Heading-Coupled STAR-RIS Assisted Multi-UAV Networks for Semantic ISCC

Abstract

This letter proposes semantic integrated sensing, communication, and computation in a multi-UAV network whose rigidly mounted STAR-RIS service half-spaces rotate with UAV yaw. We couple exact no-sideslip mobility to panel normals, local user/target coordinates, and phase-aware side-exclusive channels, then jointly optimize mobility, STAR-RIS control, association, sensing, and three-tier DNN inference. The proposed graph-attention hierarchical semantic multi-agent optimization (GAT-HSMAO) uses GAT only in an outer parameter-shared multi-agent controller; a deterministic inner layer performs safety shielding and analytical resource allocation. Five-seed matched-holdout, trajectory, out-of-distribution, and runtime experiments quantify the resulting tradeoffs without assuming uniform dominance.