Low-Cost Gimbal-Less Target Localization for Expendable Fixed-Wing UAVs: Algorithm Design, HILS Validation, and Flight Test Evaluation

Abstract

Expendable fixed-wing uncrewed aerial vehicles (UAVs) require low-cost onboard perception and localization capabilities, but conventional gimbal-stabilized payloads remain too heavy and expensive for expendable operations. This paper presents a gimbal-less target localization method for such platforms and demonstrates it across the full development chain from algorithm design through Hardware-In-the-Loop Simulation (HILS) pre-validation to flight-test evaluation. The target geolocation pipeline combines YOLOv8n-based object detection with a geometric back-projection algorithm using a body-fixed forward-facing monocular camera. Because expendable UAVs must minimize both hardware and development costs, we also develop a HILS testbed for pre-flight evaluation. We use it together with flight tests to validate the proposed pipeline over an altitude range of 10–200 m on two embedded platforms: a GPU-based Jetson Orin Nano ( $\approx {\$}400$ ) and an NPU-based Raspberry Pi 5 (RPi 5) with Hailo-8 ( $\approx {\$}200$ ). Under ideal sensor conditions in HILS, the pipeline achieves a CEP50 of 1.9 m. In flight tests over an altitude range of 10–200 m, it achieves a CEP50 of 7.2 m. The flight-test error is primarily driven by a longitudinal bias along the flight direction, caused by oblique viewing geometry and terrain elevation mismatch. On the low-cost RPi 5 with Hailo-8, the pipeline runs in real-time at 22.08 frames per second. Overall, these results show that real-time target geolocation is feasible on low-cost expendable fixed-wing UAVs, with an accuracy appropriate for coarse-localization tasks such as search-and-rescue and reconnaissance.