Malmo Collaborative AI Challenge

The Malmo Collaborative AI Challenge

Presenting the winners of the Project Malmo Collaborative AI Challenge > (opens in new tab)

A long-standing goal of researchers in Artificial Intelligence (AI) is to develop technology that collaborates with humans to achieve their goals. But, research in Collaborative AI poses unique challenges:

  • How can AI agents learn to recognize someone’s intent (that is, what they are trying to achieve)?
  • How can AI agents learn what behaviors are helpful when working toward a common goal?
  • And, how can they coordinate or communicate with another agent to agree on a shared strategy for problem-solving?

While these and other challenges in Collaborative AI remain unsolved, we believe that recent advances have created unique opportunities for finding answers to key questions.

That’s the thinking behind the Malmo Collaborative AI Challenge, designed to encourage research relating to various problems in Collaborative AI. The Challenge takes the form of a collaborative mini-game in which players need to work together to achieve a common goal. Challenge participants are invited to develop and train collaborative AI solutions that learn to achieve high scores across a range of partners.

Innovative AI solutions to the Challenge can lead to breakthroughs in Collaborative AI technology, and we’d like your help!

Are you interested in developing innovative AI solutions? Read how it works!

The Malmo Collaborative AI Challenge uses Project Malmo (opens in new tab) [1], an artificial intelligence experimentation and research platform that is built on top of the popular Minecraft (opens in new tab) game. Project Malmo was created to inspire a new generation of research into problems presented by this unique environment.

The Malmo Collaborative AI Challenge takes the form of a collaborative mini-game, based on an extension [2] of the game theoretic “stag hunt (opens in new tab).” The stag hunt is a classic example modeling the trade-offs between choosing to collaborate and going solo. We have implemented this concept in the form of the Malmo Pig Chase. The task and sample code are freely available on GitHub.

And what’s in it for you?

Besides the joy of problem-solving, Microsoft is offering the following prizes:

During registration, Challenge participants can sign up for either or both prizes. Note that your participation is subject to Challenge Rules (see separate tab on this page).

Here’s how to participate!

  1. Check the Challenge Rules (see separate tab on this page). Note that the Challenge is targeted at participants affiliated with a university or nonprofit research organization.
  2. Form a team (of up to three participants), decide on a team lead, and create a GitHub repository for your submission. Note that you will need to provide a valid link to the GitHub repository at the time of registration. We recommend that you start by creating a fork of https://github.com/Microsoft/malmo-challenge (opens in new tab).
  3. Decide on whether you want to use Microsoft Azure for developing and training your agent. If you do, note that we will make available, at no charge, Microsoft Azure Passes (market value $1,500 USD) to the first 150 entries (subject to requirements outlined in the Challenge Rules). Please indicate this during registration.
  4. Register your team here (opens in new tab). Please make sure that your team’s registration form is complete before submitting. Once submitted to Microsoft via your registration form, accepting any changes to your GitHub repository and registration details will be at the sole discretion of Microsoft.
  5. Write code and train your agent, who will take the role of one of two players. You can update your submission on your GitHub repository until the entry submission deadline.
  6. Before the entry submission deadline, create a one-page summary of your GitHub submission that describes your approach, and create a one-to-three-minute video that shows off your agent. Include the summary and link to the video on your GitHub overview page.

After the entry submission deadline, we will assess all Challenge submissions. The best agent is the one who plays best with all other players. Check the winner selection section in the Challenge Rules for further details. We will notify the finalist on June 5, 2017.

Before you sign up, please make sure to study the Challenge Rules (opens in new tab) carefully.

GitHub

Important dates

  • Registration deadline: April 21, 2017, 23:59 UK Time
  • Entry submission deadline: May 22, 2017 23:59 UK Time
  • Finalists notified: June 5, 2017

Useful links

Contact us

For any queries please see our FAQ section or send email to [email protected] (opens in new tab). For technical questions please use the Malmo Forum (opens in new tab).

References

[1] Johnson M., Hofmann K., Hutton T., Bignell D. (2016) The Malmo Platform for Artificial Intelligence Experimentation (opens in new tab). Proc. 25th International Joint Conference on Artificial Intelligence (opens in new tab), Ed. Kambhampati S., p. 4246. AAAI Press, Palo Alto, California USA.

[2] Yoshida, W., Dolan, R.J. and Friston, K.J., (2008) Game Theory of Mind (opens in new tab). PLoS Computational Biology (opens in new tab), 4 (12), Article e1000254.