Room 1

Take 20 minutes to stretch your legs, grab a coffee and decide what talks to check out next!

Implementing AI for tactical games is challenging, especially when the AI opponent must make decisions akin to a human player and provide a challenge in a one-on-one match. And if you add geometric considerations like planning for movement of physical objects, you get a recipe for an AI programmer’s headache. However, with the right tools, nothing is impossible! This presentation showcases a case study on AI bots developed for the unique tactical sports game, Soccer Kids by Acid Wizard. It will be invaluable for AI programmers, particularly those interested in strategic and tactical genres.

Takeaways:

• Post-mortem of a unique game.
• Introduction to Monte Carlo Tree search.
• Handling physical collisions and other geometric aspects in MCTS simulations.
• Tips and tricks for spatial analysis used for determining possible unit movements in a continuous space and evaluating the overall tactical situation.

The talk gives an overview of components used to build the AI behavior of machines for Horizon Forbidden West. It follows the examples of the Waterwing from the Burning Shores DLC, a semi-aquatic flying machine that fights the player in any medium it encounters them in. The talk will cover topics such as in-air and underwater navigation and movement, and combat behaviors. Most other AI characters in the game share these components.

Takeaways:

• Gain insight into what goes into building the AI behavior of the Waterwing. These parts generalize to the other machines built for Horizon Forbidden West.
• Learn about the navigation and movement solutions used in Horizon Forbidden West in-air and underwater.
• Learn about the AI components used to build challenging machines to fight against.

In this talk we explore the AI design philosophy for Space Marine 2, our various solutions that helped to achieve the desired vision and overcome obstacles. As well as a brief introduction of our AI framework implementation.

Takeaways:

• Opportunistic and Reactive AI design and techniques
• A way to approach Many vs Many combat scenarios
• Moveset and behaviour design with Player Gameplay Graph approach
• Why Behaviour Trees and Utility AI make a great mix

Machine learning for non-player character control is often unwieldy and takes a lot of time. In this talk we will share some reliable design patterns I have learnt researching and developing learning agents from collaborations on Age of Empires 4, Bleeding Edge, Minecraft and more.

Takeaways:

• A method to stabilise reinforcement learning to reliably improve Non-Player Character (NPC) performance;
• A simple way to make the NPC’s learnt behaviour more human-like;
• How to combine both to make NPC’s collaborate with any human player.

In Rain World, we used AI as an fundamental building block for our core game loop. The game is essentially about interacting with AI and AI agents interacting with each other. In this talk we give a breakdown of the systems employed and how this enabled for the design of Rain World to be achieved.

Takeaways:

• Multiple AI agents interacting will be unpredictable
• The level at which you comminicate what’s going on inside an AI agent is a delicate decision, with upsides and downsides to both extremes
• Making AI and emergent behaviour the core feature of your game will almost definitely make it “interesting” rather than “smooth”

We’ll explore the promise and challenges of using generative techniques to create new experiences in established worlds. Large language models are a fraught technology, including in their lack of regard for the boundaries of intellectual property. At Hidden Door, we’ve been combining them in constrained ways with procedural generation, templating, and “classical” natural language techniques to allow stories that feel meaningfully different every time, but which respect the world in which they occur.

We’ll speak to the kind of input that creative partners (authors, film producers, and other creators) want, and discuss the risks and pitfalls of generative technology (oh hi, bias) and how we mitigate those. We can also share how we’ve approached these challenges in our game, with examples like:

• Using language models to choose which templated story beat happens next, instead of generating them from scratch.
• Treating language as a unique affordance for player characters (my strong character is different than your strong character).
• Creating lots of content using procgen and language models together, with a game designer’s curation.
• When we take the reigns off and let worlds get weird (cybernetic mech skeleton in the Wizard of Oz?).
• How to do this ethically, from a technology, data, and business perspective.

Takeaways:

• A grounded perspective on the use of large language models: both boons and pitfalls.
• Some approaches to mitigating bias when using those models.
• Examples of non-LLM machine learning techniques and their use in generative stories.