Arc Raiders Uses ML for Enemy Movement

The real story here is not “AI enemies learning on the fly” but a polished production pipeline that makes game combat feel more physical and less canned.

• –Reinforcement learning appears to handle locomotion and recovery, while designers still author goals and combat logic
• –The layered setup matters: learned movement adds unpredictability without giving up control over encounter design
• –This is closer to robotics-inspired animation tech than generative AI, which makes it technically interesting but easy to overhype
• –For game teams, the takeaway is that ML can improve feel and scale, especially for complex multi-legged enemies, without replacing the core AI stack
• –The controversy around ARC Raiders’ “AI” use mostly comes from people conflating locomotion ML, TTS, and generative content