PadLoc: Training-Free Floor-Plan Localization by Sparse Geometric Matching
PadLoc: Training-Free Floor-Plan Localization by Sparse Geometric Matching
This paper presents PadLoc, a training-free localization back-end that places a moving sensor within a known floor plan using sparse geometric matching. The method replaces dense pose grids with a small set of pose candidates defined by dominant wall orientations, masked normalized cross-correlation scores, and a Gaussian-mixture Bayes filter. Evaluated on the Bonn LiLocBench, PadLoc improves pooled SR@1m over UnLoc by 23-32 percentage points under wheel odometry and 35-43 percentage points with ground-truth motion. On the Hilti-Trimble Localization Challenge, it achieves 0.90-0.98 online SR@2m on three of five sessions using a camera-only fisheye rig, covering all five sessions at 0.67-0.92 when adding learned wall isolation.
PadLoc introduces a training-free localization back-end that replaces dense pose grids with sparse geometric matching, achieving superior accuracy on the Bonn LiLocBench and competitive performance on the Hilti-Trimble Localization Challenge while maintaining sensor modularity.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
hallenge PadLoc localizes a camera-only fisheye rig on construction floors hundreds of metres across at 0.90-0.98 online SR@2m on three of five sessions; adding learned wall isolation, applied to every session, covers all five at 0.67-0.92 with the localizer unchanged throughout. PadLoc thus matches or outperforms dense grid back-ends while preserving sensor modularity and a compact state that does not grow with traj
Open source S5
ength, and UnLoc's offline refinement from $T=35$ on; it exceeds Z-FLoc's published results with a shared MapAnything front-end. On Bonn LiLocBench, PadLoc improves pooled SR@1m over UnLoc by 23-32 pp under wheel odometry and 35-43 pp with ground-truth motion; with full-FOV LiDAR plus wheel odometry the same readout reaches 100% SR@1m at 9-10 cm median error on the static sequence. On the Hilti-Trimble Localization C
Open source S4
Source：arXiv Robotics — research abstracts · arxiv.org