Pseudo-Random UAV Test Generation Using Low-Fidelity Path Simulator
Anas Shrinah, Kerstin Eder
TL;DR
The paper addresses the high computational cost of UAV safety verification in high-fidelity simulators by proposing a pseudo-random test generator that uses a low-fidelity path simulator to estimate flight paths. It reuses the 3DVFH* planner to compute lookahead paths and pre-screens candidate tests by predicted minimum obstacle clearance, forwarding tests with potential safety violations to the HFS for final verification. Key contributions include a geometry-based obstacle-design arena for challenging trajectories, a concrete LFS pipeline using Open3D for depth-to-point-cloud generation, and an explicit motion model with $v_{max}=3~\mathrm{m}/\mathrm{s}$, $\omega_{max}=13.5^{\circ}/\mathrm{s}$, and $t_{plan}\approx 8~\mathrm{ms}$, plus a practical demonstration in SBFT 2025. These results indicate substantial resource savings and scalable, selective HFS validation for UAV safety testing, enabling faster iteration without sacrificing fidelity.
Abstract
Simulation-based testing provides a safe and cost-effective environment for verifying the safety of Uncrewed Aerial Vehicles (UAVs). However, simulation can be resource-consuming, especially when High-Fidelity Simulators (HFS) are used. To optimise simulation resources, we propose a pseudo-random test generator that uses a Low-Fidelity Simulator (LFS) to estimate UAV flight paths. This work simplifies the PX4 autopilot HFS to develop a LFS, which operates one order of magnitude faster than the HFS.Test cases predicted to cause safety violations in the LFS are subsequently validated using the HFS.