Intelligent 3D Trajectory Planning and Resource Allocation for Heterogeneous Multi-UAV Communications Using Deep Learning

Abstract

In this paper, we propose a deep learning (DL)-based framework for optimizing three-dimensional (3D) trajectory planning and resource allocation in unmanned aerial vehicle (UAV)-enabled wireless networks composed of heterogeneous UAVs with varying capabilities. Unlike conventional approaches that assume uniform UAV characteristics, we consider a realistic operational environment where UAVs differ in maximum transmit power, speed, and flight altitude, necessitating an adaptive load balancing strategy among UAVs for tasks. Our objective is to maximize the minimum average spectral efficiency (SE) among ground users (GUs) served by multiple heterogeneous UAVs, subject to practical constraints, including no-fly zones (NFZs), stochastic GU mobility, and UAV mobility limitations, such as maximum acceleration bounds. To efficiently solve the formulated optimization problem with low computational complexity, we propose a novel DL framework with specialized deep neural network (DNN) architectures. The framework employs an unsupervised learning-based training methodology that eliminates the need for labeled data and enables fast and adaptive trajectory planning and resource allocation, tailored to diverse UAV-specific capabilities. Through extensive performance evaluations, we demonstrate that the proposed scheme effectively handles UAV heterogeneity by adaptively balancing communication tasks, while ensuring NFZ compliance and adapting to the stochastic movement of GUs. Moreover, the simulation results demonstrate that the proposed framework achieves a higher minimum average SE than the ablation baselines and remains competitive with the per-instance optimization benchmark, while requiring only millisecond-level inference time, supporting its applicability to real-time UAV-based wireless networks.