A Game AI Competition to foster Collaborative AI research and development

A Game AI Competition to foster Collaborative AIresearch and development

Table of Contents


Game AI competitions are important to foster research and development on Game AI and AI in general. These competitions supply different challenging problems that can be translated into other contexts, virtual or real. They provide frameworks and tools to facilitate the research on their core topics and provide means for comparing and sharing results. Competition is also a way to motivate new researchers to study these challenges. In this document, we present the GeometryFriends Game AI Competition. Geometry Friends is a two-player cooperative physics-based puzzle platformer computer game. The concept of the game is simple, though its solving has proven to be difficult. While the main and apparent focus of the game is cooperation, it also relies on other AI-related problems such as planning, plan execution, and motion control, all connected to situational awareness. All of these must be solved in real-time. In this paper, we discuss the competition and the challenges it brings, and present an overview of the current solutions. Index Terms—game AI, collaboration, cooperation, team AI, AI competition, task and motion planning, physics-based game.

  • Author Keywords

    • game AI ,
    • collaboration ,
    • cooperation ,
    • team AI ,
    • AI competition ,
    • task and motion planning ,
    • physics-based game
  • IEEE Keywords

    • Games ,
    • Artificial intelligence ,
    • Geometry ,
    • Collaboration ,
    • Diamond ,
    • Task analysis ,
    • Planning


Game Artificial Intelligence (AI) has been around for several decades. Many computer games make use of AI, whether to serve as the players’ opponents or allies, to create real-time challenges by altering the environment or even generate entire new levels. Although the main value of computer games entertainment, they can also be used to improve other aspectsof the real or virtual world. For example, some games have educational value as well. Games based on vehicle simulators, for instance, can teach and train piloting skills that players can later apply in real-life. A similar approach could be used with artificial autonomous agents. An agent that can pilot a simulated car in a game, so it can be used as a rival to the player, during a race, could later be used as a base for self-driving cars. It all depends on the level of detail of said simulation, but in the end, this kind of development is done with several steps of incremented complexity.

There is a wide variety of skills that games can teach. Multiplayer games, for instance, can teach social skills. Inthose games, players may find themselves competing againstother players or cooperating with them. While the competition may focus more on individual skills, cooperation allows players to develop skills as team members.

Competitions of Game AI, similarly to other AI fields[1], have been used frequently and successfully to foster the development of new and innovative research. Competitions motivate many researchers to work on difficult problems and provide, at the same time, a framework for common ground tocontextualise and compare research. Also, competitions maybe used to support teaching of Game AI (and AI in general)as they present very concrete problems for students to address and, typically, share openly the submissions and results.

In this paper, we present the Geometry Friends Game AICompetition [2] that runs around a physics based collaborativepuzzle-platformer game (Geometry Friends). Because of itscooperative nature, the game promotes interesting AI challenges that are not often seen in game AI competitions. Itprovides challenges of collaboration and team AI at differentlevels. In the paper, we discuss some of the results obtainedso far and the implications and potential importance of thecompetition for game AI research. We start by explaining theGeometry Friends game in section III, after some discussionof related work on competitions of game AI, in section II. Insection IV we enumerate the AI problems that can be exploredwith the game. We then present the competition and frameworkin the respective sections V and VI. Afterwards, we describebriefly the solutions developed so far, in section VII, most ofwhich competed in previous competitions. Finally, we drawsome conclusions in section VIII and discuss future work.


Game AI competitions are important motivators for researchand development on Game AI, which deals with issues thatcan be applied to more general AI problems.

We believe that the Geometry Friends Game AI Competitionis a strong competition for it deals with several AI problems,related to cooperation, task and motion planning, and control,all this with the need to work real-time. Some of theseproblems already explored by other competitions, but ourcompetition is unique in the way it assembles them at thesame time.

he Geometry Friends is also a game that can be laterexpanded to had more challenges and deal with other AIproblems.First of all, the game has some features that have yet tobe introduced, such as, moving platforms. These features areimplemented in the game engine but left out of the competitionlevels, as we are currently waiting to get better results in thetracks as they are before adding more complexity to the baseproblem.The competition may include, in the future, a level genera-tion track as well, which we believe would be an interestingscenario for the development of PCG (Procedural ContentGeneration) algorithms. The creation of puzzle game scenariosis not something completely new in the PCG community,although we believe that the Geometry Friends game willprove an interesting test-bed for these algorithms. The mainnovelty is the cooperative nature of Geometry Friends, whichadds some interesting challenges, as levels should to be funfor both players, for example. Additionally, making sure thatthe levels are solvable may actually be a hard problem.We already have a clear specification for the levels, whichwould facilitate the development of this track. Nevertheless,we would still require an adequate framework to evaluate thelevels generated. One possibility would be to follow the currentevaluation practices in the level generation tracks where theranking process consist of interleaving artificially generatedlevels with human-created levels and allow human players toplay all.

This may lead to the creation of a level generation tool o create unlimited number of levels automatically. This maybe quite important for machine learning approaches that needlarge number of examples to be able to generalise the perfor-mance. And remove the bias of human designers as well, thatmay not include enough diversity in the levels.It would also be interesting to explore a systematic way(eventually automatic) to categorize the levels in terms ofthe challenges they provide. This can guide the definition ofspecific sets of levels and support a better comparison of theagents being developed. In particular, it would be interestingto assess the difficulty of the levels.

Another extension that has been considered is an AgentBelievability track. This track would consist of creating agentsthat would act in a human-believable way. The ranking processcould consist of a Turing-like test where users would viewvarious human and agent game-play sessions and try toidentify which are human and which are artificial.Finally, the last idea currently being developed, which is infact one of the main motivations for the development of thewhole agent framework, is the Human-AI Cooperation track.This will consist of developing a single AI agent (Circle orRectangle) that can play cooperatively with a human player.Besides the challenges presented in the previous sections, thistrack would require agents to effectively communicate withhuman players, predict their movements and interact with theircharacters in an entertaining way. The ranking of this trackwould require tracking the performance of both players (suchas the number of levels solved and how long did it take tosolve them) and randomly have users play with other usersor agents (without their knowledge) and ask them how theirplaying experience was. This will include similar concerns asthe believability track.


This work was supported by national funds through FCT, Fundac ̧ ̃ao para a Ciˆencia e a Tecnologia, under project UID/50021/2020

About KSRA

The Kavian Scientific Research Association (KSRA) is a non-profit research organization to provide research / educational services in December 2013. The members of the community had formed a virtual group on the Viber social network. The core of the Kavian Scientific Association was formed with these members as founders. These individuals, led by Professor Siavosh Kaviani, decided to launch a scientific / research association with an emphasis on education.

KSRA research association, as a non-profit research firm, is committed to providing research services in the field of knowledge. The main beneficiaries of this association are public or private knowledge-based companies, students, researchers, researchers, professors, universities, and industrial and semi-industrial centers around the world.

Our main services Based on Education for all Spectrum people in the world. We want to make an integration between researches and educations. We believe education is the main right of Human beings. So our services should be concentrated on inclusive education.

The KSRA team partners with local under-served communities around the world to improve the access to and quality of knowledge based on education, amplify and augment learning programs where they exist, and create new opportunities for e-learning where traditional education systems are lacking or non-existent.

FULL Paper PDF file:

A Game AI Competition to foster Collaborative AIresearch and development



A. Salta, R. Prada, and F. Melo,




A Game AI Competition to foster Collaborative AI research and development

Publish in

A Game AI Competition to foster Collaborative AI research and development,” in IEEE Transactions on Games,



PDF reference and original file: Click here

+ posts

Somayeh Nosrati was born in 1982 in Tehran. She holds a Master's degree in artificial intelligence from Khatam University of Tehran.

Website | + posts

Professor Siavosh Kaviani was born in 1961 in Tehran. He had a professorship. He holds a Ph.D. in Software Engineering from the QL University of Software Development Methodology and an honorary Ph.D. from the University of Chelsea.

Website | + posts

Nasim Gazerani was born in 1983 in Arak. She holds a Master's degree in Software Engineering from UM University of Malaysia.