Navigation and Localization for Railway Inspection Drone in GPS-denied Environments: An Investigative Study of Modular SLAM Baselines and End-to-End Learning-based Approaches

Sammanfattning

Autonomous navigation in GPS-denied environments remains a critical challengefor unmanned aerial vehicles performing infrastructure inspection. This thesis investigatesthe feasibility of learning-based navigation for railway-inspection dronesby first constructing and analyzing a state-of-the-art modular baseline and thenevaluating emerging end-to-end paradigms.A high-performance navigation system combining FasterLIO SLAM with Fast-Plannertrajectory generation is implemented in high-fidelity simulation and used as ananalytical baseline. While the system successfully navigates dense forest environments,controlled experiments reveal three structural failure modes—SLAM localizationdrift, flight-controller tracking limitations, and planner-induced trajectoryconstraints—highlighting deeper challenges such as cumulative error propagationand real-time sensor–compute bottlenecks.Building on these insights, the thesis conducts an experimentally grounded feasibilitystudy of three dominant end-to-end learning directions: predictive worldmodelarchitectures, self-supervised representation learning pipelines, and visionreinforcement-learning approaches. By implementing prototype models and stresstestingtheir stability and data requirements, the study identifies several infeasibleor unstable directions—such as feature-forecasting models and self-distillation objectives—and reveals simulator limitations that currently block scalable vision-RLfor UAVs.Rather than delivering a complete end-to-end navigation system, this work providesa systematic evaluation of the landscape, clarifies the fundamental obstacles facinglearning-based navigation in GPS-denied environments, and establishes concretedesign requirements and a research roadmap for future PhD-level research.