Terrifying New Machine-Learning Enemy Revealed in Arc Raiders

Did you know that the Leaper—one of the many hostile surface threats players face in Arc Raiders—actually picks up movement patterns through machine learning? Speaking at a GDC presentation earlier this year, Embark Studios’ machine learning research lead, Martin Singh-Blom, explained that the adversary relies on a reward-and-recognition system. This allows the creature to move with natural fluidity and, crucially, adapt dynamically when overcoming obstacles, such as compensating for the loss of a limb.

While it isn’t strictly traditional artificial intelligence, the underlying concepts share a strong resemblance. The Leaper isn’t the sole combatant in Arc Raiders to utilize this adaptive behavior, though its spindly limbs and agile traversal make it the most prominent example. However, with the arrival of the game’s massive Frozen Trail update on October 8, a second machine-learning-driven foe is joining the roster: the Bully.

When you picture a stereotypical brute, what comes to mind? Raw aggression? Imposing physical strength? That sums up this fresh adversary coming to Arc Raiders. We caught a glimpse of it in action during an exclusive virtual preview session hosted for press and content creators ahead of the Frozen Trail launch. While it lacks the towering height of the Leaper, this ground-based combatant makes up for it with a remarkably stocky and relentless build. It pursues targets relentlessly, refusing to back down and leaving very little room to breathe. The design goal is much like the Leaper’s: to spark chaotic, emergent moments where enemies stumble, trip, and collide with players right in the heat of battle.

We asked Arc Raiders executive producer Aleksander Grøndal about the team’s philosophy regarding machine learning and whether we can expect to see the technology integrated into even more adversaries down the line:

“We view machine learning as a cutting-edge technology that is still somewhat in its infancy,” he shared. “For us, it remains an advanced research endeavor. The fascinating aspect of it is that you cannot fully micromanage the outcomes. We try to instruct the system on how to behave, but sometimes the results are completely nonsensical. In those instances, we have to tweak the underlying models to resolve specific anomalies.”

No more monthly updates, Embark says

He also noted that these machine-learning-powered units in Arc Raiders are “meant to fulfill specific archetype roles within the lineup,” adding that “as the technology matures and our understanding deepens, we will leverage it to craft entirely unique player experiences.”

During the subsequent Q&A session, we pressed Grøndal further regarding how the Bully leverages this adaptive tech: “It operates primarily on the ground and will attempt to conquer any terrain in its path,” he noted. “The great thing about utilizing machine-learning opponents is that their actions are entirely physics-driven. Coupled with our new tether system—which we dive into extensively in our complete Frozen Trail overview—players can interact with the physical presence of these foes in completely unprecedented ways.”

“Enemies driven by machine learning shine brightest when they possess limbs and can figure out how to walk,” Grøndal continued. “We train models across several different environments, and our most successful results so far involve creatures with legs that players can actively manipulate, observing how they react to pressure—whether that means losing a limb, getting stuck, stumbling, or tumbling—to deliver those unpredictable, high-energy encounters.”

The Frozen Trail expansion launches as a complimentary update for Arc Raiders on Thursday, October 8, across PlayStation 5, Windows PC, and Xbox Series X/S.

“Different players have different needs”

Source: Polygon