GDC 2017 Flash Forward: Deterministic vs. Replicated AI: Building the Battlefield of 'For Honor'

22 Feb 20172 min video

Transcript

So hello, my name is Gav Gilbo. I'm gameplay tech lead at Ubisoft Montreal. And I'm Frederick D, AI programmer on Foreigner. So we want to talk to you about the the upcoming presentation that we're going to be doing at GDC, which is called uh replicated versus deterministic AI building the battlefield of Forunner. So what we want to do is really showcase to you guys the failures, the pitfalls, and hopefully the successes that we had in building the game. uh it's a talk aimed mostly at AI programmers and AI designers. So on one side I'm going to be talking about the deterministic AI. What we mean about that how we ended up building a fully deterministic system for all our AI to have a distributed uh AI working on over the network but without any information sent over the network for the synchronization how we what tools we ended up building uh for navigation to combat to giving designers the proper levers for that and uh so that's basically how we built the the

armies and how the clashes happened. And on my side I will talk about the replicated AI systems that we build for the game. So in the game we have different characters. All the playable characters have a lot of specificities and we manage to extract uh all these specificities to expose them for the AI. So this will be the first part. Secondly, we will talk about how we managed to create a tool to empower the designers to be able to use all the specificities and control the characters, try everything they wanted and finally ship the game with all the AI that we needed for the PVP and for the campaign. And also we will expose all the validation system that we build for the AI uh to be able to uh validate that the difficulty of the AI that we built are what the designer wants to do to to create and if they are Korean for for all these difficulty levels. So that's it. So see you soon in San Francisco. Hope to see you there. [Music]

Auto-generated captions from YouTube — names and numbers may be misheard.

Key insights

6 takeaways · ~2 min read

01
Ubisoft implemented a dual-architecture AI system for 'For Honor' that separates large-scale battlefield actors from high-level, hero-class combatants.
02
The battlefield AI utilizes a distributed peer-to-peer deterministic simulation, allowing dozens of actors to synchronize across the network without requiring additional bandwidth for state updates.
03
Deterministic AI behavior is managed through metadata-driven world navigation and rule-based systems that provide designers with granular control over complex actor interactions.
04
Heroic AI opponents use a replicated decision-making system that extracts fight logic directly from the game's core combat mechanics to ensure reactive, high-level decision-making.
05
The development team built specialized tools to empower AI designers to manipulate character specificities and test combat behaviors across both PVP and campaign modes.
06
An automated validation pipeline was integrated to ensure AI difficulty levels consistently matched designer intent and remained coherent across varying skill levels.