GNN-Guided Surrogate-Accelerated Route Optimization for QoS- and QoE-Aware Urban Air Mobility

View at publisher Summarise with VIRAT
Authors: Yurika Maeda
Cited by: 0
References: 10

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

UAM route planning for aerial tourism must balance passenger experience, communication reliability, and community acceptability. These objectives often conflict and are computationally expensive to evaluate, as they depend on highresolution urban geometry and geospatial data. We formulate route design as a four-objective optimization problem under flight-time constraints. The objectives are: (1) maximizing along-route landmark-exposure reward as a QoE proxy, (2) maximizing non-residential overflight ratio as a noiseacceptability proxy, (3) maximizing the 10th-percentile RSRP as a QoS robustness proxy, and (4) minimizing route length. To reduce computational cost, we propose a GNN-guided NSGAIII framework in which a graph neural network predicts expensive objectives from a route graph during evolutionary search, and the final non-dominated set is re-evaluated using the original simulator. In a Yokohama waterfront case study with 24 POIs and open 3D city data, the surrogate achieves strong predictive accuracy, with coefficients of determination of 0.94, 0.92, and 0.88 for Scene, Quiet, and Comm, respectively, and reduces expensive simulation calls by approximately 89%. After true re-evaluation, the proposed approach produces a substantially larger true non-dominated set (92 solutions vs. 20 for the baseline) while maintaining comparable best-achieved objective values, enabling richer trade-off exploration for decision support in UAM route design.

Fields of Study & Subjects

Air Traffic Management and Optimization Advanced Multi-Objective Optimization Algorithms UAV Applications and Optimization

Publication details

Journal / source
International Conference on Artificial Intelligence in Information and Communication (ICAIIC)
Pages
584–589
Publication date
24 February 2026
Language
EN