Multi-Objective ISAC for Low-Altitude Economy Based on Multi-Task Deep Reinforcement Learning With Mixture of Experts

This paper studies a low-altitude economy (LAE)-oriented terrestrial integrated sensing and communication (ISAC) system, where a ground base station (GBS) provides communication and navigation services for authorized unmanned aerial vehicles (UAVs), as well as senses the low-altitude airspace to monitor the unauthorized mobile target. Both the communication sum-rate and sensing signal-to-noise ratio are maximized through joint optimization of GBS's beamforming and authorized UAVs' trajectories, subject to constraints including UAV mission completion and collision avoidance, as well as the GBS's maximum transmit power. To address this problem, we propose an intelligent joint optimization scheme, termed Multi-task mixture Of Drl Experts (MODE), based on multi-task deep reinforcement learning (DRL) with mixture of experts (MoE). Specifically, the MODE scheme first introduces the deep deterministic policy gradient algorithm to handle all continuous-domain variables and incorporates an objective-preference weight to capture the trade-off between communication and sensing performance. Thereafter, a multi-task MoE architecture is developed as the neural network of MODE, which treats each optimization problem with a specific objective-preference weight as a distinct task and trains these tasks concurrently within shared gating and expert networks. As such, MODE can generalize well to unseen weights. To facilitate efficient learning of the multi-task MoE architecture, we further put forth a hybrid experience replay mechanism that utilizes multi-task experience sets for training; each set contains all experiences from a complete episode under a specific task. Simulation results show that MODE attains more Pareto-efficient solutions than other schemes while maintaining stronger robustness across various simulation setups. Besides, we demonstrate the generalization ability of MODE for unseen tasks. © 2026 IEEE.