PrivateMap-Bench: A Privacy-Utility Benchmark for Indoor SLAM Map Sharing

Indoor robots often share maps, trajectories, and derived navigation artifacts. These artifacts can reveal private information even when raw video is not shared: maps can expose entrances, room layouts, and possible room functions, while repeated robot paths can reveal routines, frequently visited areas, and dwell zones. Most SLAM benchmarks focus on mapping accuracy, localization, and navigation, but they do not provide a reproducible way to measure both privacy leakage and task utility when robot maps are filtered and shared. We introduce PrivateMap-Bench, a TurtleBot4 benchmark for studying the privacy–utility trade-off in indoor SLAM map sharing. The benchmark treats map release as a filtering problem: it applies filters to maps and trajectories, then measures how much information remains recoverable and how useful the filtered artifacts remain for navigation. PrivateMap-Bench includes 15 real robot runs across 3 layout variants, producing 300 filtered artifacts, 2310 privacy rows, 5580 utility rows, and 1126 aggregate rows. It evaluates floor-plan recovery, room-structure recovery, trajectory leakage, routine leakage, map compactness, and A* navigation utility. The results show that perturbation alone does not guarantee privacy. Noise preserves most map structure, with an occupied-cell F1 of 0.991, while blur reduces floor-plan F1 to 0.650 and room-boundary skeleton IoU to 0.260. For trajectories, downsampling keeps routine order mostly intact, with similarity of 0.993 at a factor of two, whereas 1.0 m spatial quantization reduces visit-order similarity to 0.382. PrivateMap-Bench provides a practical protocol for auditing privacy leakage and utility loss in indoor robot map sharing, releasing code, configs, reviewed processed maps and trajectories, aggregate CSVs, figures, tables, and documentation while withholding raw ROS bags and unreviewed images by default.