Robust Robot Localization with Pole-centric Descriptors

This paper addresses the challenge of robust robot localization in urban environments, where the reliability of pole-like structures as landmarks is compromised by distance. It introduces a specialized evaluation framework using the Small Pole Landmark (SPL) dataset, which is a significant contribution. The comparative analysis of Contrastive Learning (CL) and Supervised Learning (SL) paradigms provides valuable insights into descriptor robustness, particularly in the 5-10m range. The work's focus on empirical evaluation and scalable methodology is crucial for advancing landmark distinctiveness in real-world scenarios.