Game AI North 17 was a two-day conference and community event focused on design, systems, and production for professionals in the field of game AI development. The conference was held October 17-18th in Copenhagen, Denmark.
The backbone of GAIN17's schedule consisted of 14 sessions where industry-leading speakers provided insight into their professional work. These talks were supplemented by roundtables and workshops where attendees could take part in discussions focused on Game AI.
In this talk by Adrien Allard and Sébastien Broussaud, you will learn how Amplitude Studios developed an AI that offers a challenging gameplay experience while maintaining narrative coherence in Endless Space 2.
Adrien has been with Amplitude Studios since its creation in 2011. He has worked on many different 4X game elements from gameplay and audio integration to artificial intelligence. For the last several years he has been working on the design and architecture of the Endless Space 2 AI framework and the tools to understand and implement it. He is currently working on developing advanced content for diplomatic interactions.
Sébastien is a programmer with over 10 years of industry experience. He worked at Dontnod, Heliceum, and Lightbulbcrew before joining Amplitude Studios, always with a focus on Artificial Intelligence technologies and related software architecture. He is currently working on Amplitude's Endless Space2 project, focusing on creating a complete AI for 4X Games.
For Guerrilla, Horizon Zero Dawn was a massive change in direction. Transitioning from the level-based, tactical, first person shooter genre to open world, action RPG required us to develop many new systems and to redesign existing ones. In this session we will describe the challenges we faced when developing the AI for Horizon Zero Dawn and the solutions we chose. Some of the subjects we will cover are navigation for agents of varying size in a dynamic environment, how we use a combination of HTN planning and utility based decision making, and the systems we developed to coordinate groups of agents.
Arjen Beij decided at an early age that making games would be the best job in the world. After studying computer science at the Hogeschool van Amsterdam he joined Lost Boys Interactive in 1999, where he got the opportunity to develop his first commercially released games. Two years later he joined Guerrilla as an AI programmer where he worked on the AI for Killzone, Killzone 2, Killzone 3, Killzone Shadow Fall and Horizon Zero Dawn. For the past six years he has led the AI team at Guerrilla.
Sea of Thieves is a shared world pirate game with AI running on a dedicated server. In this talk Rob Masella will describe how Rare's extensive use of automated testing meant that Sea of Thieves could ship frequently and with confidence that AI agents would behave correctly in a number of challenging multiplayer scenarios. Rob will lay out the advantages of a good testing infrastructure, the ability to test multiplayer network interactions and, with lots of examples from Sea of Thieves development, talk about how the team used tests to find and fix a number of issues before they shipped to customers.
Rob Masella has been a gameplay engineer at Rare for 12 years, working on titles like Banjo Kazooie: Nuts and Bolts, Kinect Sports and Sea of Thieves. In this time, he's been involved in many feature areas such as physics, character control and animation. In the last couple of years, he's been focusing on AI, bringing enemies to life in Sea of Thieves.
In his talk, Christian will provide practical insights into Crytek's Universal Query System. Learn how this fully customizable generic system for doing spatial queries was built from the ground up, how it works, and how it can be used by AI characters and gameplay systems in general In particular, the audience will gain an in-depth understanding of the architectural decisions and design considerations that went into the development of this system. Also, the talk will explore how queries can be debugged and how programmers and designers alike can analyze the decisions that were made throughout a query.
Christian joined the games industry in 2011 after he had been working on all kinds of B2B software for 8 years. He has developed the AI systems for a 3rd person stealth/action game on XBox360 + PC. In particular he built a Behavior Tree system, Perception system, Tactical Positioning system, Cover management and all Enemy behaviors as well as sophisticated search behaviors. Since 2013 he is working at Crytek as AI Programmer on various aspects of AI, most notably on their new spatial query system.
The field called deep learning, i.e. methods using deep neural networks, has seen explosive progress in the last couple of years. Neural nets has enabled significant breakthroughs in everything from computer vision, voice generation, voice recognition, translation, and self-driving cars. Neural nets will also be an amazing enabler for game development. This presentation will give an overview of the potential of neural nets as game making tools, as well as provide an in depth look at how we can use neural nets combined with reinforcement learning for new types of game AI.
Magnus published his first game in 1983. Since then he has spent 25 years doing computer science and software engineering in a large number of projects and companies. He is currently heading up the deep learning and AI research team of SEED, an EA R&D division.
This presentation provides a practical look into how Warpzone Studios is building a game with autonomous agents on an Entity Component Systems architecture. Specifically it will delve into how the AI fits into that. The presentation will also go into detail about how a tile-based, platformer-style map with destructible terrain affects decision making, path finding and movement. Moddability has been a large focus on the development of the engine and the game architecture. The talk will show how all parts of the AI are extensible.
Anders is a game programmer and designer with 10 years of professional experience and dozens of hobby games under his belt. As a generalist, he's worked on gameplay, AI, core engine development, network and multiplayer code, UI, tools, optimization and physics. Having worked at ten studios and some high-profile projects, notably Avalanche Studios (Just Cause 2), Rockstar North (GTA 5) and Fatshark (Warhammer: Vermintide), he finally decided to join two old friends at Warpzone Studios as a cofounder. Anders ambition as an indie developer is to be close to the creative direction and of his projects, and to improve his skills by doing what he likes the most: learn new things.
TIn this presentation Derek will describe the workings of the Commander AI of Halo Wars 2. He will talk about the design and implementation of this AI from the tactical through to the strategic level.
Derek Fagan is an AI programmer on the Total War team at Creative Assembly. He made his debut in the games industry in 2015 when he joined CA to develop the commander AI of Halo Wars 2. Derek's main interests in AI are in the realms of NPC behaviour and machine learning. He received his PhD in computer science from Trinity College Dublin in 2016 for his research on Multi-Agent Reinforcement Learning. When not actively engaged in his endeavours to infuse AI systems with a sense of intelligence and life he is likely to be found at home spending time with his wife and six children.
Very few games get close to extreme scale and complexity of large scale player space combat in EVE online. The talk will discuss the work where multiple fleets of NPCs are trying to defend giant space station using similar mechanics as regular player vs player structure fighting.
Freyr Magnússon is a Senior Programmer at CCP Games in Reykjavik. He joined as an Associate Programmer working on EVE Online in 2008. In 2014 he introduced event driven behavior trees to EVE Online. Since then he has continued to work on upgrading NPC behaviors and spawn mechanics to break content out of old stagnant molds. Recently he has been working with his team to create large scale fleet fights with player fleets. Freyr graduated with a bachelor degree in computer science from Reykjavik University.
Mixed Reality is the mixture of the real world and the virtual worlds, so that one understands the other. MR allows us to bring games and interaction to the real world around us. To develop truly immersive experiences, AI and data parsing are huge areas for this new medium. In this talk, we will be discussing the What, Why and How of Mixed Reality.
Aleissia is a game programmer & designer who has been working as a developer for 17 years and developing games at Ubisoft for 10 of them. She was Lead AI on multiple Assassin’s Creed games. Her passion for design has pushed her teams to think about why we are developing our games rather than just how we develop them. She empowers her teams to think outside of the box through game jams and experimentation. Recently she has made the jump from video games to Mixed Reality. Joining Magic Leap as Interaction Director for their new San Francisco office. Where her team focuses on finding new and innovative ways to create rich immersive experiences focused on player interaction and narrative. She is also an AI Summit advisor for GDC. And is an advocate for Women in Games and Tech.
After spending more than 20 years as a gameplay and AI programmer in the game industry, Brian Schwab joined a little Florida company who's mission was not so little: to fundamentally change how humans use and interact with their computers. So, while in his past he shipped everything from a simple Sega Genesis jrpg to the international blockbuster Hearthstone at Blizzard, he now runs the Interaction Lab, a systems research group within Magic Leap. In addition to using crazy new hardware to explore interactions from low level to hugely complex, his team also helps bring compelling experience to the forefront of the Magic Leap engineering efforts. Throughout this, his mission has never changed: to find fun, interactive experiences and share them with the world.
In this talk, we explain the technology behind the aerial navigation in Horizon Zero Dawn. In Horizon, we've represented the flyable air space by use of a run-time generated height map. Queries can be done on this height map for positional information and navigation. We present a hierarchical path planning algorithm for finding a progressively more detailed path between two points. Additionally, we will touch on some gameplay related subjects, to show the additional challenges we faced in implementing the different flying behaviors, such as transitioning from air to ground and guided crash-landing.
Wouter Josemans is a senior AI programmer at Guerrilla Games. He graduated from the University of Amsterdam in 2012, holding an MSc in Artificial Intelligence. After graduating, he started working at Guerrilla on the various AI systems of Killzone: Shadow Fall and later Horizon Zero Dawn, where he focused mainly on combat behavior and aerial navigation. He's currently working on the Frozen Wilds expansion for Horizon Zero Dawn.
Reinforcement Learning (RL) has long been popular in academic research on game AI, but is much less commonly applied to AAA games. This talk will present an initiative within Ubisoft Montreal to use RL as a tool for game productions, by building agents that are able to play (parts of) the game during its development. Preliminary results for two applications will be shown: learning a fighting AI in For Honor to assist with game balancing and new character development, and learning a driving AI in Watch_Dogs to help reach realistic behavior with a large variety of vehicles. The talk will cover the challenges we faced when applying state-of-the-art RL algorithms to modern games, the main lessons learned, and how this project was actually helpful (or not) to designers, testers and programmers on the production floor. And, as a bonus, you may expect videos of games that look better than Pong or Doom for a change!
Olivier is currently part of the analytics team in Ubisoft Montreal’s User Research Lab, where he can have fun with tons of player data to build machine learning models from. It took him a while to get there: after graduating from the Ecole Polytechnique de Paris (2001) and from the Ecole Nationale Superieure des Télécommunications de Paris (2003), he first worked as a research assistant at Université de Montréal. He then decided a Ph.D in machine learning could always be handy, so he started one, and even finished it when Ubisoft offered to hire him in 2012.
Adrien is working at Ubisoft Montreal as an AI programmer and more precisely at La Forge, the R&D department, after two MEng (one in France and one in Montreal). His old passion for video games creation and his new one in AI offered him an opportunity to join Ubisoft in 2017. His latest work is in teaching AIs to play video games.
Our goal with Assassin's Creed Origins has been to modernize the franchise and reinventing our AI was a focal point towards reaching this objective. This challenged our team to deliver systems to make the citizens of Egypt feel like they have an agenda in the world beyond the Player. However, there are many challenges in both design and programming that we had to tackle to accomplish this goal. We had to find a way to make every AI do something coherent beyond player centric reactions at all time, make our reaction system adapt more based on what our AI previously experienced and make the search loop more about outsmarting a group rather singular AIs.
Brian Fillion is a AI Programmer at Ubisoft Montreal. Winner of the Ubisoft Game Lab Competition in 2015. Brian’s main interest on AI includes reasoning systems, systemic realization and dynamic music. He was involved in Assassin’s Creed Syndicate and is currently working on Assassin’s Creed Origins. Currently trying to push new standards for systemically driven AI realization.
This presentation will cover the implementation of the AI teammates in the systemic open world of Ghost Recon Wildlands, in order for them to be reliable and engaging for the player. I’ll quickly introduce the mainly data-driven architecture before diving in the essentials systems ruling their behavior, designed under strong creative rules : How do they acquire information regarding their surroundings, and how do they react to it while taking in consideration the player’s situation, without hindering his experience. This presentation will show us how striking a balance between purely systemic solutions and constrained player-centric rules is key to obtain a satisfying result that makes sense from the player’s perspective.
My name is Mathias Rolland, I’m 29 and I’ve been a developer for close to seven years, and working in the industry as a gameplay programmer for Ubisoft for a bit more than four. I’ve worked mostly on AI and gameplay elements, notably on the main DLCs for the two previous Assassin’s Creed games, and was in the team responsible of the Teammates AI on Ghost Recon Wildlands.
There are no weight classes in game development. There’s no division to keep the fights fair. You, your team, or your studio are going to be competing with games from vastly different team sizes, with vastly different budgets. You can use UI, AI, and performance tricks to make your games stand out among those from Goliath teams. In Race for the Galaxy, our neural network creates its own training data with which to hone itself. We run our AIs on a separate thread, to keep framerate up while it's executing. We conduct the flow of UI events through locking attention tokens to make a complex system intuitive. A few smart innovations can give a small team the edge to stand out in the ring.
Theresa Duringer is the cofounder of Temple Gates Games, the studio behind Race for the Galaxy for iOS/Android. She has been exploring UX/AI/UI particularly in the social spaces of digital boardgames. She got her start in the boardgame space creating Ascension VR, based on the deckbuilding tabletop game from Stone Blade. Up next she is partnering again with Rio Grande Games bringing Roll for the Galaxy to iOS and Android. Before hopping on board the hype-train of futurism between AI and VR, she created Cannon Brawl, an artillery RTS for XBOX, PS4 and PC and worked at Maxis on titles including Spore, Sim City and the Sims.