Physics Informed Gaussian Process Radio Maps for Risk Aware UAV Trajectory Planning
Cellular-connected unmanned aerial vehicles (UAVs) flying at low altitude experience line-of-sight (LoS) interference from many cells, altitude-dependent shadowing and reception through the side lobes of down-tilted base station antennas, so safe operation requires an accurate three-dimensional (3D) radio map built from few measurements. This paper proposes SKYMAP, a framework that couples an antenna-aware, LoS-conditioned path-loss prior with a multicell Gaussian process residual field whose kernel encodes inter-site shadowing correlation and LoS- and altitude-dependent shadowing variance. The posterior map is converted into per-location probabilities that the signal-to-interference-plus-noise ratio falls below a threshold, and these outage probabilities drive two planners: a risk-weighted shortest-path planner and a new length-budgeted risk-aware planner (LB-RAP) that uses bisection to find the most outage-averse route whose length stays within a prescribed budget. In simulations over five statistically generated cities with the Third Generation Partnership Project aerial channel model, SKYMAP reduces the root-mean-square map error by 10.5 to 13.1% relative to physics-residual kriging and by 48 to 57% relative to ordinary kriging at sampling ratios of 0.5 to 5%. Its learned hyper-parameters recover the simulated channel statistics, and its predictive uncertainty is well calibrated. Risk-aware planning lowers the fraction of flight in outage from 37.3% on the shortest path to 2.5%, while LB-RAP met every length budget and reached 5.4% outage with only 7.6% extra path length. Limitations under model mismatch and for adaptive sampling are also reported.
Authors
- G. Arul Dalton
Institutions
- Saveetha University (IN)
Publication Details
- Journal
- Research Square
- Published
- 2026-11-20
- DOI
- https://doi.org/10.21203/rs.3.rs-11480747/v1
- Primary Topic
- UAV Applications and Optimization
- Type
- preprint