Multi-Agent DRL-Driven Age-Energy Efficiency Optimization in STAR-RIS Assisted Consumer-Grade Autonomous Vehicle Applications

Simultaneously transmitting and reflecting reconfigurable intelligent surface (STAR-RIS)-assisted autonomous vehicle networks represent a promising solution for enhancing data transmission efficiency and extending communication coverage in consumer-grade applications. Among them, autonomous aerial vehicles have attracted significant research interest as a representative application scenario. However, in multi-STAR-RIS scenarios, the coupling between STAR-RIS configuration and device scheduling not only presents optimization challenges in high-dimensional spaces but also leads to a trade-off between energy consumption and age of information (AoI). To address the above issues, this paper proposes a multi-agent deep reinforcement learning-based algorithm, termed CORAL. The proposed algorithm jointly optimizes STAR-RIS coefficient matrices, autonomous aerial vehicle trajectories, device association and scheduling variables to maximize age-energy efficiency (AEE) in multi-STAR-RIS-assisted multi-autonomous aerial vehicle data collection networks. First, we introduce the AEE to quantify the achievable AoI gain per unit of energy consumption. Moreover, we formulate an AEE maximization problem and model it as a Markov decision process. Then, we develop the CORAL algorithm incorporating joint action-value estimation, natural gradient updates, and prioritized experience replay mechanisms to improve learning efficiency. Finally, experimental results demonstrate that the proposed CORAL algorithm improves average AEE by 5.37% and 5.94% over reflecting/transmitting-only RIS under varying data transmission times and data upload thresholds, respectively.