Overview
- Niantic Spatial says its Large Geospatial Model was trained on roughly 30 billion geolocated images captured by players through optional Pokéstop and gym scans processed with anonymization.
- On March 10, the company announced a partnership with Coco Robotics to use its Visual Positioning System so delivery robots can navigate dense urban areas where GPS performs poorly.
- CTO Brian McClendon says the system can locate devices to within centimeters and determine a user’s viewing direction, enabling precise AR and navigation.
- Coverage describes the model’s ability to infer unseen parts of spaces by drawing on similarities across many scanned locations, strengthening performance in incomplete scans.
- Niantic reorganized in 2025 to create Niantic Spatial from its tech unit, and recent disclosures about the dataset and capabilities have intensified scrutiny over data reuse and potential law-enforcement or military applications.