Застосування методів машинного навчання у сучасних комп’ютерних іграх

Abstract

This work investigates the application deep reinforcement learning (DRL) for creating adaptive behaviors in modern video game development. The paper analyzes the evolution from rigid scripted AI to autonomous agents capable of physically accurate locomotion. Using the development expirience of ARC Raiders created by Embark Studio as a primary case study, an evaluation is performed on how agents develop emergent movement strategies based on mass, inertia, and sensory observations. Additionally, skill-balancing mechanisms, such as the Skilled Experience Catalogue, are briefly explored to demonstrate the adjustment of NPC proficiency in real-time. The study concludes that machine learningdriven approaches significantly optimize development pipelines by reducing the reliance on manual animation authoring, while also enchancing gameplay realism and fostering unpredictable interactive experiences