During July 5-9th, 2021, the 3rd International Summer School took part online, and I had the chance to participate in it as an attendee and the honour to be one of the winners of the Game Jam. From here, a big shout out to the organizers, that had careful attention to detail, making all feel welcome.
The event gathers experts from academia and industry each year to share, learn, and Jam about the uses of AI in and for games. The agenda covers Industry lectures, essential techniques in AI and Games for playing, generating content and modelling players. Main school lectures are professors Georgios N. Yannakakis and Julian Togelious. I had the opportunity to meet them thanks to their authorship and insightful contribution to the field with their book. This book covers with an organized structure all the algorithmic approaches to different aspects of games and game development and the main milestones given by the field. It is, indeed, a worthwhile contribution to the field, passing, with elegance, the test of time.
!(https://lh6.googleusercontent.com/h9AADILxsKydgHkD0UWFPJ9hzzAUlrulVpcgliXKiWcp0GNqwzRPmpKA_CzsowHkBUvo4j8XQK_IbEnO10ylKmb3iya9dUeb4swnODEounUZqkY_ssECtadVFHbcF7SbezjTPgUv "3rd International Summer School on Artificial Intelligence and Games: AI Game Jam winner Garden of AIden overview 3")
Visual of StarCraft II Taxonomy presented during GameLab bcn 2020 based on Artificial Intelligence and Games book
The agenda was carefully crafted into days that explored different topics. As my video games research interests focus on player modelling, the first day was one of the most interesting for me: the most exciting highlights were learning from communication strategy coming from Tom Schaul that explained Reinforcement Learning fundamentals, carefully translating concepts in a hierarchical and nested manner. Duygu Cakmak AI in Strategy Games from Creative Assembly and the AI challenges faced in Total War faced, especially the decision domain of diplomacy with the Utility-Based AI handling considerations, divided into the Deal generation in terms of actions utility as the Deal evaluation technique, was beautifully structured. With their new multiagent cooperative set up, the following workshop of Unity ML agents glimpsed a new exciting direction regarding the future in player modelling. Besides, at the end of the session, I had the opportunity to participate in Summer School on AI and Games TV, fantastically hosted gratefully by Tommy Thomson and Enrica Loira.
As for the project process, five dioramas were created by an artist to settle a training dataset. A scanning tool takes images from the diorama and converts them into an array data structure, using the root training component set size and number of sample parameters. These arrays are then piped over stdin to a python process that parses them into NumPy arrays and starts training with the GAN conditional autoencoder architecture.
Capture of GAN Conditional Architecture
If you are willing to reproduce the project, feel free to reach out to the repository!
The Jam itself was a fantastic experience, and it was encouraging to receive the popular vote win. There were also other interesting projects that I personally would like to highlight, such as the project from Straw Hat Pirates, with multiagent cooperative NPC design with ML-Agents, and VAERIO-Bros, a level generation for SuperMario video game.
The 3rd edition of International Artificial Intelligence and AI Game Jam Garden of Aiden has both been an insightful opportunity to grow and learn, and even it was fully online, there were many opportunities to connect with attendees, speakers, lectures, and organizers.
Congrats also to David Melhart , that carefully worked for all been smooth!
See you at future editions of the International Artificial Intelligence and Games summer school.