Fluid Antenna Enabled Low-Altitude Economy Networks via Deep Reinforcement Learning: An ISAC Design With Antenna Trajectory Optimization

Abstract

The rapid development of the low-altitude economy (LAE) has fostered various aerial applications such as cargo delivery and surveillance, while imposing growing demands on integrated sensing and communications (ISAC) services in dynamic environments with moving uncrewed aerial vehicles (UAVs). However, conventional ISAC systems employing fixed position antennas suffer from performance bottlenecks due to the limited spatial degrees of freedom. To address this issue, this paper investigates a fluid antenna (FA)-enabled LAE system with an ISAC design, where a base station (BS) leverages FAs to simultaneously provide communication services for multiple cargo UAVs and track an unauthorized UAV. We formulate a joint optimization of the ISAC beamforming, FA trajectories at the BS, and the flight speeds of the cargo UAVs, to maximize the system-wise task-aware utility of all cargo UAVs. This problem is a high-dimensional, constrained sequential decision-making process under environmental uncertainty. Given its complexity, we transform it into a constrained Markov decision process, and we propose an improved deep reinforcement learning algorithm, i.e., the deep deterministic policy gradient for the FA-enabled low-altitude sensing and communication (FLSCDDPG) algorithm, to tackle it. Specifically, a symmetry-based experience augmentation mechanism is designed to improve sample efficiency and accelerate training by exploiting the symmetries of homogeneous UAVs and FAs. Simulation results show that the FLSCDDPG algorithm improves both task-aware utility and learning efficiency while consistently satisfying sensing and delivery constraints across diverse scenarios. These gains reveal the benefits of jointly exploiting FA reconfigurability and experience augmentation, thereby enabling adaptive ISAC design in dynamic LAE networks.