Deep Reinforcement Learning Over RIS-Assisted Integrated Sensing and Communication: Challenges and Opportunities

The advancement of deep learning (DL) significantly accelerates the development of future integrated sensing and communication (ISAC) systems. Deep reinforcement learning (DRL), as a promising DL approach, has emerged to leverage a distributed personalized dataset from different reconfigurable intelligent surface (RIS) nodes. However, the high costs associated with data offloading and model training pose challenges to implementing network intelligence within existing ISAC frameworks, particularly at the network edges. To address this limitation, a paradigm of RIS-enabled DRL technology is developed that can overcome the arithmetic, high-frequency transmission, and coverage region problems. The fundamental studies with respect to RIS-assisted ISAC modeling and its solution are investigated, which can provide insights into the design of RIS-enabled DRL in ISAC networks. To facilitate the corresponding implementation, key techniques are proposed to integrate the communication, sensing, and computation capabilities of the ISAC network. Moreover, future trends of RIS-enabled DRL technology for ISAC networks, such as potential applications and open issues, are discussed.