Using Simulations to Develop Terrain-Relative Navigation for Mars-UAVs: Is it realistic enoughƒ

Download PDF
Open Webpage

Max Hofacker, Michael Schleiss, Thomas Pany, Roger Förstner

Published: 2025, Last Modified: 01 Mar 2026PLANS 2025EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: This paper presents the development of a robust Terrain-Relative Navigation (TRN) system for Mars Unmanned Aerial Vehicles (UAVs), integrating camera-based localization with a focus on the Extended Kalman Filter (EKF) for fault-tolerant performance in GNSS-denied environments. The navigation system’s performance is optimized using Bayesian Optimization (BO), enabling automatic tuning of filter parameters based on real and simulated datasets using a state of the art hyperparameter optimization software framework. The system is validated using the Gazebo-Classic simulation framework and real-world testing. A detailed analysis compares simulated and real sensor data, evaluating the accuracy of TRN updates and the effects of simulated Inertial Measurement Unit (IMU) noise. The results demonstrate the usability of Gazebo-Classic as a tool for TRN algorithm development, highlighting its limitations in replicating real-world navigation results and providing insights into advancing simulation environments for space exploration applications.