ScholarWorks@성균관대학교

초록

The evolution toward 6G wireless communications is driving the development of intelligent, autonomous, and adaptive networks. Advanced deep reinforcement learning (DRL) has emerged as a core enabler of agentic artificial intelligence (AI), particularly when integrated with large language models (LLMs), by providing autonomous decision-making, reasoning, and action capabilities. Compared with conventional AI approaches, DRL-powered agentic AI exhibits stronger adaptability to dynamic environments and more effective sequential decision-making, making it well suited for addressing complex optimization, control, and management challenges in future wireless networks. We provide a comprehensive review of the foundations, architectures, and applications of advanced DRL in agentic AI for wireless communications. We first discuss the design process of agentic AI and LLMs, and the role of DRL in fine-tuning, decision-making, and reasoning within LLMs, with a focus on overcoming current limitations of wireless DRL under emerging 6G requirements. We then examine state-of-the-art DRL algorithms and architectures for wireless networks as well as new DRL–LLM synergy approaches that enable autonomous learning and long-term, goal-driven decision-making in agentic AI, such as LLM-enhanced RL and agentic RL. We further explore the use of DRL-powered agentic AI in wireless communication scenarios. We also summarize available open-source resources for DRL, LLM-enhanced RL, and agentic RL in wireless networks. Finally, we highlight key challenges and provide insights into promising research directions for developing robust, efficient, and intelligent wireless networks driven by advanced DRL-enabled agentic AI.

키워드