Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Efficient LiDAR Bundle Adjustment for Multi-Scan Alignment Utilizing Continuous-Time Trajectories

T0 review · 3 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Modeling each laser beam's measurement time lets bundle adjustment align 11,702 scans into one consistent map.

desk verdict Solid engineering contribution: continuous-time LiDAR bundle adjustment that scales to thousands of scans with 8 GB GPU memory, but the piecewise-linear motion model is untested under aggressive motion and no error bars are reported. read the letter →

arxiv 2412.11760 v1 pith:T4GOYDSE submitted 2024-12-16 cs.RO

Louis Wiesmann , Elias Marks , Saurabh Gupta , Tiziano Guadagnino , Jens Behley , Cyrill Stachniss This is my paper · ORCID
classification cs.RO
keywords LiDARbundleadjustmentcontinuous-timetrajectorymulti-scanalignmentpointcloudregistrationlarge-scalemappingmulti-sessionout-of-coreprocessingleastsquaresoptimization
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper is trying to establish that LiDAR bundle adjustment can be built on a continuous-time trajectory, so that every laser beam in every scan gets its own pose in the global optimization instead of treating each scan as a rigid body. A sympathetic reading: the claim is that this one change fixes two problems at once — motion distortion within a scan is modeled without an IMU or odometry, and large multi-scan, multi-session alignment becomes a single sparse least-squares problem. The paper argues the method achieves this efficiently, reporting sharper maps and lower absolute and relative trajectory errors than previous alignment approaches on its own car dataset and competitive results on a campus dataset, using modest GPU memory by keeping only a circular buffer of point clouds in memory. If true, this matters because precise global maps are the foundation for localization, surveying, monitoring, and digital twins, and odometry-based mapping drifts over long routes.

What carries the argument

The central object is the continuous-time trajectory with per-scan start and end poses connected by slerp for rotation and linear interpolation for translation (Eqs. 2–4). This gives every beam a pose indexed by its measurement time, so motion distortion within a scan is absorbed directly into the least-squares adjustment rather than pre-corrected by an external motion model. The key work it does is turning the map-alignment problem into a single sparse set of normal equations: each point-to-plane residual couples only the start and end poses of the two scans involved, the derivatives (Eqs. 7–14) distribute the correction between those four poses, and the resulting sparsity is what lets the method solve for tens of thousands of pose unknowns. Around this core, the voxel hash map makes correspondence search nearly constant-time per point and the circular buffer bounds memory.

What would settle it

A decisive test would record a sequence with known ground-truth trajectory while deliberately braking or accelerating hard within single scans (for example, sharp jerks on a robot arm or vehicle), then compare the optimized map against a static terrestrial reference; if wall thickness and trajectory error stay as low as in smooth motion, the piecewise-linear model is adequate, but if they grow, the missing in-scan acceleration term is the cause.

Watch

Extended reading notes

Core claim

On its own terms, the paper claims that jointly optimizing a continuous-time trajectory over all scans yields globally and locally consistent point cloud maps. Each scan's pose is represented by its start and end poses, and the pose at any beam's measurement time is obtained by spherical linear interpolation of the rotation and linear interpolation of the translation, with the end pose of one scan tied to the start pose of the next. Residuals are point-to-plane distances between corresponding points across overlapping scans, and the optimization adjusts the start and end poses of every scan at once, with each residual contributing Jacobians to two scans. To make this tractable for thousands of scans, the method randomly samples a fixed number of neighboring scans for correspondences, searches with a GPU voxel hash map, and streams scans through an out-of-core circular buffer. The experiments claim this produces the most accurate trajectory on the car dataset (0.90 m and 0.63 deg absolute error, 0.014 m and 0.055 deg relative error), competitive results on a campus dataset, a large reduction in inter-session alignment error, and a peak GPU memory of 8 GB on 11,702 scans.

Load-bearing premise

The method assumes that within each scan the sensor moves exactly along a straight-line path interpolated between the scan's start and end poses, so any unmodeled acceleration or speed change during a scan biases every corrected beam's position and corrupts the surface-matching correspondences the whole optimization depends on.

Editorial extensions

If this is right

  • Mapping systems can refine an odometry-plus-loop-closure or odometry-plus-GPS initial guess into a globally consistent map without needing a feature extractor or a specific scan pattern.
  • Multi-session datasets can be aligned jointly by optimizing both trajectories together, with only the time discontinuity between sessions handled separately, producing one unified map.
  • The per-beam pose model removes the need for a separate deskewing step, so scans recorded during motion are corrected inside the optimization.
  • Because correspondences are pruned by spatial radius and scans stream through a circular buffer, memory use grows with buffer size rather than total scan count, making very long sequences feasible on a single GPU.
  • The approach can be applied to both handheld and vehicle-mounted LiDAR sensors while remaining independent of the sensor's scan pattern.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • An inference the paper leaves implicit: the piecewise-linear motion model is the main accuracy ceiling; if a platform accelerates hard inside a single scan, the bias in every corrected beam would show up as thickening or ghosting of walls, so the method would likely benefit from an IMU term or a higher-order interpolation when applied to aggressive motion.
  • A natural testable extension is to replace the random choice of neighboring scans with a covisibility-based selection using the current map estimate, which should improve convergence on routes where overlap is uneven.
  • Since the paper reports the largest rotational gains, much of the benefit likely comes from correcting orientation drift; a useful ablation would freeze translation and optimize rotation only, to see whether the same map sharpening appears.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes a LiDAR bundle adjustment method that globally aligns large sets of LiDAR scans by optimizing a continuous-time trajectory. The trajectory is represented per scan by interpolating translation linearly and rotation with slerp between scan boundary poses, and the objective is a point-to-plane least-squares cost over correspondences found in a sampled subset of nearby scans. To scale to thousands of scans, the authors use randomized scan sampling within a radius, voxel-hash correspondence search, geometric subsampling, and an out-of-core circular buffer. Experiments on a self-recorded car dataset (IPB-Car) and the MCD dataset compare ATE/RPE against KISS-ICP, PIN-SLAM, HBA, and PBA, and a multi-session experiment aligns two MCD sessions.

Significance. If the reported results hold, the paper provides a scalable offline LiDAR bundle adjustment tool that gives competitive trajectory accuracy and map quality while using modest GPU memory (8 GB on 11,702 scans, versus 43 GB for PBA and 160 GB CPU memory for HBA). The use of TLS-based ground truth for the IPB-Car evaluation and the comparison with several recent baselines are strengths, as is the explicit treatment of motion distortion through a continuous-time formulation. The main novelty is the system-level integration of known components -- per-scan interpolation from CT-ICP, voxel hashing, scan sampling, and out-of-core buffering -- rather than a new theoretical formulation; the Jacobians and least-squares structure are standard. The stochastic correspondence sampling and the lack of error bars, together with the simplistic motion model, mean that the quantitative claims currently need additional support before they can be fully relied upon.

major comments (3)
  1. [Section III-A, Eqs. (2)-(4)] The continuous-time trajectory is a piecewise-linear model in translation and slerp in rotation with exactly one interval per scan, and the end pose of scan i is set equal to the start pose of scan i+1. This first-order motion model is load-bearing because Eq. (1) transforms every beam and the correspondences in Eq. (5) are computed on these corrected positions; under acceleration, braking, or motion during scan gaps, every residual is systematically biased, and the formulation provides no mechanism to absorb the unmodeled dynamics. The experiments only cover a car and a handheld device under mild motion. Please add a stress test with stronger dynamics, for example an accelerating platform or a handheld sequence with fast rotations, or explicitly qualify claim (i) to this piecewise-constant-velocity model class.
  2. [Section IV, Tables I-III and Eq. (15)] The correspondence search randomly samples Nmatches scans within a radius tau, but the paper reports single runs with no random seed, no repeated runs, and no error bars. Because different random samples can change the correspondence set and the objective landscape, the reported margins over HBA on IPB-Car (0.90 m vs 1.29 m ATE) and the MCD results cannot be assessed for statistical significance. Please report means and standard deviations over multiple runs, or fix and disclose the seed and provide evidence that the results are insensitive to the random sampling.
  3. [Section IV-A, runtime and memory paragraph] The paper's title and claim (ii) emphasize efficiency, but the only runtime information given is '45 min per iteration' and 'around two days' for the IPB-Car dataset, with no wall-clock time reported for HBA or PBA and no memory comparison on equal hardware. The memory advantage of the out-of-core buffer is clear, but the efficiency claim needs either a direct runtime comparison with the baselines or a more precise statement that the contribution is memory efficiency rather than overall speed.
minor comments (5)
  1. [Section III-A, Eqs. (2)-(3)] Equation (2) mixes indices: it should read R_{t_j} = slerp(R_{t_b(j)}, R_{t_e(j)}, alpha_j), and Eq. (3) should use alpha_j rather than alpha.
  2. [Section III-A, robust loss] The Geman-McClure kernel is mentioned but no formula or scale parameter is given; without this detail the experiments are not fully reproducible.
  3. [Section IV-A, evaluation protocol] The IPB-Car evaluation does not state how many LiDAR scans fall within the TLS reference locations or the time span covered by the reported ATE/RPE statistics; please add these details.
  4. [Section IV-D] The heading 'Mutli-Session Alignment' contains a typo and should read 'Multi-Session Alignment'.
  5. [Section IV-D, Table III] The multi-session experiment compares only against the initial guess; adding a state-of-the-art multi-session baseline would strengthen the claim that the method handles multi-session alignment.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the continuous-time LiDAR bundle adjustment is a direct least-squares fit evaluated against independent TLS ground truth, and no prediction reduces by construction to a fitted input.

full rationale

The paper's central derivation is a standard point-to-plane least-squares objective (Eqs. 5-6) over beam poses interpolated by slerp/linear interpolation (Eqs. 2-4). The optimized quantities are the scan boundary poses; the reported ATE/RPE numbers are computed against TLS-based reference poses generated independently of the optimization (Sec. IV-A), so the evaluation is not forced by the objective. The continuous-time trajectory is a modeling assumption rather than a result derived from the data; it does not define the target error metric in terms of itself. The method does rely on the authors' own KISS-ICP and loop-closure detector for initial guesses and as baselines, but those are external systems with their own published evaluations, and the BA refinement is not equivalent to re-fitting them: the objective, correspondence search, and evaluation are all separate from the initial guess. No equation in the paper defines a predicted quantity as a function of a fitted parameter of the same quantity, and no claimed 'prediction' is equal by construction to an input. The piecewise-linear motion model is a potential correctness limitation in dynamic scenarios, but it is an assumption, not a circular step. Therefore no circularity is found.

Assumptions & free parameters 9 free parameters · 7 assumptions · 0 invented entities

No new physical entities are introduced. The method rests on standard ICP closest-point assumptions, a linear/slerp scan-interpolation model, the continuity of poses across scans, and the accuracy of the external initial guess and TLS reference. The hand-chosen hyperparameters (Nmatches, tau, resolutions, buffer size, kernel scale) are tuned without ablation, and the missing kernel scale limits exact replication.

free parameters (9)
  • Nmatches = 10
    Number of neighboring scans sampled for correspondence search per scan (Sec III-B, Eq. 15). Chosen by hand; directly controls how many constraints each scan receives.
  • tau = 30 m
    Radius for selecting candidate scans for correspondence search (Sec III-B). If initial pose error exceeds tau for overlapping scans, true correspondences are not considered.
  • downsampling resolution = 15 cm
    Grid subsampling resolution for each scan before the adjustment (Sec IV). Affects density and runtime, chosen without ablation.
  • voxel hash map grid size = 30 cm
    Voxel size used in the hash map for nearest-neighbor search (Sec IV).
  • correspondence search neighborhood = 3x3x3 = 27 voxels
    Only adjacent voxels are checked for correspondences (Sec IV).
  • Nbuffer = 1000
    Size of the out-of-core circular buffer for point clouds (Sec III-C).
  • Niter = 100
    Maximum number of least-squares iterations (Sec IV); the authors note that 10 to 30 iterations usually suffice.
  • normal neighborhood = 30 nearest neighbors
    Number of nearest neighbors used to compute normals in each scan (Sec IV).
  • Geman-McClure robust loss scale = not reported
    A robust kernel is used to reduce outlier influence (Sec III-A), but its scale parameter is not specified, making exact replication impossible.
assumptions (7)
  • domain assumption For each sampled point, the closest point in the neighborhood of other scans is the true corresponding surface point.
    Used to define data associations in Eq. (5) and Sec III-A. This is the standard ICP assumption and fails for large initial errors, dynamic objects, or occlusions.
  • domain assumption Within each scan, the sensor pose at any beam time is exactly the slerp/linear interpolation between the scan's start and end poses (Eqs. 2-4).
    The continuous-time model has no acceleration or higher-order motion; aggressive motion within a scan would violate this and bias all corrected points.
  • domain assumption The end pose of one scan equals the start pose of the next scan.
    Stated explicitly in Sec III-A after Eq. (4). It ties scans into one trajectory but is false when scans are dropped or when there are time gaps.
  • domain assumption The provided initial poses are accurate enough that closest-point correspondences are mostly correct.
    Sec III-A requires a sufficiently accurate initial guess; experiments use KISS-ICP fusions or loop closures. If the initial guess is poor, the optimized result depends on its basin of convergence.
  • domain assumption Normals computed once from each raw scan's 30 nearest neighbors remain a valid local surface model after poses change.
    Sec IV precomputes normals and only rotates them per pose; neighborhoods are not recomputed during optimization, which can degrade for large pose changes.
  • ad hoc to paper Sampling only Nmatches scans within radius tau (Eq. 15) retains all correspondences needed for global consistency.
    This pruning is central to the complexity reduction. If two scans truly overlap but are not sampled, no constraint ties them, so correctness relies on the sampling and the radius.
  • domain assumption TLS-based reference poses are accurate enough to validate ATE and RPE at the reported magnitudes.
    Sec IV-A describes centimeter-accurate global TLS reference; the evaluation inherits any residual error in aligning scans to the TLS ground truth.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Efficient LiDAR Bundle Adjustment for Multi-Scan Alignment Utilizing Continuous-Time Trajectories." pith.science (2026). https://pith.science/paper/T4GOYDSE

@misc{pith2026241211760,
  author       = {Pith},
  title        = {Pith review of: Efficient LiDAR Bundle Adjustment for Multi-Scan Alignment Utilizing Continuous-Time Trajectories},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/T4GOYDSE}},
  note         = {Machine review of arXiv:2412.11760}
}
read the original abstract

Constructing precise global maps is a key task in robotics and is required for localization, surveying, monitoring, or constructing digital twins. To build accurate maps, data from mobile 3D LiDAR sensors is often used. Mapping requires correctly aligning the individual point clouds to each other to obtain a globally consistent map. In this paper, we investigate the problem of multi-scan alignment to obtain globally consistent point cloud maps. We propose a 3D LiDAR bundle adjustment approach to solve the global alignment problem and jointly optimize the available data. Utilizing a continuous-time trajectory allows us to consider the ego-motion of the LiDAR scanner while recording a single scan directly in the least squares adjustment. Furthermore, pruning the search space of correspondences and utilizing out-of-core circular buffer enables our approach to align thousands of point clouds efficiently. We successfully align point clouds recorded with a handheld LiDAR, as well as ones mounted on a vehicle, and are able to perform multi-session alignment.

Figures

Figures reproduced from arXiv: 2412.11760 by the authors.

Figure 1
Figure 1. We propose a LiDAR bundle adjustment approach for [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 3
Figure 3. The estimated trajectory of our approach on the self-recorded [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Shown are two close-up parts of the IPB-car dataset to allow qualitative evaluation. The initial guess for the methods was obtained [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figures from the paper (1 more)
Figure 5
Figure 5. Figure 5: For the multi-session alignment, we jointly optimize the [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. PINGS: Gaussian Splatting Meets Distance Fields within a Point-Based Implicit Neural Map

    cs.RO 2025-02 conditional novelty 6.0 of 10

    PINGS jointly builds a signed distance field and a Gaussian splatting radiance field in one point-based neural map, using geometric consistency to improve both.

Reference graph

Works this paper leans on

51 extracted references · 41 canonical work pages · cited by 1 Pith paper

  1. [1]

    Behley and C

    J. Behley and C. Stachniss. Efficient Surfel-Based SLAM using 3D Laser Range Data in Urban Environments. In Proc. of Robotics: Science and Systems (RSS) , 2018

    work page 2018

  2. [2]

    Benjemaa and F

    R. Benjemaa and F. Schmitt. A solution for the registration of multiple 3d point sets using unit quaternions. In Proc. of the Europ. Conf. on Computer Vision (ECCV) , 1998

    work page 1998

  3. [3]

    Bergevin, M

    R. Bergevin, M. Soucy, H. Gagnon, and D. Laurendeau. Towards a general multi-view registration technique. IEEE Trans. on Pattern Analysis and Machine Intelligence (TPAMI) , 18(5):540–547, 1996

    work page 1996

  4. [4]

    Besl and N

    P. Besl and N. McKay. A Method for Registration of 3D Shapes. IEEE Trans. on Pattern Analysis and Machine Intelligence (TPAMI) , 14(2):239–256, 1992

    1992

  5. [5]

    Biber and W

    P. Biber and W. Straßer. The normal distributions transform: A new approach to laser scan matching. In Proc. of the IEEE/RSJ Intl. Conf. on Intelligent Robots and Systems (IROS) , 2003

    work page 2003

  6. [6]

    Chen and G

    Y . Chen and G. Medioni. Object Modelling by Registration of Multiple Range Images. In Proc. of the IEEE/RSJ Intl. Conf. on Intelligent Robots and Systems (IROS) , 1991

    1991

  7. [7]

    J. Dai, L. Yan, H. Liu, C. Chen, and L. Huo. An offline coarse-to- fine precision optimization algorithm for 3d laser slam point cloud. Remote Sensing, 11(20):2352, 2019

    work page 2019

  8. [8]

    A. Das, J. Servos, and S. Waslander. 3D Scan Registration Using the Normal Distributions Transform with Ground Segmentation and Point Cloud Clustering. In Proc. of the IEEE Intl. Conf. on Robotics & Automation (ICRA), 2013

    work page 2013

Show all 51 references
  1. [9]

    Dellenbach, J

    P. Dellenbach, J. Deschaud, B. Jacquet, and F. Goulette. CT-ICP Real- Time Elastic LiDAR Odometry with Loop Closure. In Proc. of the IEEE Intl. Conf. on Robotics & Automation (ICRA) , 2022

    2022

  2. [10]

    Deschaud

    J. Deschaud. IMLS-SLAM: scan-to-model matching based on 3D data. In Proc. of the IEEE Intl. Conf. on Robotics & Automation (ICRA) , 2018

    2018

  3. [11]

    Di Giammarino, E

    L. Di Giammarino, E. Giacomini, L. Brizi, O. Salem, and G. Grisetti. Photometric LiDAR and RGB-D Bundle Adjustment. IEEE Robotics and Automation Letters (RA-L) , 8(7):4362–4369, 2023

    2023

  4. [12]

    Droeschel and S

    D. Droeschel and S. Behnke. Efficient Continuous-Time SLAM for 3D Lidar-Based Online Mapping. In Proc. of the IEEE Intl. Conf. on Robotics & Automation (ICRA) , 2018

    2018

  5. [13]

    Einhorn and H.M

    E. Einhorn and H.M. Gross. Generic ndt mapping in dynamic environments and its application for lifelong slam. Journal on Robotics and Autonomous Systems (RAS) , 69:28–39, 2015

    2015

  6. [14]

    Geiger, P

    A. Geiger, P. Lenz, and R. Urtasun. Are we ready for Autonomous Driving? The KITTI Vision Benchmark Suite. In Proc. of the IEEE Conf. on Computer Vision and Pattern Recognition (CVPR) , 2012

    2012

  7. [15]

    Gelfand, N.J

    N. Gelfand, N.J. Mitra, L.J. Guibas, and H. Pottmann. Robust Global Registration. In Proc. of the Symp. on Geometry Processing , 2005

    2005

  8. [16]

    Gupta, T

    S. Gupta, T. Guadagnino, B. Mersch, I. Vizzo, and C. Stachniss. Effectively Detecting Loop Closures using Point Cloud Density Maps. In Proc. of the IEEE Intl. Conf. on Robotics & Automation (ICRA) , 2024

    2024

  9. [17]

    Johnson and S.B

    A.E. Johnson and S.B. Kang. Registration and Integration of Textured 3D Data. Journal on Image and Vision Computing (IVC) , 17(2):135– 147, 1999

    1999

  10. [18]

    Kaess, A

    M. Kaess, A. Ranganathan, and F. Dellaert. iSAM: Incremental smoothing and mapping. IEEE Trans. on Robotics (TRO), 24(6):1365– 1378, 2008

    2008

  11. [19]

    Keller, D

    M. Keller, D. Lefloch, M. Lambers, and S. Izadi. Real-time 3D Reconstruction in Dynamic Scenes using Point-based Fusion. In Proc. of the Intl. Conf. on 3D Vision (3DV) , 2013

    2013

  12. [20]

    C. Kerl, J. Sturm, and D. Cremers. Robust Odometry Estimation for RGB-D Cameras. In Proc. of the IEEE Intl. Conf. on Robotics & Automation (ICRA), 2013

    2013

  13. [21]

    J. Li, T.M. Nguyen, S. Yuan, and L. Xie. Pss-ba: Lidar bun- dle adjustment with progressive spatial smoothing. arXiv preprint , arXiv:2403.06124, 2024

    2024 arXiv

  14. [22]

    X. Liu, Z. Liu, F. Kong, and F. Zhang. Large-scale lidar consistent mapping using hierarchical lidar bundle adjustment. IEEE Robotics and Automation Letters (RA-L) , 8(3):1523–1530, 2023

    2023

  15. [23]

    Liu and F

    Z. Liu and F. Zhang. Balm: Bundle adjustment for lidar mapping. IEEE Robotics and Automation Letters (RA-L), 6(2):3184–3191, 2021

    2021

  16. [24]

    W. Lu, Y . Zhou, G. Wan, S. Hou, and S. Song. L3-net: Towards learning based lidar localization for autonomous driving. In Proc. of the IEEE/CVF Conf. on Computer Vision and Pattern Recognition (CVPR), 2019

    2019

  17. [25]

    Newcombe, S

    R.A. Newcombe, S. Izadi, O. Hilliges, D. Molyneaux, D. Kim, A.J. Davison, P. Kohli, J. Shotton, S. Hodges, and A. Fitzgibbon. KinectFusion: Real-Time Dense Surface Mapping and Tracking. In Proc. of the Intl. Symp. on Mixed and Augmented Reality (ISMAR) , 2011

    2011

  18. [26]

    Nguyen, S

    T.M. Nguyen, S. Yuan, T.H. Nguyen, P. Yin, H. Cao, L. Xie, M. Woz- niak, P. Jensfelt, M. Thiel, J. Ziegenbein, and N. Blunder. MCD: Diverse Large-Scale Multi-Campus Dataset for Robot Perception. In Proc. of the IEEE/CVF Conf. on Computer Vision and Pattern Recognition (CVPR), 2024

    2024

  19. [27]

    Nießner, M

    M. Nießner, M. Zollh ¨ofer, S. Izadi, and M. Stamminger. Real-time 3D Reconstruction at Scale using V oxel Hashing. In Proc. of the SIGGRAPH Asia, 2013

    2013

  20. [28]

    Y . Pan, X. Zhong, L. Wiesmann, T. Posewsky, J. Behley, and C. Stachniss. Pin-slam: Lidar slam using a point-based implicit neural representation for achieving global map consistency. IEEE Trans. on Robotics (TRO), 2024

    2024

  21. [29]

    C. Park, P. Moghadam, S. Kim, A. Elfes, C. Fookes, and S. Sridharan. Elastic LiDAR Fusion: Dense Map-Centric Continuous-Time SLAM. In Proc. of the IEEE Intl. Conf. on Robotics & Automation (ICRA) , 2018

    2018

  22. [30]

    C. Park, P. Moghadam, J.L. Williams, S. Kim, S. Sridharan, and C. Fookes. Elasticity Meets Continuous-Time: Map-Centric Dense 3D LiDAR SLAM. IEEE Trans. on Robotics (TRO) , 38(2):978–997, 2022

    2022

  23. [31]

    J. Park, Q. Zhou, and V . Koltun. Colored Point Cloud Registration Revisited. In Proc. of the IEEE Intl. Conf. on Computer Vision (ICCV), 2017

    2017

  24. [32]

    Rusinkiewicz and M

    S. Rusinkiewicz and M. Levoy. Efficient variants of the ICP algorithm. In Proc. of Intl. Conf. on 3-D Digital Imaging and Modeling , 2001

    2001

  25. [33]

    Saarinen, T

    J. Saarinen, T. Stoyanov, H. Andreasson, and A. Lilienthal. Fast 3D Mapping in Highly Dynamic Environments Using Normal Dis- tributions Transform Occupancy Maps. In Proc. of the IEEE/RSJ Intl. Conf. on Intelligent Robots and Systems (IROS) , 2013

    2013

  26. [34]

    Schneider, F

    J. Schneider, F. Schindler, T. L ¨abe, and W. F¨orstner. Bundle adjustment for multi-camera systems with points at infinity. In ISPRS Annals of Photogrammetry, Remote Sensing and Spatial Information Sciences , 2012

    2012

  27. [35]

    Segal, D

    A. Segal, D. Haehnel, and S. Thrun. Generalized-ICP. In Proc. of Robotics: Science and Systems (RSS) , 2009

    2009

  28. [36]

    T. Shan, B. Englot, D. Meyers, W. Wang, C. Ratti, and D. Rus. LIO- SAM: Tightly-coupled Lidar Inertial Odometry via Smoothing and Mapping. In Proc. of the IEEE/RSJ Intl. Conf. on Intelligent Robots and Systems (IROS) , 2020

    2020

  29. [37]

    Shan and B

    T. Shan and B. Englot. LeGO-LOAM: Lightweight and Ground- Optimized Lidar Odometry and Mapping on Variable Terrain. In Proc. of the IEEE/RSJ Intl. Conf. on Intelligent Robots and Systems (IROS), 2018

    2018

  30. [38]

    Skuddis and N

    D. Skuddis and N. Haala. Dmsa–dense multi scan adjustment for lidar inertial odometry and global optimization. arXiv preprint , arXiv:2402.19044, 2024

    2024 arXiv

  31. [39]

    Steinbr ¨ucker, J

    F. Steinbr ¨ucker, J. Sturm, and D. Cremers. V olumetric 3D Mapping in Real-Time on a CPU. In Proc. of the IEEE Intl. Conf. on Robotics & Automation (ICRA) , 2014

    2014

  32. [40]

    Stoyanov, M

    T. Stoyanov, M. Magnusson, H. Andreasson, and A.J. Lilienthal. Fast and accurate scan registration through minimization of the distance between compact 3D NDT representations. Intl. Journal of Robotics Research (IJRR), 31(12):1377–1393, 2012

    2012

  33. [41]

    St ¨uckler and S

    J. St ¨uckler and S. Behnke. Multi-Resolution Surfel Maps for Efficient Dense 3D Modeling and Tracking. Journal of Visual Communication and Image Representation (JVCIR) , 25(1):137–147, 2014

    2014

  34. [42]

    Thrun and M

    S. Thrun and M. Montemerlo. The graph SLAM algorithm with applications to large-scale mapping of urban structures. Intl. Journal of Robotics Research (IJRR) , 25(5-6):403, 2006

    2006

  35. [43]

    Triggs, P.F

    B. Triggs, P.F. McLauchlan, R.I. Hartley, and A.W. Fitzgibbon. Bundle adjustment - a modern synthesis. In Proc. of the Intl. Workshop on Vision Algorithms: Theory and Practice , 1999

    1999

  36. [44]

    Vizzo, T

    I. Vizzo, T. Guadagnino, B. Mersch, L. Wiesmann, J. Behley, and C. Stachniss. KISS-ICP: In Defense of Point-to-Point ICP – Simple, Accurate, and Robust Registration If Done the Right Way. IEEE Robotics and Automation Letters (RA-L) , 8(2):1029–1036, 2023

    2023

  37. [45]

    H. Wang, C. Wang, C. Chen, and L. Xie. F-LOAM: Fast LiDAR Odometry and Mapping. In Proc. of the IEEE/RSJ Intl. Conf. on Intelligent Robots and Systems (IROS) , 2021

    2021

  38. [46]

    Y . Wang, N. Funk, M. Ramezani, S. Papatheodorou, M. Popovic, M. Camurri, S. Leutenegger, and M. Fallon. Elastic and Efficient LiDAR Reconstruction for Large-Scale Exploration Tasks. In Proc. of the IEEE Intl. Conf. on Robotics & Automation (ICRA) , 2021

    2021

  39. [47]

    Whelan, M

    T. Whelan, M. Kaess, H. Johannsson, M. Fallon, J.J. Leonard, and J. McDonald. Real-time large scale dense RGB-D SLAM with volumetric fusion. Intl. Journal of Robotics Research (IJRR) , 34(4- 5):598–626, 2014

    2014

  40. [48]

    Whelan, S

    T. Whelan, S. Leutenegger, R.S. Moreno, B. Glocker, and A. Davison. ElasticFusion: Dense SLAM Without A Pose Graph. In Proc. of Robotics: Science and Systems (RSS) , 2015

    2015

  41. [49]

    Y . Wu, T. Guadagnino, L. Wiesmann, L. Klingbeil, C. Stachniss, and H. Kuhlmann. LIO-EKF: High Frequency LiDAR-Inertial Odometry using Extended Kalman Filters. In Proc. of the IEEE Intl. Conf. on Robotics & Automation (ICRA) , 2024

    2024

  42. [50]

    W. Xu, Y . Cai, D. He, J. Lin, and F. Zhang. FAST-LIO2: Fast Direct LiDAR-Inertial Odometry. IEEE Trans. on Robotics (TRO) , 38(4):2053–2073, 2022

    2022

  43. [51]

    Zhang and S

    J. Zhang and S. Singh. LOAM: Lidar Odometry and Mapping in Real-time. In Proc. of Robotics: Science and Systems (RSS) , 2014

    2014

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.