Preprint
Article

This version is not peer-reviewed.

PRISM-Loc: Three LiDAR Localization Backends Behind One ROS 2 Contract, with Middleware-Free Estimator Cores

Submitted:

19 November 2026

Posted:

19 November 2026

You are already at the latest version

Abstract
Open-source LiDAR localization tools are organized by sensor suite. We describe PRISM-Loc, an open-source ROS 2 stack that puts three established estimator families behind one contract: laser2d, a likelihood-field Monte Carlo localizer; ndt3d, a planar NDT-MCL backend (Saarinen et al., 2013); and fusion3d, a 15-state error-state Kalman filter (ESKF) fusing LiDAR, IMU, and RTK-GNSS. All backends publish the same map→odom transform and pose with covariance and accept the same /initialpose. The filter mathematics live in Eigen-only C++17 cores tested without ROS, and laser2d relocalizes by branch-and-bound scan matching with multi-scan verification that can report a place as ambiguous. We claim no new estimator theory. Seeded synthetic experiments show a 0.105 m mean tracking error, 97 of 100 room relocalizations, verification cutting false accepts on repeated-structure maps from 21 to 1 of 240 queries, and an overconfident ESKF covariance. On public data we test only starts from a wrong initial pose. On NCLT, ndt3d with the ground grid and tempered likelihood converges from wrong initial headings where a 6-DoF scan-to-map NDT tracker does not (2 m and 45◦: 12 of 12 runs against 4 of 12; 2 m and 90◦: 12 against 0; 10 m and 90◦: 11 against 0), although the particle filter is given the initial spread and the converged tracker is more accurate (median RMSE 0.15 against 0.30 m). On the MIT Stata Center dataset, branch-and-bound relocalization finds the pose in 9 of 12 runs against 3 of 12 for nav2_amcl global localization, and laser2d converges faster than AMCL in the runs where it converges, but AMCL converges as often or more often from offset starts. AMCL with Nav2 defaults fails on the sparse NCLT 2D projection even without offset. fusion3d has no public evaluation. Samples are small, Stata odometry is synthesized, and no field-accuracy claim is made. Code: https://github.com/kjungmo/prism_loc.
Keywords: 
;  ;  ;  ;  ;  
Copyright: This open access article is published under a Creative Commons CC BY 4.0 license, which permit the free download, distribution, and reuse, provided that the author and preprint are cited in any reuse.