AI-driven decision-making is becoming integral to managing complex distributed networks such as IoT systems, cyber-physical environments, and intelligent edge networks. However, traditional AI approaches often rely on centralization, which introduces inefficiencies in energy, communication, and computation, making them unsuitable for decentralized networks. This project aims to explore Federated Reinforcement Learning (FRL) techniques that enable agents across distributed nodes to collaboratively learn optimal strategies without sharing raw data, thereby enhancing resource efficiency. By leveraging the NAISS computational infrastructure, we will develop and test novel FRL algorithms that optimize resource allocation while ensuring high performance in distributed AI tasks. Our focus will be on addressing the challenges of communication constraints, model synchronization, and energy efficiency.