Adaptive Task-Oriented Semantic Communication for UAVs With Joint Trajectory and Resource Optimization

Abstract

The transmission of large-volume sensed images during uncrewed aerial vehicle (UAV) inspections poses significant challenges due to the UAV’s limited bandwidth and energy constraints. This paper proposes a UAV-adaptive task-oriented semantic communication (UA-TSC) framework for low-altitude inspection tasks, which extracts task-relevant semantic features on the UAV and jointly optimizes flight trajectory and communication resource allocation to enhance transmission efficiency and task performance. We first develop a joint source-channel coding (JSCC) model for UAV semantic communication and formulate a unified optimization problem that maximizes a weighted sum of task execution accuracy, energy consumption, and trajectory reward under system constraints. To efficiently solve this problem, we decompose it into two coupled subproblems. The trajectory design subproblem is tackled using a Diffuser-based offline reinforcement learning algorithm, while the resource allocation subproblem is handled by a proximal policy optimization (PPO)-based algorithm. Simulation results demonstrate that the proposed semantic transmission scheme improves task execution accuracy by 15% over the local inference (LI) scheme and achieves a 5-fold improvement over the raw image offloading (RIO) scheme under low-SNR condition. Furthermore, the PPO-based resource allocation strategy consistently outperforms two baseline methods, while the learned trajectory optimization algorithm yields smoother paths and better adaptability across diverse environments.