The system works by teaching the Arc to move and act in ways that appear dynamic and believable. Their cloud-based AI evaluates each action, categorizing outcomes as good or bad. Good actions increase the Arc’s reward value, while bad actions reduce it. Over time, this reinforcement learning loop allows the Arc to identify which behaviors lead to successful interactions with players, resulting in enemies that adapt and act in increasingly convincing and challenging ways.
In Arc Raiders, the enemy AI—called the Arc—is powered by a sophisticated machine learning setup that effectively creates a feedback loop. According to Martin Singh-Blom, Embark Studios’ machine learning research lead, the Arc are trained to seek out players because doing so generates “rewards,” which the neural network “brains” controlling them are designed to pursue. In short, the Arc exist to maximize these rewards.