Pith. sign in

REVIEW 4 major objections 6 minor 26 references

GaRLIO: Gravity enhanced Radar-LiDAR-Inertial Odometry

T0 review · 4 major / 6 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read GaRLIO uses radar Doppler velocity to estimate gravity directly, cutting vertical drift in LiDAR-inertial odometry on long downhill drives.

desk verdict A genuine method-level advance—using radar Doppler velocity as a gravity constraint in LiDAR-inertial odometry—with convincing ATE results, but the gravity-accuracy benchmark measures distance to the filter's own reference and needs reworking. read the letter →

arxiv 2502.07703 v2 pith:QBAI347L submitted 2025-02-11 cs.RO

Chiyun Noh , Wooseong Yang , Minwoo Jung , Sangwoo Jung , Ayoung Kim This is my paper · ORCID
classification cs.RO
keywords gravityestimationradar-LiDARfusionDopplervelocityverticaldriftinertialodometrydynamicobjectremoval4Dradarstate
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 argues that a LiDAR-inertial odometry system can sharply reduce vertical drift by using radar Doppler velocity measurements to estimate gravity directly, rather than inferring gravity by double-integrating IMU acceleration through pose estimates. The proposed system, GaRLIO, adds a velocity-aware gravity residual that constrains roll and pitch after a first stage fuses LiDAR and radar pointwise residuals. Because radar gives direct velocity, the gravity estimate avoids the bias amplification that the paper identifies in velocity-ignorant approaches. The authors report lower translation and rotation errors than three recent LiDAR-inertial odometry baselines on several public car, handcart, e-bike, and drone sequences, with the largest gains on long downhill paths where vertical drift dominates. If correct, the core insight is that any sensor providing direct velocity can play the same role in future inertial state estimators.

What carries the argument

The load-bearing object is a velocity-aware gravity residual. After a first Kalman update fuses LiDAR point-to-plane residuals and radar pointwise Doppler velocity residuals, the optimal velocity estimate feeds into an IMU preintegration identity (equation 13) that relates two consecutive radar-keyed velocities to acceleration bias and gravity. Solving for gravity and comparing it on the $S^{2}$ sphere with the initial gravity vector via cosine similarity yields a residual (equation 15) whose Jacobian enters a second-stage iterated Kalman update. The second stage therefore separately corrects roll and pitch, the degrees of freedom that drive vertical drift.

What would settle it

Take a dataset where the true gravity direction is known from an independent high-precision attitude reference, but initialize the estimator on a platform tilted by a few degrees (or with IMU bias) so the stationary gravity vector is wrong. If GaRLIO's estimated roll and pitch converge toward the biased initialization rather than toward the known true gravity, the central claim that the residual improves gravity estimation fails for biased initializations. A simpler check is to deliberately add a constant bias to the initial gravity vector and measure whether the z-error over a downhill sequence improves or degrades.

Watch

Extended reading notes

Core claim

GaRLIO's central claim is that world-frame gravity can be estimated online from a short interval of radar-corrected velocity plus IMU preintegration, and that constraining the state with the cosine-similarity residual between this predicted gravity and a stationarily initialized gravity vector removes vertical drift without requiring double integration. The paper derives a gravity residual on the $S^{2}$ manifold from equations 13-15, shows that adding pointwise radar Doppler velocity residuals improves gravity estimation accuracy relative to velocity-ignorant estimation, and reports that the full system reduces z-axis error substantially, for example 1.21 m mean vertical error over a 2.045 km path where baselines show 3.28-5.34 m. It also claims that radar-based dynamic-point removal improves robustness in scenes with moving objects. The paper states this is the first method to use radar velocity for gravity estimation.

Load-bearing premise

The method relies on the initial stationary IMU reading giving a world-frame gravity vector that is accurate enough to serve both as the reference in the gravity residual and as the ground truth when the paper evaluates gravity error; if that initialization is biased, the residual pulls roll and pitch toward the biased value, and the evaluation cannot see the mistake.

Editorial extensions

If this is right

  • Vertical drift on long downhill or elevated paths is reduced because the gravity residual corrects roll and pitch from radar-velocity-derived gravity rather than from double-integrated acceleration.
  • Platform vibration and acceleration bias affect gravity estimation less, since the velocity term comes from direct radar Doppler readings.
  • Dynamic scenes are handled in a single pass: radar flags moving objects, which are removed from the LiDAR sweep before geometric matching.
  • The reported gains grow with speed: faster car sequences show larger improvements from velocity residuals than slow handcart sequences.
  • Translation and rotation errors on the evaluated car, handcart, e-bike, and drone sequences are lower than the three compared LiDAR-inertial baselines, with the biggest margins on sequences prone to vertical drift.

Reading between the lines

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

  • One implication the paper leaves implicit is that the direct-velocity source need not be radar: wheel encoders, Doppler odometry, or other velocity-capable sensors could feed the same gravity residual in a visual-inertial or wheel-inertial estimator.
  • The two-stage update suggests gravity can be treated as a separate constraint layer, which might allow online correction of the initial gravity magnitude or frame alignment if the reference vector is later refined.
  • A testable extension would be to initialize on a tilted or sloped platform and check whether the residual corrects toward true gravity or toward the biased initialization; the paper's evaluation assumes the initialization is accurate.
  • Because the dynamic-removal projection works in the xy plane, future work could extend it to full 3D using radar point uncertainty to remove objects at height.
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

4 major / 6 minor

Summary. GaRLIO proposes a tightly-coupled radar-LiDAR-inertial odometry system that uses radar Doppler velocity for two purposes: computing pointwise velocity residuals in a Kalman filter, and constructing a velocity-aware gravity residual that constrains roll and pitch drift. The paper also uses radar detections to remove dynamic objects from LiDAR scans before the point-to-plane update. The method is evaluated on the NTU4DRadLM and Snail-Radar datasets against FAST-LIO2, Point-LIO, and DLIO, reporting improved ATE on several sequences (loop2, loop3, sl, iaef), and a separate gravity-deviation evaluation on additional sequences. The stated contribution is a first velocity-aware gravity estimator for LIO that mitigates vertical drift.

Significance. If the claims are validated, the work would be a useful contribution to sensor fusion and gravity-constrained odometry, and the open-source implementation is a strength. The proposed dynamic object removal using radar is a practical addition, and the reported trajectory improvements on public benchmarks are notable. However, the central gravity-accuracy evaluation contains a circularity: the same stationary-initialized gravity vector serves as both the reference in the gravity residual and as the ground truth for the gravity evaluation. The ATE results are not affected by this circularity, but the paper's headline claim of improved gravity estimation accuracy is not currently supported by independent evidence. Additionally, the gravity residual formula and its Jacobian are presented without sufficient derivation and contain apparent dimensional inconsistencies.

major comments (4)
  1. [§IV-D, Eq. (15) and §V-B] The gravity residual in Eq. (15) drives the estimated gravity toward Gg, the world-frame gravity obtained by stationary IMU initialization, and the same Gg is declared to be the ground truth for the gravity evaluation in Fig. 3 and Table I. Therefore the reported gravity deviation only measures convergence to the filter's own setpoint, not absolute gravity accuracy in the world frame. The abstract's claim to 'improve gravity estimation accuracy substantially' is not supported by the current evidence. Please provide an external gravity reference, for example a higher-grade IMU, a different estimation method, or evaluation on sequences with independently known attitude, or reframe the claim to focus on the demonstrated improvements in trajectory ATE.
  2. [§IV-D, Eq. (15)] The residual h_g = 1 - Gg^T Gbgr(xk, vg) is described as a cosine similarity, but Gbgr is not explicitly normalized in the equation. If Gbg is a R^3 vector whose norm can vary, the residual is not scale invariant and does not equal 1 - cos(theta). Also, the Jacobian Hg is written as -1/(||Gbgr|| Δt) [ (⌊Gg⌋× Gbv_{Ik+1})^T - Gg^T 0 0 0 ], which is not dimensionally consistent with a 1×15 block for the error state δbx; the non-zero blocks appear to be row vectors rather than 3×3 matrices, and the dependence of β on the IMU biases and the derivative of the normalization are not included. Please derive the residual and its Jacobian in full detail, either in the paper or in a supplement, and correct the formula.
  3. [§IV-B, §IV-C, and experiments] Several key filter parameters are not reported: the dynamic-removal distance threshold ε in Eq. (7), the velocity residual noise parameters R_v and α in Eq. (11), and the gravity residual noise R_g in Eq. (15). These directly affect both the proposed residuals and the ablation study, and without them the experiments are not reproducible even with the open-source code. Please state the exact values used for all datasets or clearly identify the configuration file in the released code.
  4. [§V-B, Table I] The reported improvement in gravity deviation from adding velocity measurements is very small (e.g., loop2 mean 0.035 vs 0.037 with standard deviations 0.053 and 0.055), and no statistical significance or confidence intervals are provided. Combined with the circular reference issue, this table does not convincingly demonstrate that velocity-aware gravity estimation is more accurate. Consider using an external accuracy metric, such as the vertical component of trajectory error over segments where a ground-truth height profile is available, or reporting repeated-run statistics.
minor comments (6)
  1. [§IV-D, Eq. (12-14)] The derivation of the predicted gravity Gbg is hard to follow: the preintegration term β in Eq. (12) and the velocity relationship in Eq. (13) should more clearly state the assumed IMU bias constancy and how the bias estimates enter the preintegration. A short derivation or a pointer to the VINS-Mono preintegration formulation would help.
  2. [§V-A] The comparison omits the radar-LiDAR-inertial odometry DR-LRIO due to unavailability of its code; this is reasonable, but the limitation should be stated in the experiments section rather than only in the comparison paragraph.
  3. [Fig. 3] The figure shows time series of gravity deviation but no quantitative metric in the caption; adding the RMSE or mean error for each method would make the reader's assessment easier.
  4. [§IV-C] The term 'pointwise velocity residual' is used for the scalar residual in Eq. (10); consider clarifying that each radar point contributes one scalar residual and that the state velocity is shared across points.
  5. [Eq. (11)] The Jacobian HRj is written with a block-diagonal-like notation; please ensure that the dimensions are consistent with the 15-dimensional error state and that each block is a 3×3 matrix.
  6. [Conclusion] The conclusion says the method is anticipated to 'advance robust SLAM systems based on UGV,' but the experiments include cars, handcarts, e-bikes, and a drone; the wording should reflect the broader platform coverage.

Circularity Check

1 steps flagged · score 2.0 of 10

Gravity accuracy is evaluated against the same stationary-initialized Gg used as the residual reference, so the gravity-improvement claim is partly an internal benchmark; ATE results remain externally grounded.

  1. self definitional [Section IV-D, Eq. (15) and Section V-B]
    "World frame gravity Gg is calculated through static initialization using IMU acceleration. ... Gravity residual is defined using cosine similarity, and its Jacobian is represented as follows: hg(xk, vg k) = 1 − Gg⊤Gbgr(xk, vg k) ... For gravity evaluation, we measured the deviation from the ground truth using the ⊟ operation on S 2 [16]. The global gravity ground truth is obtained through stationary initialization."

    The gravity residual in Eq. (15) explicitly drives the estimated gravity Gbg_r toward Gg, the same vector that Sec. V-B declares to be the gravity ground truth obtained by stationary initialization. As a result, the reported gravity deviation measures convergence to the filter's own setpoint; any bias in the stationary-initialized Gg appears identically in both the control target and the evaluation reference. The gravity-accuracy claim in Table I and Fig. 3 is therefore an internal consistency check rather than an absolute world-frame validation. This does not affect the ATE comparisons, which use external trajectory ground truth, and the velocity residual is derived from kinematics, so the circularity is limited to the gravity-evaluation benchmark.

full rationale

The paper's main trajectory claims are evaluated against external ground truth and are not circular: the radar Doppler velocity residual (Eq. 10) and the gravity residual (Eq. 15) are derived from standard kinematic and preintegration relationships, not fitted to the target benchmark. The one identifiable circular element is the gravity-evaluation protocol: the same stationary-initialized Gg serves both as the reference vector in the gravity residual and as the 'ground truth' for gravity deviation in Sec. V-B and Table I. This makes the reported gravity accuracy a measure of consistency with the initialized gravity, not an independent absolute check. However, this is a moderate internal-benchmark issue rather than a construction-level circularity: the velocity residual's benefit on ATE is empirically demonstrated against external trajectory ground truth, and the ablation results distinguish contributions of the residuals. No problematic self-citation chain or imported uniqueness theorem is present. Overall circularity score is 2.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new physical entities. It depends on standard sensor models and a stationary initialization assumption, plus several hand-set noise and threshold parameters that are not reported.

free parameters (3)
  • epsilon (dynamic removal distance threshold)
    Used in Eq. (7) to classify LiDAR points as dynamic if Mahalanobis distance d < epsilon. Value not reported in the paper.
  • velocity residual noise parameters R_v and alpha
    In Eq. (11), the measurement noise R_R^j is set to R_v + alpha * M_j for outliers. Values not reported.
  • gravity residual noise R_g
    Used in the second-stage Kalman update with the gravity residual. Value not reported.
assumptions (4)
  • domain assumption Global gravity vector Gg is known from stationary IMU initialization
    Used in Sec. IV-D as the reference for the gravity residual and in Sec. V-B as the evaluation ground truth. A biased initialization would bias both filter and evaluation.
  • domain assumption IMU preintegration model in Eqs. (12)-(13) correctly captures velocity evolution with only gravity and accelerometer bias
    The derivation ignores other forces such as vibrations and non-gravitational accelerations beyond bias. Used throughout Sec. IV-D.
  • domain assumption Radar static/dynamic classification via 3-Point RANSAC-LSQ and z-score weighting correctly separates static points
    Used for both dynamic removal (Sec. IV-B) and the velocity residual (Sec. IV-C). Failures would corrupt velocity updates.
  • standard math Invariant error-state Kalman filter framework from [16] is applicable
    Borrowed from prior work and assumed correct for this state representation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of GaRLIO: Gravity enhanced Radar-LiDAR-Inertial Odometry." pith.science (2026). https://pith.science/paper/QBAI347L

@misc{pith2026250207703,
  author       = {Pith},
  title        = {Pith review of: GaRLIO: Gravity enhanced Radar-LiDAR-Inertial Odometry},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QBAI347L}},
  note         = {Machine review of arXiv:2502.07703}
}
read the original abstract

Recently, gravity has been highlighted as a crucial constraint for state estimation to alleviate potential vertical drift. Existing online gravity estimation methods rely on pose estimation combined with IMU measurements, which is considered best practice when direct velocity measurements are unavailable. However, with radar sensors providing direct velocity data-a measurement not yet utilized for gravity estimation-we found a significant opportunity to improve gravity estimation accuracy substantially. GaRLIO, the proposed gravity-enhanced Radar-LiDAR-Inertial Odometry, can robustly predict gravity to reduce vertical drift while simultaneously enhancing state estimation performance using pointwise velocity measurements. Furthermore, GaRLIO ensures robustness in dynamic environments by utilizing radar to remove dynamic objects from LiDAR point clouds. Our method is validated through experiments in various environments prone to vertical drift, demonstrating superior performance compared to traditional LiDAR-Inertial Odometry methods. We make our source code publicly available to encourage further research and development. https://github.com/ChiyunNoh/GaRLIO

Figures

Figures reproduced from arXiv: 2502.07703 by the authors.

Figure 1
Figure 1. Top: Trajectories of GaRLIO and other methods with ground truth (white) overlaid on the terrestrial laser scanner (TLS) map. Bottom: Elevation plot along path length. Our method (blue) reported only 1.21 m vertical mean error over 2.045 km path length. Since traditional LIO lacks direct velocity observations at the raw measurement level, they rely on pose estimation and IMU measurements to estimate gravity. Yet, we … view at source ↗
Figure 2
Figure 2. GaRLIO is divided into four primary modules. Each module contributes to achieving the optimal state by removing LiDAR dynamic points and calculating both pointwise [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Gravity estimation using the velocity-ignorant (yellow) and velocity-aware [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (2 more)
Figure 5
Figure 5. Figure 5: Estimated trajectory for the iaef sequence and GaRLIO demonstrated superior performance over such a long sequence. segments, GaRLIO achieves accurate estimation results in such challenging scenarios. This exhibits the robustness of our velocity-aware gravity estimation…
Figure 6
Figure 6. Figure 6: z, pitch, and roll error in sl. The RMSE values for z, pitch, and roll are above each plot. Our method shows the lowest error along the baselines. ● : filtered dynamic point ■ : static point [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

26 extracted references · 19 canonical work pages

  1. [1]

    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 Trans. Robot., vol. 38, no. 4, pp. 2053–2073, 2022

    work page 2022

  2. [2]

    LINS: A lidar-inertial state estimator for robust and efficient navigation,

    C. Qin, H. Ye, C. E. Pranata, J. Han, S. Zhang, and M. Liu, “LINS: A lidar-inertial state estimator for robust and efficient navigation,” in Proc. IEEE Intl. Conf. on Robot. and Automat. , 2020, pp. 8899–8906

    work page 2020

  3. [3]

    Point-LIO: Robust high-bandwidth light detection and ranging inertial odometry,

    D. He, W. Xu, N. Chen, F. Kong, C. Yuan, and F. Zhang, “Point-LIO: Robust high-bandwidth light detection and ranging inertial odometry,” Advanced Intelligent Systems , vol. 5, no. 7, p. 2200459, 2023

    work page 2023

  4. [4]

    Lidar measurement bias estimation via return waveform modelling in a context of 3d mapping,

    J. Laconte, S.-P. Desch ˆenes, M. Labussi `ere, and F. Pomerleau, “Lidar measurement bias estimation via return waveform modelling in a context of 3d mapping,” in Proc. IEEE Intl. Conf. on Robot. and Automat. , 2019, pp. 8100–8106

    work page 2019

  5. [5]

    Gravity- constrained point cloud registration,

    V . Kubelka, M. Vaidis, and F. Pomerleau, “Gravity- constrained point cloud registration,” in Proc. IEEE/RSJ Intl. Conf. on Intell. Robots and Sys. IEEE, 2022, pp. 4873–4879

    work page 2022

  6. [6]

    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 Trans. Multimedia, vol. 25, pp. 3905–3920, 2023

    work page 2023

  7. [7]

    Joint on- manifold gravity and accelerometer intrinsics estima- tion for inertially aligned mapping,

    R. Nemiroff, K. Chen, and B. T. Lopez, “Joint on- manifold gravity and accelerometer intrinsics estima- tion for inertially aligned mapping,” in Proc. IEEE/RSJ Intl. Conf. on Intell. Robots and Sys. IEEE, 2023, pp. 1388–1394

    work page 2023

  8. [8]

    Wildcat: Online continuous-time 3d lidar-inertial slam,

    M. Ramezani, K. Khosoussi, G. Catt, P. Moghadam, J. Williams, P. Borges, F. Pauling, and N. Kot- tege, “Wildcat: Online continuous-time 3d lidar-inertial slam,” arXiv preprint arXiv:2205.12595 , 2022

    arXiv 2022

Show all 26 references
  1. [9]

    Nebula: Quest for robotic autonomy in challenging environments; team costar at the darpa sub- terranean challenge,

    A. Agha, K. Otsu, B. Morrell, D. D. Fan, R. Thakker, A. Santamaria-Navarro, S.-K. Kim, A. Bouman, X. Lei, J. Edlund et al., “Nebula: Quest for robotic autonomy in challenging environments; team costar at the darpa sub- terranean challenge,” arXiv preprint arXiv:2103.11470 , 2021

    2021 arXiv

  2. [10]

    On-manifold preintegration for real-time visual– inertial odometry,

    C. Forster, L. Carlone, F. Dellaert, and D. Scaramuzza, “On-manifold preintegration for real-time visual– inertial odometry,” IEEE Trans. Robot. , vol. 33, no. 1, pp. 1–21, 2016

    2016

  3. [11]

    Degradation resilient lidar-radar-inertial odometry,

    M. Nissov, N. Khedekar, and K. Alexis, “Degradation resilient lidar-radar-inertial odometry,” in Proc. IEEE Intl. Conf. on Robot. and Automat. , 2024, pp. 8587– 8594

    2024

  4. [12]

    RoLM: Radar on lidar map localization,

    Y . Ma, X. Zhao, H. Li, Y . Gu, X. Lang, and Y . Liu, “RoLM: Radar on lidar map localization,” in Proc. IEEE Intl. Conf. on Robot. and Automat. , 2023, pp. 3976–3982

    2023

  5. [13]

    Radar localiza- tion and mapping for indoor disaster environments via multi-modal registration to prior lidar map,

    Y . S. Park, J. Kim, and A. Kim, “Radar localiza- tion and mapping for indoor disaster environments via multi-modal registration to prior lidar map,” in Proc. IEEE/RSJ Intl. Conf. on Intell. Robots and Sys. , 2019, pp. 1307–1314

    2019

  6. [14]

    RaLL: End-to-end radar localization on lidar map using dif- ferentiable measurement model,

    H. Yin, R. Chen, Y . Wang, and R. Xiong, “RaLL: End-to-end radar localization on lidar map using dif- ferentiable measurement model,” IEEE Trans. Intell. Transport. Sys., vol. 23, no. 7, pp. 6737–6750, 2022

    2022

  7. [15]

    Contact-aided invariant extended kalman filtering for robot state estimation,

    R. Hartley, M. Ghaffari, R. M. Eustice, and J. W. Griz- zle, “Contact-aided invariant extended kalman filtering for robot state estimation,” Intl. J. of Robot. Research , vol. 39, no. 4, pp. 402–430, 2020

    2020

  8. [16]

    Kalman filters on differ- entiable manifolds,

    D. He, W. Xu, and F. Zhang, “Kalman filters on differ- entiable manifolds,” arXiv preprint arXiv:2102.03804 , 2021

    2021 arXiv

  9. [17]

    Invariant extended kalman filtering for tightly coupled lidar- inertial odometry and mapping,

    P. Shi, Z. Zhu, S. Sun, X. Zhao, and M. Tan, “Invariant extended kalman filtering for tightly coupled lidar- inertial odometry and mapping,” IEEE/ASME Trans. Mechatronics, vol. 28, no. 4, pp. 2213–2224, 2023

    2023

  10. [18]

    SR- LIVO: Lidar-inertial-visual odometry and mapping with sweep reconstruction,

    Z. Yuan, J. Deng, R. Ming, F. Lang, and X. Yang, “SR- LIVO: Lidar-inertial-visual odometry and mapping with sweep reconstruction,” IEEE Robot. and Automat. Lett. , vol. 9, no. 6, pp. 5110–5117, 2024

    2024

  11. [19]

    An ekf based approach to radar inertial odometry,

    C. Doer and G. F. Trommer, “An ekf based approach to radar inertial odometry,” in 2020 IEEE International Conference on Multisensor Fusion and Integration for Intelligent Systems (MFI) . IEEE, 2020, pp. 152–159

    2020

  12. [20]

    4DRadarSLAM: A 4d imaging radar slam system for large-scale environments based on pose graph optimization,

    J. Zhang, H. Zhuge, Z. Wu, G. Peng, M. Wen, Y . Liu, and D. Wang, “4DRadarSLAM: A 4d imaging radar slam system for large-scale environments based on pose graph optimization,” in Proc. IEEE Intl. Conf. on Robot. and Automat. IEEE, 2023, pp. 8333–8340

    2023

  13. [21]

    ikd-tree: An incre- mental kd tree for robotic applications,

    Y . Cai, W. Xu, and F. Zhang, “ikd-tree: An incre- mental kd tree for robotic applications,” arXiv preprint arXiv:2102.10808, 2021

    2021 arXiv

  14. [22]

    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 Trans. Robot. , vol. 34, no. 4, pp. 1004–1020, 2018

    2018

  15. [23]

    Ntu4dradlm: 4d radar-centric multi-modal dataset for localization and mapping,

    J. Zhang, H. Zhuge, Y . Liu, G. Peng, Z. Wu, H. Zhang, Q. Lyu, H. Li, C. Zhao, D. Kircali et al., “Ntu4dradlm: 4d radar-centric multi-modal dataset for localization and mapping,” in Proc. IEEE Intell. Transport. Sys. Conf. IEEE, 2023, pp. 4291–4296

    2023

  16. [24]

    Snail-Radar: A large-scale diverse dataset for the evaluation of 4d-radar-based slam systems,

    J. Huai, B. Wang, Y . Zhuang, Y . Chen, Q. Li, Y . Han, and C. Toth, “Snail-Radar: A large-scale diverse dataset for the evaluation of 4d-radar-based slam systems,” arXiv preprint arXiv:2407.11705 , 2024

    2024 arXiv

  17. [25]

    Direct lidar- inertial odometry: Lightweight lio with continuous-time motion correction,

    K. Chen, R. Nemiroff, and B. T. Lopez, “Direct lidar- inertial odometry: Lightweight lio with continuous-time motion correction,” in Proc. IEEE Intl. Conf. on Robot. and Automat. , 2023, pp. 3983–3989

    2023

  18. [26]

    evo: Python package for the evaluation of odometry and slam

    M. Grupp, “evo: Python package for the evaluation of odometry and slam.” https://github.com/MichaelGrupp/ evo, 2017

    2017

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.