Adaptive Multiple AI by Leveraging Reinforcement Learning

Abstract

This paper is about utilizing reinforcement learning (RL) for the development of adaptive AI systems utilizing Unreal Engine 5. This paper focuses on how reinforcement learning can be used to create more efficient artificial intelligence or Non- Playable Characters (NPCs) that dynamically adjust their behavior based on any given in-game context. This provides not only more immersive and challenging gameplay, but also an easy way for AI to mimic human behavior. The research has a detailed analysis of integrating RL algorithm with unreal engine’s blueprint and C++ systems. This provides a framework for developers to implement adaptive AI that can evolve over time. The overall performance of this AI system is evaluated through various gameplay scenarios, that demonstrate significant improvements in player engagement. In games now a days, multiplayer experiences continue to grow in complexity and popularity. Game developers are seeking more adaptive and strategically efficient Non-Playable Characters (NPCs) to mimic human-like behavior. This paper showcases an efficient AI system where two opposing teams of agents which are trained using Proximal Policy Optimization (PPO) engage in open-field combat within a game level under Unreal Engine. Each agent aims to maximize individual and collective team rewards by eliminating opponent team members. Through iterative training and carefully designed reward structures, this AI system learns robust combat strategies which leads to an efficient combat AI that exhibits sophisticated coordination behaviors. Results indicate that PPO-driven agents can adapt to various geometrical game levels more effectively compared to conventional scripted NPCs or enemy AI. This paper demonstrates the potential of deep reinforcement learning for complex multi-agent in diverse game environments.