arxiv: 2605.07741 · v1 · submitted 2026-05-08 · 💻 cs.RO

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

Jiahua Ren , Kai Shen , Muhua Zhang , Lei Ma This is my paper

Pith reviewed 2026-05-11 02:48 UTC · model grok-4.3

classification 💻 cs.RO

keywords global relocalizationLiDAR3D localizationhierarchical frameworkdescriptor retrievalsynthetic sensingpose estimationrobot navigation

0 comments

The pith

An offline-online hierarchy uses synthetic LiDAR descriptors to relocalize robots in large 3D spaces within seconds at 8 cm accuracy.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper targets the problem of slow global relocalization in expansive 3D environments, where exhaustive pose search creates high computational costs. It decouples this by precomputing candidate positions and geometric descriptors offline through simulated LiDAR scans inside a grid map, then retrieves a coarse match online before refining with point cloud registration. This produces precise 6-DoF estimates. Real-world tests show the approach delivers 3-second average times and 8 cm accuracy, yielding an order-of-magnitude efficiency boost over prior global methods while matching their precision levels.

Core claim

The offline-online hierarchical framework decouples the pose search space by generating candidate positions and their geometric descriptor indices offline via simulated LiDAR scans within the grid map, enabling online global retrieval of a coarse pose estimate that is then refined through point cloud registration to produce precise 6-DoF outputs. Experiments confirm average relocalization times of 3 seconds and localization accuracies of 8 cm in 3D settings, with an order-of-magnitude improvement in computational efficiency compared to existing global relocalization techniques at comparable accuracy.

What carries the argument

The offline generation of synthetic LiDAR scan descriptors from a grid map to support fast descriptor-space retrieval of candidate poses, followed by online registration for refinement.

If this is right

Robots operating in large 3D spaces can perform global relocalization frequently without prohibitive delays.
The separation of offline precomputation and online retrieval allows handling of massive pose spaces that defeat exhaustive search.
Localization accuracy stays at 8 cm while computational cost drops by roughly ten times relative to prior global methods.
The framework supports 6-DoF pose output by chaining descriptor retrieval with standard registration.
Prebuilt descriptor indices make repeated relocalizations in the same mapped area efficient after initial setup.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Shared descriptor databases across multiple robots could reduce per-robot setup time in team operations.
Periodic map updates would require regenerating the synthetic descriptor index to maintain retrieval reliability.
The method's efficiency gains suggest it could enable relocalization as a background service rather than a rare event.

Load-bearing premise

Synthetic LiDAR descriptors generated inside the grid map will reliably retrieve a candidate pose close enough to the true location for subsequent registration to converge successfully despite real sensor noise and map differences.

What would settle it

A series of real-world trials in which retrieved candidates lie more than 50 cm from the ground-truth pose on average, causing registration to fail or yield errors exceeding 20 cm.

Figures

Figures reproduced from arXiv: 2605.07741 by Jiahua Ren, Kai Shen, Lei Ma, Muhua Zhang.

**Figure 1.** Figure 1: Experimental UAV platform. A Livox MID360 LiDAR is mounted on top of the UAV for 3D perception, while an Intel NUC12 onboard computer handles real-time mapping and relocalization. difficult to obtain accurately, which limits the applicability of these local relocalization schemes. On the other hand, the data scale and computational burden in 3D environments are also major limiting factors. Whether using vi… view at source ↗

**Figure 2.** Figure 2: Overall pipeline of the proposed LiDAR-based global 3D relocalization framework. The system takes as input a prior global point cloud map and real-time LiDAR scans. In the offline processing stage (orange), valid robot poses are uniformly sampled in the prior 3D occupancy grid map, a virtual LiDAR model is used to generate synthetic scans, and a descriptor database is constructed from the resulting virtual… view at source ↗

**Figure 3.** Figure 3: Constraint-aware uniform sampling in a 3D occupancy grid map. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

**Figure 4.** Figure 4: Sampling on the 3D occupancy grid map with the same number of [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗

**Figure 5.** Figure 5: Pipeline of ray-casting-based scan synthesis and egocentric descriptor [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗

**Figure 6.** Figure 6: Two outdoor experimental scenarios and representative evaluation poses. (a) Wide corridor between buildings, approximately [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗

**Figure 8.** Figure 8: Qualitative comparison of Scan Context images produced by local [PITH_FULL_IMAGE:figures/full_fig_p006_8.png] view at source ↗

read the original abstract

3D global relocalization is one of the key capabilities for mobile robots in practical applications. However, in large scale spaces, existing methods often suffer from prolonged online relocalization time due to factors such as the massive pose search space and high computational overhead. To address these issues, this paper proposes an offline-online hierarchical framework that decouples the search space. In the offline phase, candidate positions and their corresponding geometric descriptor indices are generated in the map by simulating LiDAR scans within the grid map. In the online phase, a coarse pose estimate is first obtained via global retrieval, followed by point cloud registration to output precise 6-DoF pose estimates. Real-world experiments demonstrate that the proposed method achieves an average relocalization time of 3 s and an average localization accuracy of 8 cm in 3D environments. Compared with existing global relocalization methods, the proposed method achieves an order-of-magnitude improvement in computational efficiency while delivering comparable relocalization accuracy.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper gives a practical offline-online split for 3D relocalization by precomputing synthetic LiDAR descriptors from a grid map, but the reported 3 s / 8 cm numbers rest on unexamined retrieval success rates.

read the letter

The core contribution is an engineering refinement: generate candidate poses and descriptors offline by ray-casting synthetic LiDAR inside the existing grid map, then online retrieve a rough match by descriptor similarity and refine with registration. This decouples the expensive global search and produces the claimed order-of-magnitude speed-up while keeping accuracy comparable to prior methods. The approach is sensible for large environments where exhaustive online search is the bottleneck, and the offline phase re-uses the map already built for navigation, so no new sensors or mapping steps are required. That part is cleanly executed and directly addresses a real deployment pain point. The real-world averages of 3 seconds and 8 cm are the strongest selling point if they generalize. The main weakness is that the abstract supplies no retrieval recall, no breakdown of how often the descriptor step lands inside the registration basin, and no tests against sensor noise, beam divergence, or small map errors. Without those, the averages could reflect only the trials where retrieval succeeded by chance. The stress-test concern about synthetic-to-real mismatch therefore lands; the paper would be stronger with at least a table of retrieval success rate versus map size and a few failure cases. The method itself is not conceptually new, but the specific pipeline and the concrete timing numbers are worth checking. This is useful reading for anyone implementing large-scale robot localization who already has a grid map and wants faster global recovery. It is not a foundational advance, but the implementation is reproducible enough that a referee could ask for the missing robustness data and still find the work worth publishing after revision. I would send it to peer review.

Referee Report

2 major / 1 minor

Summary. The paper proposes an offline-online hierarchical framework for 3D global relocalization of mobile robots in large-scale environments. In the offline phase, synthetic LiDAR scans are simulated inside a grid map to precompute candidate positions and geometric descriptor indices. In the online phase, descriptor-space retrieval yields a coarse pose estimate that is refined via point cloud registration to produce precise 6-DoF localization. Real-world experiments are claimed to deliver an average relocalization time of 3 s and average accuracy of 8 cm, together with an order-of-magnitude improvement in computational efficiency relative to prior global relocalization methods.

Significance. If the reported performance numbers are shown to be robust across varied map sizes, sensor conditions, and failure modes, the hierarchical decoupling of search space via synthetic descriptors could meaningfully advance practical global relocalization for field robotics, where exhaustive online search remains prohibitive.

major comments (2)

[Abstract] Abstract: the central performance claims (3 s average time, 8 cm accuracy, order-of-magnitude efficiency gain) are stated without any accompanying information on map sizes, number of trials, baseline methods, error distributions, or failure cases. Because these quantitative assertions constitute the primary evidence for the method's utility, their lack of supporting experimental context is load-bearing and prevents evaluation of whether the averages generalize.
[Abstract] The method's success hinges on the offline synthetic LiDAR descriptors (generated inside the grid map) retrieving a candidate pose close enough for subsequent registration to converge. No retrieval recall rates, sensitivity tests to real-world LiDAR noise, beam effects, intensity variations, or grid-map discretization errors are reported; without these, it is impossible to determine whether the reported averages reflect reliable operation or only favorable trials.

minor comments (1)

[Abstract] The abstract would be clearer if it briefly named the specific geometric descriptor and the registration algorithm employed.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment below and indicate where revisions will be made to strengthen the manuscript.

read point-by-point responses

Referee: [Abstract] Abstract: the central performance claims (3 s average time, 8 cm accuracy, order-of-magnitude efficiency gain) are stated without any accompanying information on map sizes, number of trials, baseline methods, error distributions, or failure cases. Because these quantitative assertions constitute the primary evidence for the method's utility, their lack of supporting experimental context is load-bearing and prevents evaluation of whether the averages generalize.

Authors: We agree that the abstract, owing to length constraints, omits key experimental context that supports the reported averages. The full manuscript (Section IV) provides these details, including the sizes of the grid maps used, the number of independent trials, the baseline methods, and error statistics. In the revised version we will expand the abstract with a concise clause noting the scale of the tested environments and that results are averaged over multiple runs, thereby making the central claims more self-contained while remaining within abstract length limits. revision: yes
Referee: [Abstract] The method's success hinges on the offline synthetic LiDAR descriptors (generated inside the grid map) retrieving a candidate pose close enough for subsequent registration to converge. No retrieval recall rates, sensitivity tests to real-world LiDAR noise, beam effects, intensity variations, or grid-map discretization errors are reported; without these, it is impossible to determine whether the reported averages reflect reliable operation or only favorable trials.

Authors: The referee correctly highlights that explicit retrieval recall rates and sensitivity analyses to sensor noise, beam effects, intensity, and discretization are not reported. While the end-to-end real-world results demonstrate practical success, these intermediate metrics would better substantiate the reliability of the hierarchical retrieval step. We will add a dedicated paragraph and table in the revised manuscript reporting top-K retrieval recall on the collected datasets together with ablation results on synthetic-scan fidelity under added noise and discretization variations. revision: yes

Circularity Check

0 steps flagged

No circularity in empirical engineering framework

full rationale

The paper describes an offline-online hierarchical method that generates synthetic LiDAR descriptors in a grid map during an offline phase and performs descriptor retrieval plus registration online. No equations, derivations, parameter fittings, or uniqueness theorems appear in the abstract or summary. Claims rest on reported real-world experimental averages (3 s time, 8 cm accuracy) rather than any reduction of outputs to inputs by construction. The central premise is an engineering workflow whose success is evaluated externally via experiments, with no self-definitional steps, fitted predictions renamed as results, or load-bearing self-citations that collapse the argument.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central performance claims rest on the unstated premise that synthetic LiDAR data generated from a grid map will produce descriptors sufficiently similar to real sensor data for reliable coarse retrieval.

axioms (1)

domain assumption Synthetic LiDAR scans inside the grid map produce descriptors that match real-world scans closely enough for global retrieval to return a usable coarse pose.
The offline candidate generation step depends on this fidelity assumption to decouple the search space.

pith-pipeline@v0.9.0 · 5473 in / 1311 out tokens · 46513 ms · 2026-05-11T02:48:07.771349+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

38 extracted references · 38 canonical work pages · 1 internal anchor

[1]

Efficient and reliable lidar-based global localization of mobile robots using multiscale/resolution maps,

J. Meng, L. Wan, S. Wang, L. Jiang, G. Li, L. Wu, and Y . Xie, “Efficient and reliable lidar-based global localization of mobile robots using multiscale/resolution maps,”IEEE Transactions on Instrumentation and Measurement, vol. 70, pp. 1–15, 2021

work page 2021
[2]

Fast and accurate deep loop closing and relocalization for reliable lidar slam,

C. Shi, X. Chen, J. Xiao, B. Dai, and H. Lu, “Fast and accurate deep loop closing and relocalization for reliable lidar slam,”IEEE Transactions on Robotics, vol. 40, pp. 2620–2640, 2024

work page 2024
[3]

Global localization of a mobile robot using lidar and visual features,

Z. Su, X. Zhou, T. Cheng, H. Zhang, B. Xu, and W. Chen, “Global localization of a mobile robot using lidar and visual features,” in2017 IEEE international conference on robotics and biomimetics (ROBIO). IEEE, 2017, pp. 2377–2383

work page 2017
[4]

Fusednet: End-to-end mobile robot relocalization in dynamic large- scale scene,

F.-x. Chen, Y . Tang, C. Tai, X.-p. Liu, X. Wu, T. Zhang, and L. Zeng, “Fusednet: End-to-end mobile robot relocalization in dynamic large- scale scene,”IEEE Robotics and Automation Letters, vol. 9, no. 5, pp. 4099–4105, 2024

work page 2024
[5]

Go-icp: A globally optimal solution to 3d icp point-set registration,

J. Yang, H. Li, D. Campbell, and Y . Jia, “Go-icp: A globally optimal solution to 3d icp point-set registration,”IEEE transactions on pattern analysis and machine intelligence, vol. 38, no. 11, pp. 2241–2254, 2015

work page 2015
[6]

Scan context: Egocentric spatial descriptor for place recognition within 3d point cloud map,

G. Kim and A. Kim, “Scan context: Egocentric spatial descriptor for place recognition within 3d point cloud map,” in2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2018, pp. 4802–4809

work page 2018
[7]

Real-time rgb-d camera relocalization,

B. Glocker, S. Izadi, J. Shotton, and A. Criminisi, “Real-time rgb-d camera relocalization,” in2013 IEEE International Symposium on Mixed and Augmented Reality (ISMAR). IEEE, 2013, pp. 173–179

work page 2013
[8]

maplab: An open framework for research in visual-inertial mapping and localization,

T. Schneider, M. Dymczyk, M. Fehr, K. Egger, S. Lynen, I. Gilitschen- ski, and R. Siegwart, “maplab: An open framework for research in visual-inertial mapping and localization,”IEEE Robotics and Automa- tion Letters, vol. 3, no. 3, pp. 1418–1425, 2018

work page 2018
[9]

A 3d occupancy grid based relocalization method for under-vehicle inspection robot,

M. Zhang, L. Ma, K. Shen, and Y . Sun, “A 3d occupancy grid based relocalization method for under-vehicle inspection robot,” in2022 China Automation Congress (CAC). IEEE, 2022, pp. 6173–6178

work page 2022
[10]

Indoor lidar relocalization based on deep learning using a 3d model,

H. Zhao, D. Acharya, M. Tomko, and K. Khoshelham, “Indoor lidar relocalization based on deep learning using a 3d model,”The Inter- national Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, vol. 43, pp. 541–547, 2020

work page 2020
[11]

Improving map re-localization with deep ‘movable’objects segmenta- tion on 3d lidar point clouds,

V . Vaquero, K. Fischer, F. Moreno-Noguer, A. Sanfeliu, and S. Milz, “Improving map re-localization with deep ‘movable’objects segmenta- tion on 3d lidar point clouds,” in2019 IEEE Intelligent Transportation Systems Conference (ITSC). IEEE, 2019, pp. 942–949

work page 2019
[12]

Object recognition from local scale-invariant features,

D. G. Lowe, “Object recognition from local scale-invariant features,” in Proceedings of the seventh IEEE international conference on computer vision, vol. 2. Ieee, 1999, pp. 1150–1157

work page 1999
[13]

Orb: An efficient alternative to sift or surf,

E. Rublee, V . Rabaud, K. Konolige, and G. Bradski, “Orb: An efficient alternative to sift or surf,” in2011 International conference on computer vision. Ieee, 2011, pp. 2564–2571

work page 2011
[14]

Brief: Binary robust independent elementary features,

M. Calonder, V . Lepetit, C. Strecha, and P. Fua, “Brief: Binary robust independent elementary features,” inEuropean conference on computer vision. Springer, 2010, pp. 778–792

work page 2010
[15]

Superglue: Learning feature matching with graph neural networks,

P.-E. Sarlin, D. DeTone, T. Malisiewicz, and A. Rabinovich, “Superglue: Learning feature matching with graph neural networks,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 4938–4947

work page 2020
[16]

Global localization using distinctive visual features,

S. Se, D. Lowe, and J. Little, “Global localization using distinctive visual features,” inIEEE/RSJ international conference on intelligent robots and systems, vol. 1. IEEE, 2002, pp. 226–231

work page 2002
[17]

Vins-mono: A robust and versatile monocular visual-inertial state estimator,

T. Qin, P. Li, and S. Shen, “Vins-mono: A robust and versatile monocular visual-inertial state estimator,”IEEE transactions on robotics, vol. 34, no. 4, pp. 1004–1020, 2018

work page 2018
[18]

Global visual–inertial localization for autonomous vehicles with pre-built map,

Y . Hao, J. Liu, Y . Liu, X. Liu, Z. Meng, and F. Xing, “Global visual–inertial localization for autonomous vehicles with pre-built map,” Sensors, vol. 23, no. 9, p. 4510, 2023

work page 2023
[19]

Orb-slam3: An accurate open-source library for visual, visual– inertial, and multimap slam,

C. Campos, R. Elvira, J. J. G. Rodr ´ıguez, J. M. Montiel, and J. D. Tard´os, “Orb-slam3: An accurate open-source library for visual, visual– inertial, and multimap slam,”IEEE transactions on robotics, vol. 37, no. 6, pp. 1874–1890, 2021

work page 2021
[20]

Efficient & effective prioritized matching for large-scale image-based localization,

T. Sattler, B. Leibe, and L. Kobbelt, “Efficient & effective prioritized matching for large-scale image-based localization,”IEEE transactions on pattern analysis and machine intelligence, vol. 39, no. 9, pp. 1744– 1756, 2016

work page 2016
[21]

Global localization with object-level semantics and topology,

Y . Liu, Y . Petillot, D. Lane, and S. Wang, “Global localization with object-level semantics and topology,” in2019 International Conference on Robotics and Automation (ICRA). IEEE, 2019, pp. 4909–4915

work page 2019
[22]

Visual topological slam and global localization,

A. Angeli, S. Doncieux, J.-A. Meyer, and D. Filliat, “Visual topological slam and global localization,” in2009 IEEE International Conference on Robotics and Automation. IEEE, 2009, pp. 4300–4305

work page 2009
[23]

Mobile robot localization using color signboard,

M. Hashimoto, F. Oba, and T. Tomiie, “Mobile robot localization using color signboard,”Mechatronics, vol. 9, no. 6, pp. 633–656, 1999

work page 1999
[24]

Comparative analysis of slam algorithms for mechanical lidar and solid-state lidar,

B. Zhou, D. Xie, S. Chen, H. Mo, C. Li, and Q. Li, “Comparative analysis of slam algorithms for mechanical lidar and solid-state lidar,” IEEE Sensors Journal, vol. 23, no. 5, pp. 5325–5338, 2023

work page 2023
[25]

Sectionkey: 3-d semantic point cloud descriptor for place recognition,

S. Jin, Z. Wu, C. Zhao, J. Zhang, G. Peng, and D. Wang, “Sectionkey: 3-d semantic point cloud descriptor for place recognition,” in2022 IEEE/RSJ international conference on intelligent robots and systems (IROS). IEEE, 2022, pp. 9905–9910

work page 2022
[26]

Fast point feature histograms (fpfh) for 3d registration,

R. B. Rusu, N. Blodow, and M. Beetz, “Fast point feature histograms (fpfh) for 3d registration,” in2009 IEEE international conference on robotics and automation. IEEE, 2009, pp. 3212–3217

work page 2009
[27]

M2dp: A novel 3d point cloud descrip- tor and its application in loop closure detection,

L. He, X. Wang, and H. Zhang, “M2dp: A novel 3d point cloud descrip- tor and its application in loop closure detection,” in2016 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2016, pp. 231–237

work page 2016
[28]

Aligning point cloud views using persistent feature histograms,

R. B. Rusu, N. Blodow, Z. C. Marton, and M. Beetz, “Aligning point cloud views using persistent feature histograms,” in2008 IEEE/RSJ international conference on intelligent robots and systems. IEEE, 2008, pp. 3384–3391

work page 2008
[29]

Lidar iris for loop-closure detection,

Y . Wang, Z. Sun, C.-Z. Xu, S. E. Sarma, J. Yang, and H. Kong, “Lidar iris for loop-closure detection,” in2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2020, pp. 5769–5775

work page 2020
[30]

Lidar-based global localization using histogram of orientations of principal normals,

L. Luo, S.-Y . Cao, Z. Sheng, and H.-L. Shen, “Lidar-based global localization using histogram of orientations of principal normals,”IEEE Transactions on Intelligent Vehicles, vol. 7, no. 3, pp. 771–782, 2022

work page 2022
[31]

Lidar-based initial global localization using two- dimensional (2d) submap projection image (spi),

Y . Li and H. Li, “Lidar-based initial global localization using two- dimensional (2d) submap projection image (spi),” in2021 IEEE Inter- national Conference on Robotics and Automation (ICRA). IEEE, 2021, pp. 5063–5068. 9

work page 2021
[32]

The normal distributions transform: A new approach to laser scan matching,

P. Biber and W. Straßer, “The normal distributions transform: A new approach to laser scan matching,” inProceedings 2003 IEEE/RSJ Inter- national Conference on Intelligent Robots and Systems (IROS 2003)(Cat. No. 03CH37453), vol. 3. IEEE, 2003, pp. 2743–2748

work page 2003
[33]

Crtf-moeicp: A robust coarse-to-fine reflector-based lidar indoor positioning algorithm,

M. Yao, Z. Shao, Y . Su, D. Wei, F. Zhang, and L. Wang, “Crtf-moeicp: A robust coarse-to-fine reflector-based lidar indoor positioning algorithm,” Mechatronics, vol. 104, p. 103259, 2024

work page 2024
[34]

Bevplace++: Fast, robust, and lightweight lidar global localization for unmanned ground vehicles,

L. Luo, S.-Y . Cao, X. Li, J. Xu, R. Ai, Z. Yu, and X. Chen, “Bevplace++: Fast, robust, and lightweight lidar global localization for unmanned ground vehicles,”IEEE Transactions on Robotics, 2025

work page 2025
[35]

3d lidar-based global localization using siamese neural network,

H. Yin, Y . Wang, X. Ding, L. Tang, S. Huang, and R. Xiong, “3d lidar-based global localization using siamese neural network,”IEEE Transactions on Intelligent Transportation Systems, vol. 21, no. 4, pp. 1380–1392, 2019

work page 2019
[36]

Pointloc: Deep pose regressor for lidar point cloud localization,

W. Wang, B. Wang, P. Zhao, C. Chen, R. Clark, B. Yang, A. Markham, and N. Trigoni, “Pointloc: Deep pose regressor for lidar point cloud localization,”IEEE Sensors Journal, vol. 22, no. 1, pp. 959–968, 2021

work page 2021
[37]

Global visual localization in lidar-maps through shared 2d-3d embedding space,

D. Cattaneo, M. Vaghi, S. Fontana, A. L. Ballardini, and D. G. Sorrenti, “Global visual localization in lidar-maps through shared 2d-3d embedding space,” in2020 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2020, pp. 4365–4371

work page 2020
[38]

Tackling the Kidnapped Robot Problem via Sparse Feasible Hypothesis Sampling and Reliable Batched Multi-Stage Inference

M. Zhang, L. Ma, Y . Wu, K. Shen, D. Huang, and H. Leung, “Tackling the kidnapped robot problem via sparse feasible hypothesis sampling and reliable batched multi-stage inference,”arXiv preprint arXiv:2511.01219, 2025

work page internal anchor Pith review Pith/arXiv arXiv 2025