Multiagent Deep Reinforcement Learning for AAV-RIS-Assisted Integrated Sensing and Communication
Aamer Mohamed Huroon, Getaneh Berie Tarekegn, Adam Mohamed Ahmed Abdo, Ammar Amjad, Li‐Chia Tai, Li‐Chun Wang
Abstract
The integration of unmanned aerial vehicles (UAVs) and reconfigurable intelligent surfaces (RIS) in sixth-generation (6G) networks offers significant potential for enhancing integrated sensing and communication (ISAC) systems. Motivated by the need for real-time adaptability, spectrum efficiency, and intelligent wireless infrastructure, this paper investigates a UAV-RIS-enabled ISAC framework capable of meeting the dual demands of sensing and communication in complex environments. A key challenge in such systems lies in dynamically balancing resource allocation, UAV trajectory planning, and interference management. To address this, we propose a joint optimization framework in which a UAV simultaneously conducts sensing operations and serves multiple ground users in obstructed environments. The formulated problem involves a non-convex trade-off between maximizing sensing signal-to-noise ratio (SNR) and communication throughput. To solve it, we develop a multi-agent deep reinforcement learning (MADRL) solution. Three collaborative agents independently optimize UAV trajectories using Deep Deterministic Policy Gradient (DDPG), RIS phase shifts using Twin Delayed DDPG (TD3), and beamforming matrices using Proximal Policy Optimization (PPO). These agents learn optimal policies under dynamic channel conditions and mobility constraints. Simulation results show that the proposed approach improves sensing SNR by 22% and communication rates by 20% compared to baseline methods. With equal weight factors for sensing and communication, the system achieves 95% of the maximum theoretical sensing SNR while ensuring a minimum communication rate of 1 bps/Hz, validating its effectiveness in real-time ISAC environments.