Agenda

What does it take to make a machine feel alive? In ARC Raiders, we set out to reinvent how enemies move — not through animation clips or hand-tuned logic, but through learning. Using state of the art methods that combine animation, reinforcement learning and physics-based control, we taught our AI to walk, run, stumble, and fight with intent. Each motion emerges from experience, not scripts. Combined with a perception system built on point clouds, these agents see and react to the world around them with surprising fluidity and realism. This talk covers how we trained and deployed these agents in Unreal Engine, the challenges of achieving responsiveness and stability in a high-fidelity action setting, and the lessons learned integrating learned locomotion into a large-scale production game.

Takeaway

Attendees will gain insights into bridging cutting-edge reinforcement learning research with practical, performant character animation in a shipped title.

Intended Audience

The technical talk is aimed at game developers interested in the intersection of movement, physics and emergent gameplay. It should be interesting for animators and designers interested in new tech as well as physics and gameplay programmers. Some high school math will help but the necessary background will be covered.