OpenNavMap: Structure-Free Topometric Mapping via
Large-Scale Collaborative Localization

Large-scale visual navigation relies on scalable, easy-to-maintain map representations, which are essential for enabling real-world robot deployments. While collaborative localization across multiple mapping sessions offers efficiency, conventional structure-based approaches face high maintenance costs and often fail in feature-less environments or under large viewpoint changes. To tackle these challenges, we introduce OpenNavMap: a lightweight, structure-free topometric mapping system that uses 3D geometric foundation models for on-demand reconstruction. Our method combines sequence matching, geometric verification, and robust optimization to align submaps precisely, all without needing pre-built 3D models. On the Map-Free benchmark, OpenNavMap outperforms structure-from-motion and regression baselines, achieving high accuracy in relocalization and map merging across a wide range of real-world scenarios. We additionally demonstrate reliable image-goal navigation with real and simulated robots.