GDC 2025 Session Viewer

This talk delves into the development of AI Bots in Rainbow Six Siege that mimic player behavior using machine learning (ML) and statistical techniques. It explains how the Siege team gathered extensive player data and transformed it into a format usable by the game's AI, leveraging a hybrid HTN/GOAP architecture to blend ML with traditional AI systems. Key challenges included handling the complex combinations of maps, operators, and gadgets, and the need for dynamic metadata that evolves with game updates. The talk outlines specific AI components developed: gadget placement using clustering techniques, cover position prediction via neural networks, player-like pathing, realistic aiming using ML, and droning navigation. It also details the data storage pipeline in Amazon S3, extraction processes, and architecture decisions that facilitated integration and continuous R&D.

Takeaway

How can machine learning can optimize AI behavior in complex tactical games like Rainbow Six Siege. Practical applications of clustering and predictive modeling using neural networks for enhancing gameplay mechanics. Strategies for overcoming challenges in integrating machine learning with existing game development frameworks.

Intended Audience

This talk is designed for game developers, AI engineers, and anyone interested in the intersection of machine learning and game development. Prior knowledge of machine learning concepts and game AI development will be beneficial, but not required to appreciate the practical insights and innovations presented.