|Deterministic vs. Replicated AI: Building the Battlefield of 'For Honor'
|Frederic Doll, Xavier Guilbeault
|Track / Format:
Did you know free users get access to 30% of content from the last 2 years?
This lecture will showcase the AI technologies that were developed in 'For Honor'.
This lecture will cover how, on one side, the team at Ubisoft created a believable large scale battlefield on the backbone of a distributed peer-to-peer deterministic simulation: dozens of interactive actors, all perfectly synchronized between peers with no network traffic. This session will show how this was achieved, from the importance of metadata built in the world, to levers given to designers to control many actors, including building complex and interesting behaviors by using simple but elegant rules. This lecture will also show how, on the other side, no battlefield is complete without heroic warriors; opponents worthy of the skills of a player. With requirements ranging from high-level objectives to reactive decision-making, including environmental awareness, those deep and challenging AIs required their own pipeline: a replicated decision system, using extracted data from Ubisoft's fight logic, tools to empower the AI designers, and automated testing to validate design intentions.