Transformer-based Monte Carlo Localization in Construction Meshes

Linus Kramer; William Talbot; Olga Vysotska; Marco Hutter

cs.ROSep 25, 2026

Transformer-based Monte Carlo Localization in Construction Meshes

Authors: Linus Kramer, William Talbot, Olga Vysotska, Marco Hutter

Organizations: Robotic Systems Laboratory, ETH Zurich, Zurich 8092, Switzerland

Abstract

To be able to perform inspection or digitization tasks, mobile robots on construction sites must be able to localize themselves reliably with respect to a global reference frame that is shared with a building map. Similar room layouts and low-texture surfaces pose a challenge for existing LiDAR- and vision-based localization methods. We approach this problem with a LiDAR-based global relocalization system that estimates the robot's pose relative to a building mesh and combines a PointNet++ encoder with a place recognition decoder, whose outputs serve as a learned observation model within a Monte Carlo Localization (MCL) framework. The pipeline is trained exclusively on synthetic LiDAR scans obtained by simulating the robot's sensors inside the building mesh. Our approach is robust in ambiguous environments due to an uncertainty-aware decoder that scales positional likelihoods and a resampling strategy that injects model hypotheses into the particle set, enabling recovery from potential particle depletion. Evaluations on real-world datasets show that our method outperforms both diffusion-based and ScanContext++ baselines while maintaining fast inference (18 ms per call), demonstrating the practicality of synthetic-data training for mesh-referenced global localization in construction robotics.

Figures & tables

Fig. 1: The robot’s LiDAR sensor captures a point cloud of the environment (purple points). Our trained model directly predicts the most likely positions in the building mesh that could have generated this observation. These candidate locations are visualized in green within the mesh. Figure created by Anna Bürgisser

Dataset	Model	Recall@ $4\text{\,}\mathrm{m}$ $\uparrow$		Recall@ $2\text{\,}\mathrm{m}$ $\uparrow$		Recall@ $1\text{\,}\mathrm{m}$ $\uparrow$		Recall@ $0.5\text{\,}\mathrm{m}$ $\uparrow$		Error [ $\text{\,}\mathrm{m}$ ] $\downarrow$
Dataset	Model	Cont.	Reinit	Cont.	Reinit	Cont.	Reinit	Cont.	Reinit	Cont.	Reinit
Hilti	ScanContext++	0.57	0.69	0.29	0.54	0.29	0.38	0.00	0.15	5.56	5.92
	Diffusion	0.88	0.78	0.75	0.33	0.43	0.22	0.18	0.11	2.46	3.02
	Transformer Decoder	0.93	0.89	0.92	0.67	0.49	0.44	0.13	0.11	1.54	2.57
	w/ Confidence Pred.	0.99	0.89	0.99	0.89	0.91	0.89	0.56	0.56	0.64	1.48
Apartment	ScanContext++	0.80	0.52	0.30	0.24	0.00	0.00	0.00	0.00	3.29	4.07

TABLE I: Monte Carlo Localization (MCL) results. Recall indicates the proportion of predictions within the specified radius. Two evaluation modes are compared: Continuous , where particles are initialized once and evaluated at every step, and Reinit , where particles are reinitialized every 10 steps and metrics are reported at step 10.

Panel of Figure 3 — Fig. 5: Example of the model recovering from mode collapse. Black points show the particles of the filter and the orange point marks the ground-truth position within the building mesh. From left to right: illustration of mode collapse, reseeding of the particles in resampling stage, further mcl steps.

Dataset	Model	Recall@ $4\text{\,}\mathrm{m}$ $\uparrow$	Recall@ $2\text{\,}\mathrm{m}$ $\uparrow$	Recall@ $1\text{\,}\mathrm{m}$ $\uparrow$	Recall@ $0.5\text{\,}\mathrm{m}$ $\uparrow$	Error [ $\text{\,}\mathrm{m}$ ] $\downarrow$
Hilti	Random	0.13	0.03	0.00	0.00	9.19
	ScanContext++	0.39	0.24	0.12	0.02	7.49
	Diffusion	0.70	0.49	0.23	0.05	4.15
	Transformer Decoder	0.58	0.43	0.25	0.07	5.00
	w/ Confidence Pred.	0.68	0.60	0.47	0.20	4.53
Apartment	Random	0.22	0.06	0.01	0.00	6.68

TABLE II: Place recognition results. Recall measures the proportion of predictions falling within the specified radius, evaluated with respect to the maximum-likelihood position.

Robot Localization

Robot localization estimates a robot’s position and orientation from onboard sensors, maps, and observations of its surroundings.

See all 177 papers

LEADER: Learning Reliable Local-to-Global Correspondences for LiDAR Relocalization
Apr 13, 2026Jianshi Wu, Minghang Zhu, Dunqiang Liu +5LiDAR Localization Point Cloud Learning
Fujian Key Laboratory of Urban Intelligent Sensing and Computing · Key Laboratory of Multimedia Trusted Perception and Efficient Computing, Ministry of Education of China, School of Informatics, Xiamen University, China · School of Engineering Mathematics and Technology, University of Bristol
Offline-Online Hierarchical 3D Global Relocalization With Synthetic LiDAR Sensing and Descriptor-Space Retrieval
May 8, 2026Jiahua Ren, Kai Shen, Muhua Zhang +1LiDAR Localization Map-Based Localization
School of Electrical Engineering, Southwest Jiaotong University, Chengdu 611756, China
VLM-GLoc: Vision-Language Model Enhanced Monte Carlo Localization for Robust Semantic Global Localization in Cluttered Quasi-Static Environments
May 28, 2026Shivendra Agrawal, Bradley HayesMap-Based Localization Robot Localization
University of Colorado Boulder · Boulder, Colorado, USA

Apr 13, 2026J/K move · Enter open · S save

LEADER: Learning Reliable Local-to-Global Correspondences for LiDAR Relocalization

Jianshi Wu, Minghang Zhu, Dunqiang Liu, Wen Li+4

Fujian Key Laboratory of Urban Intelligent Sensing and Computing · Key Laboratory of Multimedia Trusted Perception and Efficient Computing, Ministry of Education of China, School of Informatics, Xiamen University, China · School of Engineering Mathematics and Technology, University of Bristol

LiDAR Localization Point Cloud Learning Robust Regression

Open paper AskPDF arXiv

LiDAR relocalization has attracted increasing attention as it can deliver accurate 6-DoF pose estimation in complex 3D environments. Recent learning-based regression methods offer efficient solutions by directly predicting global poses without the need for explicit map storage. However, these methods often struggle in challenging scenes due to their equal treatment of all predicted points, which is vulnerable to noise and outliers. In this paper, we propose LEADER, a robust LiDAR-based relocalization framework enhanced by a simple, yet effective geometric encoder. Specifically, a Robust Projection-based Geometric Encoder architecture which captures multi-scale geometric features is first presented to enhance descriptiveness in geometric representation. A Truncated Relative Reliability loss is then formulated to model point-wise ambiguity and mitigate the influence of unreliable predictions. Extensive experiments on the Oxford RobotCar and NCLT datasets demonstrate that LEADER outperforms state-of-the-art methods, achieving 24.1% and 73.9% relative reductions in position error over existing techniques, respectively. The source code is released on https://github.com/JiansW/LEADER.

Abstract

Robot Localization

Explore similar work

LEADER: Learning Reliable Local-to-Global Correspondences for LiDAR Relocalization

Offline-Online Hierarchical 3D Global Relocalization With Synthetic LiDAR Sensing and Descriptor-Space Retrieval

VLM-GLoc: Vision-Language Model Enhanced Monte Carlo Localization for Robust Semantic Global Localization in Cluttered Quasi-Static Environments