M.R.F.A.M.E. Mars Robot For Autonomous Mapping and Exploration
Authors
Miron, Bogdan-Stefan ; Smau, Alexandru-Mihai
Term
4. semester
Education
Publication year
2023
Submitted on
2023-06-02
Abstract
This thesis investigates visual SLAM techniques for autonomous mapping and navigation of a Mars-analog rover in simulation. Motivated by communication delays that make teleoperation on Mars impractical, the work addresses the question of how a mobile robot can use its onboard sensors to first map and then navigate its environment. The project builds an exploration stack around an ESA ExoMy-based rover equipped with an Intel RealSense D435i and evaluates two state-of-the-art VSLAM pipelines—RTAB-Map and NVIDIA Elbrus—within NVIDIA Isaac Sim/Omniverse. Experiments are conducted in a simulated Lunar terrain provided by the University of Luxembourg and in a simulated warehouse environment from NVIDIA. The methodology includes a review of related work, integration of perception and SLAM components, and comparative testing of the algorithms in these scenarios. The results discuss each algorithm’s applicability, highlighting where they performed well and where limitations emerged, thereby offering guidance for selecting VSLAM approaches for planetary-style exploration in simulation. Findings are presented qualitatively, emphasizing applicability and trade-offs rather than detailed performance metrics.
Dette speciale undersøger visuelle SLAM-teknikker til autonom kortlægning og navigation for en Mars-analog rover i simulation. Med afsæt i de kommunikationsforsinkelser, der gør teleoperation på Mars upraktisk, adresserer arbejdet spørgsmålet: Hvordan kan en mobil robot bruge sine sensorer til først at kortlægge og derefter navigere i sit miljø? Projektet opbygger en explorationsstack omkring en rover baseret på ESA’s ExoMy og udstyret med en Intel RealSense D435i, og evaluerer to state-of-the-art VSLAM-løsninger—RTAB-Map og NVIDIAs Elbrus—i NVIDIA Isaac Sim/Omniverse. Eksperimenterne gennemføres i et simuleret måneterræn leveret af University of Luxembourg og i et simuleret lager fra NVIDIA. Metoden omfatter et litteraturstudie, integration af perceptions- og SLAM-komponenter samt komparative tests af algoritmerne i de nævnte scenarier. Resultaterne drøfter hver algoritmes anvendelighed ved at pege på, hvor de fungerede godt, og hvor der opstod begrænsninger, og giver dermed retningslinjer for valg af VSLAM-tilgange til planetarisk udforskning i simulation. Fundene præsenteres kvalitativt med fokus på anvendelighed og afvejninger frem for detaljerede målinger.
[This apstract has been generated with the help of AI directly from the project full text]