Energy Efficiency Maximization for STAR-RIS and UAV-Assisted IUA: A Multi-Agent DRL Approach

Due to the ability to improve data transmission efficiency and extend coverage, simultaneously transmitting and reflecting reconfigurable intelligent surface (STAR-RIS) and autonomous aerial vehicle (AAV)-assisted Internet of unmanned agents (IUAs) has become an attractive paradigm. However, in multi-AAV and multi-STAR-RIS coexistence scenarios, device data transmission and energy harvesting (EH) processes are highly coupled with phase and amplitude optimization of STAR-RISs, which involve a large number of coupled variables that need to be carefully decoupled and processed. Therefore, to address the above challenges, we propose a distributed scheduling algorithm, MINT, for multi-STAR-RIS and multi-AAV assisted IUA, by jointly scheduling AAV trajectories, AAV association variable, charging time allocation, and STAR-RIS coefficient matrices to maximize system energy-efficiency. First, by considering device energy constraint, AAV energy consumption, and time constraints, we formulate the energy-efficiency maximization problem and model it as a Markov decision process. Second, we design a multiagent deep reinforcement learning-based MINT algorithm to solve the formulated optimization problem. Finally, experimental results demonstrate that MINT algorithm outperforms the existing algorithms regarding energy-efficiency, the number of uploaded bits, and convergence performance.