Pith. sign in

REVIEW 5 major objections 5 minor 35 references

ROLO-SLAM: Rotation-Optimized LiDAR-Only SLAM in Uneven Terrain with Ground Vehicle

T0 review · 5 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read ROLO-SLAM claims a LiDAR-only pipeline can reduce vertical pose drift on uneven terrain by decoupling rotation from translation, and reports matching an IMU-fused system without the IMU.

desk verdict Genuine front-end decoupling idea and consistent gains on the authors' uneven-terrain datasets, but the abstract's broad superiority claim is conditional on near-constant vehicle motion — a limitation the authors themselves flag. read the letter →

arxiv 2501.02166 v1 pith:DCQLPDBA submitted 2025-01-04 cs.RO cs.CV

Yinchuan Wang , Bin Ren , Xiang Zhang , Pengyu Wang , Chaoqun Wang , Rui Song , Yibin Li , Max Q.-H. Meng This is my paper · ORCID
classification cs.ROcs.CV
keywords LiDARSLAMuneventerrainverticaldriftrotationoptimizationposeestimationgroundvehiclescanmatchingfactorgraph
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 attempts to establish that a LiDAR-only SLAM system can keep ground vehicles from drifting vertically on rough, uneven terrain, a failure mode that tilts and distorts the resulting maps. The proposed method, ROLO-SLAM, splits the front-end pose problem: first a forward location prediction roughly cancels the translation between consecutive scans, then rotation is estimated on its own by spherical alignment against voxelized Gaussian maps, and only afterward is translation refined with a continuous-time motion constraint. If the claim holds, ground vehicles could get vertical accuracy close to an IMU-assisted LiDAR system without needing the IMU, its calibration, or its failure modes. The paper's experiments on its own off-road and campus datasets report lower translation RMSE than LOAM, LeGO-LOAM, CT-ICP, and HDL-SLAM, with rotation accuracy comparable to or better than the IMU-fused FAST-LIO2 on several sequences.

What carries the argument

The load-bearing mechanism is the forward location prediction of Eq. (3), a constant-velocity extrapolation $$t_{k+1}=t_k+\frac{\tau_{k+1}-\tau_k}{\tau_k-\tau_{k-1}}(t_k-t_{k-1})$$ that coarsely removes the translation between consecutive scans. That coarse translation lets the front-end place the two sensor centers at a common origin, so rotation can be estimated by itself: each source point slides along its spherical surface into a voxelized Gaussian distribution of the target scan, with the angular residual converted to a distance via $\arcsin(\|d_i\|/\|\bar{p}'_k\|)$ and weighted by an SVD-regularized Mahalanobis covariance. Translation is then refined separately in a continuous-time optimization that reuses the same correspondences and penalizes deviation from uniform motion. The design works because the rotation estimate no longer absorbs translation error and the translation estimate no longer absorbs rotation error.

What would settle it

Run the method on an off-road course designed around stop-and-go motion with ground truth from surveyed markers; if vertical RMSE jumps sharply on segments where the per-scan translation departs from the constant-velocity prediction by more than about 20 percent, the decoupling mechanism is the bottleneck the paper acknowledges.

Watch

Extended reading notes

Core claim

ROLO-SLAM's central claim is that vertical drift on uneven terrain happens because translation and rotation are optimized together, so vehicle shaking from the ground surface breaks point correspondences and drags both estimates down. The remedy is to estimate the inter-scan translation first with a constant-velocity forward location prediction, which lets the front-end align the sensor centers of two consecutive scans and estimate rotation alone by sliding source points along spheres into voxelized Gaussian target distributions. Translation is then refined in a separate continuous-time optimization that combines geometric alignment with a uniform-motion penalty. With a scan-to-submap alignment and factor-graph back end, the paper reports that this pipeline lowers vertical drift enough that a single LiDAR can match or beat an IMU-fused system on several sequences while remaining real-time.

Load-bearing premise

Everything in the front-end depends on the vehicle keeping roughly constant velocity from one scan to the next, so the predicted translation is a good baseline; sudden acceleration, braking, or emergency maneuvers invalidate that baseline and vertical drift returns.

Editorial extensions

If this is right

  • Ground robots and vehicles could get vertical accuracy on slopes and rough ground from LiDAR alone, dropping the need for an IMU and its calibration.
  • The vertical-drift problem can be attacked in the front-end by decoupling rotation and translation, rather than by adding sensors or relying only on dense global optimization.
  • Because the front-end runs in about 100 ms per scan, the accuracy gain does not have to trade away real-time operation.
  • On datasets where an IMU-fused system such as FAST-LIO2 fails outright, a LiDAR-only system that does not depend on IMU initialization can still produce a trajectory.
  • More faithful elevation profiles and maps on off-road routes improve downstream planning, since the map no longer tilts or doubles over on slopes.

Reading between the lines

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

  • The dependence on constant-velocity prediction suggests the method's accuracy ceiling is set by controller aggressiveness; a natural next experiment is to measure vertical RMSE as a function of per-scan acceleration on the same terrain.
  • The spherical-alignment rotation step is sensor-agnostic; the same slide-points-on-a-sphere-into-Gaussian-voxels idea could be applied to radar or sonar scans whenever radial geometry is well-defined.
  • If decoupling is the true cause of the improvement, then adding an IMU to ROLO should yield little extra vertical accuracy on uneven terrain but improve robustness during sudden maneuvers; that contrast is testable with the released code.
  • A stricter stress test than the paper's is near-zero forward speed with pure rotation on rough ground, where the forward location prediction has almost no translation to cancel and residual drift should become visible.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. The paper proposes ROLO-SLAM, a LiDAR-only SLAM system for ground vehicles on uneven terrain. The front-end decouples rotation and translation estimation: a forward location prediction based on a constant-velocity assumption coarsely removes inter-scan translation, voxelized spherical-alignment rotation registration estimates orientation, and a continuous-time constraint refines translation. A keyframe-based scan-to-submap alignment and factor-graph loop closure form the back-end. Experiments on KITTI, two campus datasets, and three off-road datasets compare against LOAM, LeGO-LOAM, CT-ICP, HDL-SLAM, and FAST-LIO2, reporting lower translation and rotation RMSE on the authors' uneven-terrain datasets and similar localization performance to FAST-LIO2 without using an IMU.

Significance. If the results hold, the main contribution is a practical demonstration that explicitly separating rotation and translation in the front-end reduces vertical drift in LiDAR-only SLAM, while maintaining real-time performance and releasing open-source code. The paper provides a broad experimental comparison on multiple self-collected datasets, an ablation of the front-end registration against ICP, NDT, and point-to-plane ICP, and a comparison with a LiDAR-inertial method. The strength of the paper is this breadth of evaluation and the reproducibility offered by the public code and video. The main caveat is that the core decoupling is contingent on a constant-velocity assumption that is not stress-tested, and the quantitative claims rely on a reference trajectory generated by another SLAM system.

major comments (5)
  1. [Sec. IV-B, Eq. (3), Sec. VI] The forward location prediction in Eq. (3) assumes v_k ≈ v_{k-1} and δd_k ≈ δd_{k-1}, and this predicted translation is what allows sensor centers to be aligned before the rotation-only registration in Sec. IV-C. If the vehicle accelerates, brakes, or turns sharply, the prediction error is nonzero; the spherical-angle cost in Eq. (15) then attributes that translational residual to rotation, and the continuous-time constraint in Eqs. (19)-(21) penalizes the optimizer for correcting the erroneous translation. The authors acknowledge in Sec. VI that "The accuracy of ROLO-SLAM is now reliant on forward location prediction" and list emergency braking and congested/off-road rally driving as failure cases, but none of the datasets in Table I includes a controlled acceleration-braking protocol or aggressive maneuvers. The abstract's broad claim that ROLO "excels in pose estimation of ground vehicles" is therefore conditional on near-constant velocity. I request either added experiments that quantify degradation under acceleration and braking or a reformulated claim that explicitly limits the operating regime.
  2. [Sec. V-A, Tables II, IV, V] The ground truth for the off-road and campus datasets is generated by the multi-sensor SLAM method GLIO [33], as stated in Sec. V-A. Because GLIO is itself an estimator -- and because in off-road, tree-covered terrain the accuracy of its GNSS/IMU/LiDAR fusion is not independently certified -- the absolute RMSE values in Tables II, IV, and V are relative to GLIO's trajectory rather than to an independent reference. The paper should validate the reference trajectory, for example with surveyed ground-control points, RTK-GPS checkpoints where available, or elevation profiles compared against independent altimeter readings such as the landmarks used in Fig. 24. Without this, the magnitude of the reported improvements over the baselines cannot be separated from the accuracy of the chosen reference.
  3. [Tables II-V] The paper reports RMSE (translation (m)/rotation (deg)) without defining whether this is per-frame absolute pose error after trajectory alignment, a relative error over sub-trajectories, or a benchmark-specific metric. This definition is essential for reproducibility: for the KITTI rows, rotation RMSE values around 2.2-3.0 deg in Table II are much larger than the KITTI odometry benchmark's usual relative rotational error expressed in deg/100m, which suggests a different aggregation. Please specify the exact error metric, the alignment procedure used before computing RMSE, and, for the KITTI datasets, report the standard KITTI odometry metrics as well.
  4. [Eqs. (10)-(15)] The rotation objective replaces the spherical angle with the tangent-plane distance: Eq. (10) gives ∢ = arcsin(||d_i||/||p̄'_k||), and Eq. (11) writes ∝ ||d_i||, but the proportionality factor 1/||p̄'_k|| depends on the point's range and is omitted in Eq. (15). The paper should state explicitly that a small-angle approximation is being made and justify why omitting the range-dependent normalization does not bias the rotation estimate, especially for near-range points. Relatedly, Eq. (14) sets the two non-dominant eigenvalues to λ_max, discarding the axial covariance structure; the effect of this regularization on the cost landscape in Eq. (15) is not analyzed.
  5. [Tables II and V] The same nominal experiment appears with different numbers: ROLO on Offroad2 has rotation RMSE 0.035 deg in Table II and 0.043 deg in Table V, while all other entries for ROLO are identical between the tables. This discrepancy is unexplained and directly affects the comparison with FAST-LIO2 in Sec. V-F. Please correct the values or explain how the two tables were produced.
minor comments (5)
  1. [Fig. 15 caption] The caption contains an uninterpretable string of "/uni000000..." tokens, which appears to be an encoding artifact and should be replaced with the intended text.
  2. [Fig. 17 and Sec. V] The baseline algorithm is called "HDL-SLAM" throughout the text but is labeled "HDL-LOAM" in Fig. 17; please use one consistent name.
  3. [Eq. (16)] The optimization is written as argmin over "¬T", which appears to be a typographical artifact; the variable should be the transformation whose components are later combined in Eq. (22).
  4. [Sec. IV-E] The definition "F = Fe, Fp}" has a mismatched brace and should read "F = {Fe, Fp}".
  5. [Sec. V-C] The text says all ER and Et results are obtained by executing each method five times, but it does not identify the source of randomness across runs; please state what varies (initialization, thread scheduling, or other nondeterminism) so the robustness claim is reproducible.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: central ROLO-SLAM derivation is independent; one minor non-load-bearing self-citation yields score 2.

full rationale

ROLO-SLAM's central derivation chain is self-contained. The forward location prediction (Eq. 3) is a constant-velocity extrapolation using previous scan timestamps and pose history; it is not a parameter fitted to the pose errors it is later used to reduce. The rotation registration (Eqs. 9-15) and translation optimization (Eqs. 16-21) minimize geometric residuals against the LiDAR point cloud itself, and the outputs are validated against external baselines (LOAM, LeGO-LOAM, CT-ICP, HDL-SLAM, FAST-LIO2) and external ground truth generated by GLIO [33]. No predicted trajectory component is fed back as the ground truth, and no fitted coefficient is renamed as a prediction. The paper's own admission in Sec. VI that 'The accuracy of ROLO-SLAM is now reliant on forward location prediction' is a genuine limitation of robustness under acceleration/braking, but it is not circular: Eq. 3 is not defined in terms of the final R and t it is used to estimate, and the empirical comparison remains external. The only overlapping-author reference is [20] (prior IROS low-drift odometry work), cited as related work; ROLO does not import any uniqueness theorem, ansatz, or fitted result from it. No self-definitional step, fitted-input-as-prediction step, load-bearing self-citation, author-imported uniqueness, ansatz-via-citation, or renaming pattern is present. Score 2 reflects the minor non-load-bearing self-citation; the central claim has independent empirical content.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The method introduces no new physical entities, forces, or dimensions. Its dependence is on standard SLAM modules plus a constant-velocity motion assumption, several unspecified hyperparameters, and two mathematical approximations in the rotation cost. The constant-velocity assumption is the most fragile because the paper states the whole front-end decoupling relies on it.

free parameters (4)
  • lambda (continuous-time constraint weight) = not reported
    Appears in Eq. 16 as an adjustable weight for F_CT; value is not specified and must be hand-tuned.
  • Res (voxel resolution) = not reported
    Used in Eq. 4-5 to compute voxel indices; no value is given and it affects matching quality.
  • N+ (minimum points per voxel) = not reported
    Threshold in Alg. 2 line 5 controlling which voxels can form correspondences; not specified.
  • Keyframe interval and sliding-window size k = not reported
    Used in Sec. IV-E to select keyframes and build submaps; values are not reported.
assumptions (5)
  • domain assumption Constant velocity and constant per-scan translation between consecutive scans
    Invoked in Sec. IV-B and Eq. 3 to predict the next vehicle position. This is load-bearing and is explicitly acknowledged as a limitation in Sec. VI.
  • domain assumption Rigid connection is always maintained between vehicle and LiDAR
    Stated in Sec. III as the basis for deriving vehicle pose from LiDAR pose.
  • domain assumption Points in each voxel follow a Gaussian distribution with known mean and covariance
    Used in Sec. IV-C, Eq. 6-8, to model each voxel as a Gaussian; this is a standard but unproved modeling assumption.
  • ad hoc to paper Spherical angle can be replaced by tangent-plane distance for angles up to pi/2
    Eq. 10-11 transforms the angular metric to a distance metric; the approximation and its validity range are stated but not proved.
  • ad hoc to paper SVD regularization discards axial covariance information
    Eq. 14 regularizes the covariance by keeping only maximum eigenvalue information; this is a heuristic choice with no error analysis.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ROLO-SLAM: Rotation-Optimized LiDAR-Only SLAM in Uneven Terrain with Ground Vehicle." pith.science (2026). https://pith.science/paper/DCQLPDBA

@misc{pith2026250102166,
  author       = {Pith},
  title        = {Pith review of: ROLO-SLAM: Rotation-Optimized LiDAR-Only SLAM in Uneven Terrain with Ground Vehicle},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DCQLPDBA}},
  note         = {Machine review of arXiv:2501.02166}
}
read the original abstract

LiDAR-based SLAM is recognized as one effective method to offer localization guidance in rough environments. However, off-the-shelf LiDAR-based SLAM methods suffer from significant pose estimation drifts, particularly components relevant to the vertical direction, when passing to uneven terrains. This deficiency typically leads to a conspicuously distorted global map. In this article, a LiDAR-based SLAM method is presented to improve the accuracy of pose estimations for ground vehicles in rough terrains, which is termed Rotation-Optimized LiDAR-Only (ROLO) SLAM. The method exploits a forward location prediction to coarsely eliminate the location difference of consecutive scans, thereby enabling separate and accurate determination of the location and orientation at the front-end. Furthermore, we adopt a parallel-capable spatial voxelization for correspondence-matching. We develop a spherical alignment-guided rotation registration within each voxel to estimate the rotation of vehicle. By incorporating geometric alignment, we introduce the motion constraint into the optimization formulation to enhance the rapid and effective estimation of LiDAR's translation. Subsequently, we extract several keyframes to construct the submap and exploit an alignment from the current scan to the submap for precise pose estimation. Meanwhile, a global-scale factor graph is established to aid in the reduction of cumulative errors. In various scenes, diverse experiments have been conducted to evaluate our method. The results demonstrate that ROLO-SLAM excels in pose estimation of ground vehicles and outperforms existing state-of-the-art LiDAR SLAM frameworks.

Figures

Figures reproduced from arXiv: 2501.02166 by the authors.

Figure 1
Figure 1. Top figures show a real vehicle moving in an off-road scenario. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. A simple case of suffering correspondence problems on [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. A system pipeline of our ROLO-SLAM, incorporating front-end LiDAR odometry module and the back-end mapping module. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (20 more)
Figure 4
Figure 4. Figure 4: Snapshots of vehicle driving in uneven terrain. Here, the poses [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: Rotation alignment model. The green points are source points [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 7
Figure 7. Figure 7: Transformation of the spatial manifold representing the 3D [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]
Figure 8
Figure 8. Figure 8: Factor graph structure of ROLO-SLAM. Two types of factor, [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]
Figure 9
Figure 9. Figure 9: Ground vehicles with sensing platform in off-road and campus [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]
Figure 10
Figure 10. Figure 10: The trajectory estimation of ROLO and other odometry frameworks for KITTI-odometry dataset sequence 00, 05 and 08. (a) Offroad1 (b) Offroad2 (c) Offroad3 [PITH_FULL_IMAGE:figures/full_fig_p009_10.png]
Figure 11
Figure 11. Figure 11: The trajectory estimation of our method and alternative methods for the off-road datasets ( [PITH_FULL_IMAGE:figures/full_fig_p009_11.png]
Figure 12
Figure 12. Figure 12: The trajectory estimation of our method and alternative methods for the SDU campus datasets ( [PITH_FULL_IMAGE:figures/full_fig_p010_12.png]
Figure 13
Figure 13. Figure 13: E R and E t curves of ROLO in KITTI and Off-road datasets. The each dimension of E R and E t is separately depicted in subfigures. is further validated using SDU campus datasets. The results are shown in [PITH_FULL_IMAGE:figures/full_fig_p011_13.png]
Figure 14
Figure 14. Figure 14: Box plots of processing time per scan for all methods. Each dataset is divided into several pieces with respect to the data frames. [PITH_FULL_IMAGE:figures/full_fig_p013_14.png]
Figure 15
Figure 15. Figure 15: The overall processing time per scan for methods. Note that [PITH_FULL_IMAGE:figures/full_fig_p014_15.png]
Figure 16
Figure 16. Figure 16: Mapping result for Xinglong dataset. (a) shows the Google map, which indicates the trajectory and distance metrics. (b) displays overall point cloud map and (c) is captured at a specific scenario. In contrast, in the top figures of [PITH_FULL_IMAGE:figures/full_fig_p…
Figure 17
Figure 17. Figure 17: Mapping results of HDL-LOAM, LeGO-LOAM and ROLO in Qianfo dataset. In (b)-(d), the top figures show the overall maps while the bottom figures are captured at the side views. (a) (b) (c) (d) [PITH_FULL_IMAGE:figures/full_fig_p015_17.png]
Figure 18
Figure 18. Figure 18: Mapping results in real slope cases of Qianfo dataset. (a)-(b) show the real scenes with ground vehicles. (c)-(d) are the corresponding point cloud maps and the navigation trajectories of ROLO. right three snapshots show the real-world navigation scenes. The trajector…
Figure 19
Figure 19. Figure 19: Mapping results in Offroad2 dataset. (a) is the satellite map from Google Earth. (b) shows the point cloud and trajectory. In KITTI datasets, we compare our method and FAST￾LIO2 in KITTI seq. 05, 08, 09, where FAST-LIO2 works properly. The estimated trajectories and G…
Figure 20
Figure 20. Figure 20: Point cloud map aligned with satellite map of [PITH_FULL_IMAGE:figures/full_fig_p016_20.png]
Figure 21
Figure 21. Figure 21: Trajectory results of our method and FAST-LIO2 for the KITTI sequence 05, 08, 09. (a) Offroad1 (b) Offroad2 (c) Offroad3 [PITH_FULL_IMAGE:figures/full_fig_p017_21.png]
Figure 22
Figure 22. Figure 22: The trajectory estimation of our method and [PITH_FULL_IMAGE:figures/full_fig_p017_22.png]
Figure 23
Figure 23. Figure 23: The trajectory estimation of our method and [PITH_FULL_IMAGE:figures/full_fig_p017_23.png]
Figure 24
Figure 24. Figure 24: Mapping results on Xinglong dataset. (a) shows the satellite map from Google Earth, where light blue landmarks show absolute altitudes measured by a handheld altimeter. (b) shows the point cloud map, where the white, red and blue point clouds are generated by GT, ROLO…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

35 extracted references · 8 canonical work pages

  1. [20]

    Low-drift lidar-only odometry and mapping for ugvs in environments with non-level roads,

    X. Chen, Y . Wang, C. Wang, R. Song, and Y . Li, 20 “Low-drift lidar-only odometry and mapping for ugvs in environments with non-level roads,” in 2022 IEEE/RSJ International Conference on Intelligent Robots and Sys- tems (IROS) . IEEE, 2022, pp. 13 174–13 180, https: //doi.org/10.1109/IROS47612.2022.9982264

    arXiv 2022

  2. [33]

    Glio: Tightly-coupled gnss/lidar/imu integration for continuous and drift-free state estimation of intelligent vehicles in urban areas,

    X. Liu, W. Wen, and L.-T. Hsu, “Glio: Tightly-coupled gnss/lidar/imu integration for continuous and drift-free state estimation of intelligent vehicles in urban areas,” IEEE Transactions on Intelligent Vehicles , 2023, https: //doi.org/10.1109/TIV .2023.3323648

    arXiv 2023

  3. [1]

    Robust graph slam back-ends: A comparative analysis,

    Y . Latif, C. Cadena, and J. Neira, “Robust graph slam back-ends: A comparative analysis,” in 2014 IEEE/RSJ International Conference on Intelligent Robots and Sys- tems. IEEE, 2014, pp. 2683–2690, https://doi.org/10. 1109/IROS.2014.6942929

    arXiv 2014

  4. [2]

    Lego-loam: Lightweight and ground-optimized lidar odometry and mapping on vari- able terrain,

    T. Shan and B. Englot, “Lego-loam: Lightweight and ground-optimized lidar odometry and mapping on vari- able terrain,” in 2018 IEEE/RSJ International Con- ference on Intelligent Robots and Systems (IROS) . IEEE, 2018, pp. 4758–4765, https://doi.org/10.1109/ IROS.2018.8594299

    arXiv 2018

  5. [3]

    Gnd-lo: Ground decoupled 3d lidar odometry based on planar patches,

    A. Galeote-Luque, J.-R. Ruiz-Sarmiento, and J. Gonzalez-Jimenez, “Gnd-lo: Ground decoupled 3d lidar odometry based on planar patches,” IEEE Robotics and Automation Letters , 2023, https://doi.org/10.1109/LRA.2023.3313057

    arXiv 2023

  6. [4]

    ig-lio: An in- cremental gicp-based tightly-coupled lidar-inertial odom- etry,

    Z. Chen, Y . Xu, S. Yuan, and L. Xie, “ig-lio: An in- cremental gicp-based tightly-coupled lidar-inertial odom- etry,” IEEE Robotics and Automation Letters , 2024, https://doi.org/10.1109/LRA.2024.3349915

    arXiv 2024

  7. [7]

    Traversability analysis for autonomous driving in com- plex environment: A lidar-based terrain modeling ap- proach,

    H. Xue, H. Fu, L. Xiao, Y . Fan, D. Zhao, and B. Dai, “Traversability analysis for autonomous driving in com- plex environment: A lidar-based terrain modeling ap- proach,” Journal of Field Robotics , vol. 40, no. 7, pp. 1779–1803, 2023, https://doi.org/10.1002/rob.22209

    doi:10.1002/rob.22209 2023

  8. [8]

    Loam: Lidar odometry and mapping in real-time

    J. Zhang and S. Singh, “Loam: Lidar odometry and mapping in real-time.” in Robotics: Science and sys- tems, vol. 2, no. 9. Berkeley, CA, 2014, pp. 1–9, https://doi.org/10.15607/RSS.2014.X.007

    doi:10.15607/rss.2014.x.007 2014

Show all 35 references
  1. [9]

    F-loam: Fast lidar odometry and mapping,

    H. Wang, C. Wang, C.-L. Chen, and L. Xie, “F-loam: Fast lidar odometry and mapping,” in 2021 IEEE/RSJ International Conference on Intelligent Robots and Sys- tems (IROS). IEEE, 2021, pp. 4390–4396, https://doi. org/10.1109/IROS51168.2021.9636655

    2021

  2. [10]

    Loam livox: A fast, ro- bust, high-precision lidar odometry and mapping pack- age for lidars of small fov,

    J. Lin and F. Zhang, “Loam livox: A fast, ro- bust, high-precision lidar odometry and mapping pack- age for lidars of small fov,” in 2020 IEEE Interna- tional Conference on Robotics and Automation (ICRA) . IEEE, 2020, pp. 3126–3131, https://doi.org/10.1109/ ICRA40945.2020.9197440

    2020

  3. [11]

    Direct lidar odometry: Fast localization with dense point clouds,

    K. Chen, B. T. Lopez, A.-a. Agha-mohammadi, and A. Mehta, “Direct lidar odometry: Fast localization with dense point clouds,” IEEE Robotics and Automation Letters, vol. 7, no. 2, pp. 2000–2007, 2022, https://doi. org/10.1109/LRA.2022.3142739

  4. [12]

    Comparing icp variants on real-world data sets: Open- source library and experimental protocol,

    F. Pomerleau, F. Colas, R. Siegwart, and S. Magnenat, “Comparing icp variants on real-world data sets: Open- source library and experimental protocol,” Autonomous robots, vol. 34, pp. 133–148, 2013, https://doi.org/10. 1109/IM.2001.924446

    2013

  5. [13]

    Gicp-loam: Lidar odometry and mapping with voxelized generalized iterative closest point,

    X. Li, A. Zhang, H. Liu, C. Fan, C. Liu, and R. Zheng, “Gicp-loam: Lidar odometry and mapping with voxelized generalized iterative closest point,” in 2022 China Au- tomation Congress (CAC). IEEE, 2022, pp. 2103–2108, https://doi.org/10.1109/CAC57257.2022.10055138

    2022

  6. [14]

    Ndt-loam: A real-time lidar odometry and mapping with weighted ndt and lfa,

    S. Chen, H. Ma, C. Jiang, B. Zhou, W. Xue, Z. Xiao, and Q. Li, “Ndt-loam: A real-time lidar odometry and mapping with weighted ndt and lfa,” IEEE Sensors Journal, vol. 22, no. 4, pp. 3660–3671, 2021, https: //doi.org/10.1109/JSEN.2021.3135055

    2021

  7. [15]

    Ct-icp: Real-time elastic lidar odometry with loop closure,

    P. Dellenbach, J.-E. Deschaud, B. Jacquet, and F. Goulette, “Ct-icp: Real-time elastic lidar odometry with loop closure,” in 2022 International Conference on Robotics and Automation (ICRA) . IEEE, 2022, pp. 5580–5586, https://doi.org/10.1109/ICRA46639.2022.9811849

    2022

  8. [16]

    Fast and versatile feature-based lidar odometry via efficient local quadratic surface approximation,

    S. Choi, H.-W. Chae, Y . Jeung, S. Kim, K. Cho, and T.- w. Kim, “Fast and versatile feature-based lidar odometry via efficient local quadratic surface approximation,”IEEE Robotics and Automation Letters , vol. 8, no. 2, pp. 640– 647, 2023, https://doi.org/10.1109/LRA.2022.3227875

    2023

  9. [17]

    E-loam: Lidar odometry and mapping with expanded local structural information,

    H. Guo, J. Zhu, and Y . Chen, “E-loam: Lidar odometry and mapping with expanded local structural information,” IEEE Transactions on Intelligent Vehicles , vol. 8, no. 2, pp. 1911–1921, 2023, https://doi.org/10.1109/TIV .2022. 3151665

    1911 doi

  10. [18]

    A lidar slam with pca-based feature extraction and two-stage matching,

    S. Guo, Z. Rong, S. Wang, and Y . Wu, “A lidar slam with pca-based feature extraction and two-stage matching,” IEEE Transactions on Instrumentation and Measurement, vol. 71, pp. 1–11, 2022, https://doi.org/10.1109/TIM. 2022.3156982

    2022

  11. [19]

    Fevo-loam: Feature extraction and vertical optimized lidar odometry and mapping,

    Z. Wang, L. Yang, F. Gao, and L. Wang, “Fevo-loam: Feature extraction and vertical optimized lidar odometry and mapping,” IEEE Robotics and Automation Letters , vol. 7, no. 4, pp. 12 086–12 093, 2022, https://doi.org/10. 1109/LRA.2022.3201689

    2022

  12. [21]

    Teaser: Fast and certifiable point cloud registration,

    H. Yang, J. Shi, and L. Carlone, “Teaser: Fast and certifiable point cloud registration,” IEEE Transactions on Robotics , vol. 37, no. 2, pp. 314–333, 2020, https: //doi.org/10.1109/TRO.2020.3033695

    2020

  13. [22]

    Eil-slam: Depth-enhanced edge-based infrared-lidar slam,

    W. Chen, Y . Wang, H. Chen, and Y . Liu, “Eil-slam: Depth-enhanced edge-based infrared-lidar slam,” Journal of Field Robotics , vol. 39, no. 2, pp. 117–130, 2022, https://doi.org/10.1002/rob.22040

    2022 doi

  14. [23]

    D-liom: Tightly-coupled direct lidar-inertial odometry and map- ping,

    Z. Wang, L. Zhang, Y . Shen, and Y . Zhou, “D-liom: Tightly-coupled direct lidar-inertial odometry and map- ping,” IEEE Transactions on Multimedia , 2022, https: //doi.org/10.1109/TMM.2022.3168423

    2022

  15. [24]

    Lio-sam: Tightly-coupled lidar inertial odome- try via smoothing and mapping,

    T. Shan, B. Englot, D. Meyers, W. Wang, C. Ratti, and D. Rus, “Lio-sam: Tightly-coupled lidar inertial odome- try via smoothing and mapping,” in 2020 IEEE/RSJ in- ternational conference on intelligent robots and systems (IROS). IEEE, 2020, pp. 5135–5142, https://doi.org/10. 11...

    2020

  16. [25]

    Sdv-loam: Semi-direct visual-lidar odometry and map- ping,

    Z. Yuan, Q. Wang, K. Cheng, T. Hao, and X. Yang, “Sdv-loam: Semi-direct visual-lidar odometry and map- ping,” IEEE Transactions on Pattern Analysis and Ma- chine Intelligence , pp. 1–18, 2023, https://doi.org/10. 1109/TPAMI.2023.3262817

    2023

  17. [26]

    3dregnet: A deep neural network for 3d point registration,

    G. D. Pais, S. Ramalingam, V . M. Govindu, J. C. Nascimento, R. Chellappa, and P. Miraldo, “3dregnet: A deep neural network for 3d point registration,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2020, pp. 7193–7203, https://doi.org/1...

    2020

  18. [27]

    Lo-net: Deep real-time lidar odometry,

    Q. Li, S. Chen, C. Wang, X. Li, C. Wen, M. Cheng, and J. Li, “Lo-net: Deep real-time lidar odometry,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2019, pp. 8473–8482, https://doi.org/10.1109/CVPR.2019.00867

    2019

  19. [28]

    OverlapNet: Loop Closing for LiDAR-based SLAM,

    X. Chen, T. L ¨abe, A. Milioto, T. R ¨ohling, O. Vysotska, A. Haag, J. Behley, and C. Stachniss, “OverlapNet: Loop Closing for LiDAR-based SLAM,” in Proceedings of Robotics: Science and Systems (RSS) , 2020, https: //doi.org/10.15607/RSS.2020.XVI.009

    2020 doi

  20. [29]

    Suma++: Efficient lidar-based se- mantic slam,

    X. Chen, A. Milioto, E. Palazzolo, P. Giguere, J. Behley, and C. Stachniss, “Suma++: Efficient lidar-based se- mantic slam,” in 2019 IEEE/RSJ International Con- ference on Intelligent Robots and Systems (IROS) . IEEE, 2019, pp. 4530–4537, https://doi.org/10.1109/ IROS40897.201...

    2019

  21. [30]

    Nerf-loam: Neural implicit representation for large-scale incremental lidar odometry and mapping,

    J. Deng, Q. Wu, X. Chen, S. Xia, Z. Sun, G. Liu, W. Yu, and L. Pei, “Nerf-loam: Neural implicit representation for large-scale incremental lidar odometry and mapping,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , 2023, pp. 8218–8227, https://doi...

    2023

  22. [31]

    Slamesh: Real- time lidar simultaneous localization and meshing,

    J. Ruan, B. Li, Y . Wang, and Y . Sun, “Slamesh: Real- time lidar simultaneous localization and meshing,” in 2023 IEEE International Conference on Robotics and Automation (ICRA) , 2023, pp. 3546–3552, https://doi. org/10.1109/ICRA48891.2023.10161425

    2023

  23. [32]

    Are we ready for autonomous driving? the kitti vision benchmark suite,

    A. Geiger, P. Lenz, and R. Urtasun, “Are we ready for autonomous driving? the kitti vision benchmark suite,” in Conference on Computer Vision and Pattern Recogni- tion (CVPR) , 2012, https://doi.org/10.1109/CVPR.2012. 6248074

    2012 doi

  24. [34]

    Extending kalibr: Calibrating the extrinsics of multiple imus and of individual axes,

    J. Rehder, J. Nikolic, T. Schneider, T. Hinzmann, and R. Siegwart, “Extending kalibr: Calibrating the extrinsics of multiple imus and of individual axes,” in 2016 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2016, pp. 4304–4311, https://doi.org/ 10.11...

    2016

  25. [35]

    Autoware on board: En- abling autonomous vehicles with embedded systems,

    S. Kato, S. Tokunaga, Y . Maruyama, S. Maeda, M. Hirabayashi, Y . Kitsukawa, A. Monrroy, T. Ando, Y . Fujii, and T. Azumi, “Autoware on board: En- abling autonomous vehicles with embedded systems,” in 2018 ACM/IEEE 9th International Conference on Cyber- Physical Systems (ICCPS...

    2018

  26. [36]

    A portable three-dimensional lidar-based system for long-term and wide-area people behavior measurement,

    K. Koide, J. Miura, and E. Menegatti, “A portable three-dimensional lidar-based system for long-term and wide-area people behavior measurement,” International Journal of Advanced Robotic Systems , vol. 16, no. 2, p. 1729881419841532, 2019, https://doi.org/10.1177/ 1729881419841532

    2019

  27. [37]

    Fast-lio2: Fast direct lidar-inertial odometry,

    W. Xu, Y . Cai, D. He, J. Lin, and F. Zhang, “Fast-lio2: Fast direct lidar-inertial odometry,” IEEE Transactions on Robotics, vol. 38, no. 4, pp. 2053–2073, 2022, https: //doi.org/10.1109/TRO.2022.3141876

    2022

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.