Arc Raiders’ New Machine-Learning Enemy Looks Absolutely Terrifying

Did you know that the Leaper, one of Arc Raiders' many enemies raiders encounter topside, learns behavior through machine learning? At GDC earlier this year, Embark Studios machine learning research lead Martin Singh-Blom explained that the Leaper learns through rewards and recognition. This helps it to act conventionally and, more importantly, overcome challenges such as losing a leg.

While it isn't quite the same as saying it's AI-led, there are certainly similarities. The Leaper also isn't the only Arc Raiders enemy to feature that type of behavior, but it's certainly the most notable, thanks to its long legs and movement capabilities. With the launch of the game's biggest-ever update, Frozen Trail, on Oct. 8, there's another enemy coming with the same type of machine-learning abilities: the Bully.

When you think of a stereotypical bully, what do you picture? Aggression, perhaps? Someone quite strong? That pretty much sums up the new foe coming to Arc Raiders. Polygon saw it in action during a virtual pre-briefing with other press and creators ahead of the Frozen Trail reveal. This is another ground-based enemy, and while it isn't quite as big as the Leaper, it is very stocky. It's also absolutely relentless; it will pursue you constantly without letting up, not giving you a second to breathe. The intention is to have another enemy, like the Leaper, that will create dynamic moments: stumbling, falling over, and crashing into players mid-fight.

I asked Arc Raiders executive producer Aleksander Grøndal about the studio's approach to machine learning, and whether we'll see it applied to even more enemies as time goes on:

"We see machine learning as an emerging technology, which is still in its infancy to some degree," he told me. "It's still a bit of an advanced research task on our end. The interesting thing about it is that you can't control it super well. We try to teach it how to behave, and sometimes that doesn't make sense at all. Sometimes we need to update the model to fix particular issues with it."

He also explained that enemies that use machine learning in Arc Raiders are "supposed to fill archetype roles in the roster," and that "as the technology becomes better, and we learn more about it, we'll utilize it in the best possible way to create unique experiences."

In a Q&A session following the pre-briefing, I asked Grøndal more about how the Bully specifically uses machine learning: "It's more ground-based and will try to traverse any terrain," he explained. "The fun thing about having the machine-learning enemies is that they're all physics-driven, and since we've introduced this tether system (detailed more in our full Frozen Trail preview), players can play around with the physicality of these enemies in a completely new way.

"Machine-learning enemies are best suited to when they have legs, when they can learn to walk. We have a couple of different platforms in which we train, and the best results we've had so far are the ones that have legs we can manipulate as players, and see how they adapt to any situation; losing a leg or getting stuck, falling over or tumbling, or things like that which create those fun and dynamic moments."

Frozen Trail will be released as a free update for Arc Raiders on Thursday, Oct. 8 for PS5, Windows PC, and Xbox Series X.