Student Projects

Map Quality Evaluation for Reliable Visual SLAM

*Background* Visual SLAM enables mobile robots to localize and build maps of previously unseen environments. In industrial deployments, map quality [1, 2] strongly affects long-term localization robustness and fleet-level performance. A high-quality map is necessary to ensure reliable performance over months of operation and across hundreds of robots. Today, there are no reliable methods to automatically assess a map quality, and manual inspections are still required. Developing a systematic and automated evaluation pipeline would greatly improve fleet monitoring, localization robustness map validation, and in general deepen the understanding of SLAM maps. *Objective* This master thesis aims to establish a principled understanding of what constitutes a high-quality SLAM map and to develop quantitative metrics that capture map reliability and uncertainty. We will design and implement an evaluation pipeline capable of assessing map quality along multiple dimensions, including: -- Global and local consistency: detecting large-scale drift / distortion, but also identify noisy area, with accumulation of outlier landmarks -- Localizability: measuring how effectively a robot can relocalize in different parts of the map. -- Temporal stability: analyzing how map quality evolves as the environment is updated over time. The final outcome will be an integrated map-quality evaluation module embedded within a state-of-the-art commercial SLAM system deployed across hundreds of robots. This module will support both offline map inspection and real-time monitoring. Show detailsaddremove