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

Jianshi Wu; Minghang Zhu; Dunqiang Liu; Wen Li; Sheng Ao; Siqi Shen; Chenglu Wen; Cheng Wang

cs.CVApr 13, 2026

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

Authors: Jianshi Wu, Minghang Zhu, Dunqiang Liu, Wen Li, Sheng Ao, Siqi Shen, Chenglu Wen, Cheng Wang

Organizations: 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

Abstract

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.

Figures & tables

Figure 1: Mean position error comparisons on NCLT and Oxford dataset. Our method achieves superior relocalization accuracy on both datasets.

Figure 2: The pipeline of the proposed LEADER. Raw point clouds undergo spatial transformation to establish yaw-invariant spatial representations. Hierarchical feature extraction then derives multi-level fused features by cyclic convolution. The Regressor processes features from the Robust Projection-based Geometric Encoder (RPGE) to output predict coordinates and reliability values, followed by Cartesian Recovery. During training, the Truncated Relative Reliability (TRR) loss optimizes predictions against ground truth. During inference, high-reliability points are selected via TRR filtering for robust 6-DoF pose estimation through RANSAC-based estimator.

Panel of Figure 3 — Figure 3: Comparison of point cloud density distribution. Left: The point cloud is directly voxelized. Right: The point cloud is processed through the Spatial Transformation, voxelized, and then recovered via Cartesian Recovery.

Method		15-13-06-37	17-13-26-39	17-14-03-00	18-14-14-42	Average
APR	PointLoc [ 58 ]	10.75, 2.36	11.07, 2.21	11.53, 1.92	9.82, 2.07	10.79, 2.14
	PosePN++ [ 67 ]	4.54, 1.83	6.44, 1.78	4.89, 1.55	4.64, 1.61	5.13, 1.69
	HypLiLoc [ 57 ]	5.03, 1.46	4.31, 1.43	3.61, 1.11	2.61, 1.09	3.89, 1.27
	DiffLoc [ 36 ]	2.03, 1.04	1.78, 0.79	2.05, 0.83	1.56, 0.83	1.86, 0.87
SCR	SGLoc [ 37 ]	1.79, 1.67	1.81, 1.76	1.33, 1.59	1.19, 1.39	1.53, 1.60
	LiSA [ 62 ]	0.94, 1.10	1.17, 1.21	0.84, 1.15	0.85, 1.11	0.95, 1.14

Table 1: Mean position error (m) and mean orientation error (°) on the Quality-enhanced Oxford dataset . All values are given as (m, °). Lower values are better, with the best highlighted using bold and the second-best highlighted using underline. In both errors, LEADER achieved the best or second-best performance among all baselines.

Panel of Figure 4 — Figure 4: Visualization results of part of the methods on trajectory 17-13-26-39 in Quality-enhanced Oxford dataset. The black and red points represent the true and predicted results, respectively. The star indicates the starting point.

Method		2012-02-12	2012-02-19	2012-03-31	2012-05-26	Average
APR	PointLoc [ 58 ]	7.23, 4.88	6.31, 3.89	6.71, 4.32	10.02, 5.32	7.57, 4.60
	PosePN++ [ 67 ]	4.97, 3.75	3.68, 2.65	4.35, 3.38	9.59, 4.49	5.65, 3.57
	HypLiLoc [ 57 ]	1.71, 3.56	1.68, 2.69	1.52, 2.90	2.90, 3.47	1.95, 3.16
	DiffLoc [ 36 ]	0.99, 2.40	0.92, 2.14	0.98, 2.27	1.88 , 2.43	1.19 , 2.31
SCR	SGLoc [ 37 ]	1.20, 3.08	1.20, 3.05	1.12, 3.28	3.81, 4.74	1.83, 3.54
	LiSA [ 62 ]	0.97 , 2.23	0.91 , 2.09	0.87, 2.21	3.30, 2.84	1.51, 2.34

Table 2: Mean position error (m) and mean orientation error (°) on the NCLT dataset . All values are given as (m, °). Lower values are better, with the best highlighted using bold and the second-best highlighted using underline. In both errors, LEADER achieved the best performance among all baselines.

Panel of Figure 5 — Figure 5: Visualization results of part of the methods on trajectory 2012-05-26. The black and red points represent the true and predicted results, respectively. The star indicates the starting point.

Figure 6: All training trajectories and the 2012-05-26 test trajectory on NCLT dataset.

Figure 7: Cumulative distribution of position errors on NCLT dataset.

Method	<0.5m (%)	<1m (%)	99% Thresholds (m)
SGLoc	24.5	63.8	8.70
DiffLoc	30.8	68.7	5.73
LiSA	34.3	72.0	4.98
LightLoc	46.3	79.6	8.70
Ours	90.0	98.3	1.23

Table 3: Frame coverage comparison at critical error thresholds.

Method	ST	TRR-Train	TRR-Test	Normal Point Cloud	Yaw-Perturbed Point Cloud	Inference Time	Parameters
1	×	×	×	0.95, 1.94	8.04, 10.77	48	69.67 M
2	×	✓	×	0.89, 1.82	8.67, 10.44	48	69.67 M
3	×	✓	✓	0.88, 1.86	6.59, 10.14	48	69.67 M
4	✓	×	×	0.59, 2.12	0.59, 2.12	48	69.67 M
5	✓	✓	×	0.35, 1.80	0.35, 1.80	48	69.67 M
6	✓	✓	✓	0.31, 1.81	0.31, 1.81	48	69.67 M

Table 4: Ablation study evaluating the contributions of the RPGE and TRR modules on NCLT dataset. The ST column indicates the use of Spatial Transformation, while TRR-Train and TRR-Test denote the application of the TRR module during training and inference, respectively. Results for both normal and yaw-perturbed point clouds are reported as position/orientation errors (m, °), with inference time (ms) and parameters (M is million) included as an additional metric.

Figure 8: TRR Loss effect on Scene Point Error distribution on NCLT dataset 2012-05-26 trajectory.

Figure 9: Average Scene Point Error vs. Reliability Score Percentiles on NCLT dataset 2012-05-26 trajectory.

Test Condition	Error (m, °)
Baseline	0.31, 1.81
180° FOV	0.78, 3.51
Random Dropout (0 - 50%)	0.33, 1.86
Gaussian Noise ( $\sigma=0.05$ )	0.36, 1.93
Pitch/Roll Perturbation ( $\pm 10^{\circ}$ )	0.65, 3.17

Table 5: Robustness evaluation under various degraded sensing conditions on the NCLT dataset. All values represent the mean position and orientation errors.

Figure 10: Illustration of cylindrical projection and convolution correspondence. (a) Original point cloud

\mathcal{P}_{1}

in Cartesian coordinates. (b) Projected point cloud

\mathcal{P}_{2}

in cylindrical coordinates after Spatial Transformation. (c) The equivalent convolution operation in

\mathcal{P}_{1}

corresponding to the rectangular convolution kernel applied on

\mathcal{P}_{2}

, which forms a sector-shaped receptive field.

Figure 11: Overview of the proposed encoder architecture.

Figure 12: Illustration of the dilated cyclic sparse convolution used in stage 5 to enlarge the receptive field without reducing spatial resolution (right).

Yaw Angle	LEADER	LightLoc	DiffLoc
Original	0.31, 1.81	1.70, 2.95	1.19, 2.31
Yaw +180°	0.31, 1.81	21.70, 21.57	5.58, 36.36
Random Yaw	0.31, 1.81	30.51, 22.46	4.57, 76.50

Table 6: Localization results for the NCLT dataset at different yaw angles. All values are given as (m, °).

Method	Position Error	Orientation Error
Cylindrical projection	0.31 m	1.81°
Spherical projection	0.35 m (↑ 12.9%)	1.97°(↑ 8.8%)

Table 7: Performance comparison of coordinate systems

Figure 13: Visualization of Reliability scores. Points are color-coded by reliability (dark red: high, light/white: low).

Reliability Quartile	Error (m/°)
0-25% (Highest)	0.32, 1.96
25-50%	0.34, 1.97
50-75%	0.67, 2.33
75-100% (Lowest)	2.39, 3.63

Table 8: Mean position error (m) and mean orientation error (°) across reliability quartiles.

Metric	Position Error (m)	Orientation Error (°)
SGLoc (Original)	1.83	3.54
SGLoc + TRR	1.76 (↓ 3.83%)	2.61 (↓ 26.27%)

Table 9: Performance improvement by integrating TRR loss into SGLoc

Method	Error (m, °)
Mean Euclidean loss	0.59, 2.12
Matching loss	0.43, 1.97
TRR loss (Ours)	0.31, 1.81

Table 10: Performance comparison of different loss functions on NCLT dataset

Panel of Figure 14 — Figure 14: Trajectory visualization of the NCLT dataset.

Metric	LEADER	LEADER (xy)	RING/RING++
MPE (m)	0.31	0.28	16.34
MOE (°)	1.81	1.03	10.45
MedPE (m)	0.24	0.21	0.27
Recall@1	—	—	75.07%
Success@5m	99.72%	99.72%	92.79%

Table 11: Comparisons of different methods on the NCLT dataset.

Figure 15: Ground truth trajectory visualization of Quality-enhanced Oxford dataset.

Panel of Table 12 — Table 12: Visualization on Quality-enhanced Oxford Dataset, the star indicates the starting point.

Panel of Table 13 — Table 13: Visualzation on NCLT Dataset, the star indicates the starting point.

LiDAR Localization

LiDAR (light detection and ranging) localization estimates a robot or vehicle’s position and orientation from laser scans and maps.

See all 42 papers

Temporal-Aware Fusion for Robust Outdoor LiDAR Localization
Sep 29, 2026Minghang Zhu, Zhijing Wang, Yuxin Guo +5LiDAR Localization
Fujian Key Laboratory of Urban Intelligent Sensing and Computing, Xiamen University, Xiamen, China · School of Engineering Mathematics and Technology, University of Bristol, Bristol, United Kingdom
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
LEAR: Learning Edge-Aware Representations for Event-to-LiDAR Localization
Mar 2, 2026Kuangyi Chen, Jun Zhang, Yuxi Hu +2LiDAR Localization Camera-LiDAR Fusion
Institute of Visual Computing, Graz University of Technology, 8010 Graz, Austria · Neuromorphic Automation and Intelligence Lab (NAIL), School of Artificial Intelligence and Robotics, Hunan University, Changsha 410012, China

Sep 29, 2026J/K move · Enter open · S save

Temporal-Aware Fusion for Robust Outdoor LiDAR Localization

Minghang Zhu, Zhijing Wang, Yuxin Guo, Chen Liu+4

Fujian Key Laboratory of Urban Intelligent Sensing and Computing, Xiamen University, Xiamen, China · School of Engineering Mathematics and Technology, University of Bristol, Bristol, United Kingdom

LiDAR Localization

Open paper AskPDF arXiv

LiDAR relocalization aims to estimate the global 6-DoF pose of a sensor in the environment. However, existing regression-based approaches often encounter limitations in dynamic or ambiguous scenarios, as they typically prioritize single-frame inference, leaving the potential of spatio-temporal consistency across scans not fully explored. In this paper, we propose a Temporal-aware Localization framework (TempLoc) designed to enhance the robustness of outdoor localization by effectively modeling sequential consistency. Specifically, a Global Coordinate Estimation module is first introduced to predict point-wise global coordinates and associated uncertainties for each LiDAR scan. A Prior Coordinate Generation module is then presented to estimate inter-frame point correspondences by the attention mechanism. Lastly, an Uncertainty-Guided Coordinate Fusion module is deployed to integrate both predictions of point correspondence in an end-to-end fashion, yielding a more temporally consistent and accurate global 6-DoF pose. Experimental results on the NCLT and Oxford RobotCar benchmarks show that our TempLoc outperforms state-of-the-art methods by a large margin, demonstrating the effectiveness of temporal-aware correspondence modeling in LiDAR relocalization.

Abstract

LiDAR Localization

Explore similar work

Temporal-Aware Fusion for Robust Outdoor LiDAR Localization

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

LEAR: Learning Edge-Aware Representations for Event-to-LiDAR Localization