.com/challenges/dr-derks-mutant-battlegrounds/#evaluation-metric
Rewards
Apart from having an enthusiastic community of RL researchers and a chance to create real multi-agent RL models, we have an additional treat for you. Upon making your first successful submission, we will send you exclusive Steam code to get your own copy of Dr Derk’s Gym game!
If at any point you get stuck or need help troubleshooting, drop a message on our Dr Derk Discord channel. Over here, the community gets together to collaborate, share new approaches and discuss trends in reinforcement learning. Join here: https://discord.gg/jgJEBHKPzu
Resources
If you want to take a deep-dive in the field of multi-agent RL models in a gaming environment, we recommend checking out these research papers.
- Discrete and Continuous Action Representation for Practical RL in Video Games: https://arxiv.org/pdf/1912.11077.pdf
- Cooperative Multi-Agent Control Using Deep Reinforcement Learning: http://ala2017.it.nuigalway.ie/papers/ALA2017_Gupta.pdf
- Actor-Attention-Critic for Multi-Agent Reinforcement Learning: https://arxiv.org/pdf/1810.02912.pdf
You can also check out this Github repo that tries to implement minimal RL algorithms with few lines of code that can be trained within 30 seconds, even without GPU.
The organizing team consists of:
- Sanjay Pokkali (AIcrowd), sanjay@ext.aicrowd.com
- Yoogottam Khandelwal (AIcrowd), yoogottam@aicrowd.com
- Fredrik Noren (Mount Rouke), noren@mountrouke.com
- Florian Laurent (AIcrowd), florian@aicrowd.com
- Sharada Mohanty (AIcrowd), mohanty@aicrowd.com