Geometrically-Constrained Radar-Inertial Odometry via Continuous Point-Pose Uncertainty Modeling

arxiv: 2604.02745 · v1 · submitted 2026-04-03 · 💻 cs.RO

Geometrically-Constrained Radar-Inertial Odometry via Continuous Point-Pose Uncertainty Modeling

Wooseong Yang , Dongjae Lee , Minwoo Jung , Ayoung Kim This is my paper

Pith reviewed 2026-05-13 20:26 UTC · model grok-4.3

classification 💻 cs.RO

keywords radar odometryinertial fusionuncertainty propagationcontinuous trajectorygeometric constraintsradar mappingsensor fusionpose estimation

0 comments p. Extension

The pith

Continuous point-pose uncertainty modeling lets radar-inertial odometry down-weight noisy returns and enforce geometric constraints for higher accuracy.

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

The paper introduces a radar-inertial odometry system that tracks uncertainty in both radar points and vehicle poses across continuous time rather than at discrete instants. A continuous trajectory model propagates uncertainty from a small set of control points to any chosen timestamp, then combines that pose uncertainty with the varying noise level of each radar return during projection. The combined uncertainty score determines how much each point contributes to the optimization, enabling automatic down-weighting of unreliable observations. The same uncertainty values are used to build a higher-quality map, and the framework adds explicit geometric constraints between points to further stabilize the solution. Experiments across multiple real-world datasets show measurable gains in both localization accuracy and runtime speed over prior radar-inertial baselines.

Core claim

By propagating pose uncertainty from control points through a continuous trajectory model and fusing it with heteroscedastic radar measurement noise, the method produces a per-point that allows adaptive down-weighting during optimization; the same quantified uncertainties improve map fidelity, and explicit geometric constraints become effective once unreliable points are suppressed.

What carries the argument

Continuous trajectory model that propagates control-point uncertainties to arbitrary timestamps and integrates them with heteroscedastic radar noise to compute dynamic observation weights.

If this is right

Sparse radar returns receive lower influence when their projected uncertainty is high, reducing drift in low-feature scenes.
The uncertainty-weighted map produces fewer outliers for subsequent localization steps.
Explicit geometric constraints between radar points become usable without being dominated by noisy measurements.
The method runs faster than baselines because fewer low-confidence points enter the optimization.

Where Pith is reading between the lines

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

The same continuous-uncertainty mechanism could be applied to other continuous-time sensors such as event cameras or spinning LiDAR.
Long-term mapping consistency might improve if the uncertainty values are also used to decide when to add or remove map points.
The framework suggests a route toward tighter coupling with learned radar noise models that predict heteroscedasticity from raw data.

Load-bearing premise

The continuous trajectory model must propagate pose uncertainties accurately enough that their combination with radar noise produces unbiased weights.

What would settle it

Running the system on a dataset with precise ground-truth trajectories while disabling the uncertainty propagation step and observing whether accuracy or robustness degrades.

Figures

Figures reproduced from arXiv: 2604.02745 by Ayoung Kim, Dongjae Lee, Minwoo Jung, Wooseong Yang.

**Figure 1.** Figure 1: We jointly model and continuously consolidate point-pose uncertainty in radar odometry and mapping, enabling the effective usage of explicit local geometry. Two key considerations are indispensable for reliable uncertainty analysis for radar odometry and mapping: (i) modeling heteroscedastic, view-dependent measurement uncertainty [9], (ii) addressing pose-uncertainty-aware propagation, which affects acc… view at source ↗

**Figure 2.** Figure 2: Our method consists of preprocessing, state estimation, and mapping. In preprocessing, we remove dynamic points from radar measurements using ego-velocity estimates and perform data association for the measurements used in the state update. The state estimation module iteratively performs residual computation, uncertainty propagation, and localizability-constrained IEKF updates until convergence. After con… view at source ↗

**Figure 3.** Figure 3: Conceptual diagram of our uncertainty propagation module. Uncertainties of B-spline control points are propagated to the pose at the radar timestamp and subsequently to individual radar returns. The resulting per-point uncertainties in the global frame perform as confidence weights in the observation model for local geometryguided scan-to-submap registration. C. Scan-to-Submap Registration with Explicit L… view at source ↗

**Figure 5.** Figure 5: Comparison of odometry and mapping on Parking Lot between the proposed method and LiDAR-Inertial Odometry (FastLIO2). Our proposed method yields a well-aligned odometry and mapping result, and also achieves LiDAR-level map consistency. ground-optimized filtering. Notably, ours achieves higher ATE on Sequence_2, which involves high-dynamic maneuvers of a toy racing car. In this scenario, rapid acceleration… view at source ↗

**Figure 7.** Figure 7: Evaluation result with different uncertainty thresholds in the proposed method (orange, ATE). The gray line denotes the result from second-best baseline. The proposed method maintains stable accuracy across thresholds and still outperforms other baselines. TABLE VI: Effect of localizability-constrained update Library 01 Bridge 01 20240123_2 ATE RPEt RPEr ATE RPEt RPEr ATE RPEt RPEr w/o Loc. 13.082 0.140 0.… view at source ↗

**Figure 8.** Figure 8: Runtime analysis in 20240115_2 VI. CONCLUSION In this paper, we presented a robust uncertainty-aware continuous radar–inertial odometry and mapping framework. We proposed a novel continuous uncertainty propagation method that consolidates point and pose uncertainty in the radar system, yielding adaptive covariances that down-weight uninformative observations and preserve a consistent map for robust registr… view at source ↗

read the original abstract

Radar odometry is crucial for robust localization in challenging environments; however, the sparsity of reliable returns and distinctive noise characteristics impede its performance. This paper introduces geometrically-constrained radar-inertial odometry and mapping that jointly consolidates point and pose uncertainty. We employ the continuous trajectory model to estimate the pose uncertainty at any arbitrary timestamp by propagating uncertainties of the control points. These pose uncertainties are continuously integrated with heteroscedastic measurement uncertainty during point projection, thereby enabling dynamic evaluation of observation confidence and adaptive down-weighting of uninformative radar points. By leveraging quantified uncertainties in radar mapping, we construct a high-fidelity map that improves odometry accuracy under imprecise radar measurements. Moreover, we reveal the effectiveness of explicit geometrical constraints in radar-inertial odometry when incorporated with the proposed uncertainty-aware mapping framework. Extensive experiments on diverse real-world datasets demonstrate the superiority of our method, yielding substantial performance improvements in both accuracy and efficiency compared to existing baselines.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note private letter to a colleague

The paper adds continuous-time uncertainty propagation to radar-inertial odometry for adaptive point weighting and geometric constraints, but the gains hinge on unverified covariance approximations.

read the letter

The core idea is propagating control-point covariances through a continuous trajectory model to get pose uncertainty at any timestamp, then folding that together with per-point radar noise to down-weight unreliable observations on the fly. They also add explicit geometric constraints and use the uncertainties to build a higher-quality map. This is a step beyond typical discrete-time radar pipelines that treat noise more statically. The continuous model fits the timing of radar returns with inertial data reasonably well, and the joint point-pose treatment addresses radar sparsity directly. Real-world dataset results showing accuracy and efficiency gains over baselines are the strongest part of the presentation. The geometric constraints appear to help when paired with the uncertainty framework, which is a practical addition for mapping. The soft spot sits in the propagation mechanics. If the method relies on local Jacobians without full cross-covariance blocks between control points, or if it reduces the 3-D pose covariance to a scalar weight when fusing with heteroscedastic radar noise, the down-weighting can introduce bias rather than remove it. The abstract gives no equations or validation against the true posterior, so it is unclear whether the claimed improvements survive those common approximations. Ablations isolating the uncertainty component would clarify how much it actually drives the results versus the constraints or other tuning. This is for robotics groups working on radar or multi-sensor odometry in field conditions. Readers who need robust localization under sparse, noisy returns will find the framework worth examining. It deserves peer review to check the derivations and run the experiments in detail.

Referee Report

2 major / 2 minor

Summary. The paper proposes geometrically-constrained radar-inertial odometry that propagates control-point covariances through a continuous trajectory model to obtain pose uncertainty at arbitrary timestamps, fuses these with heteroscedastic radar measurement covariances for adaptive point weighting during projection, and uses the resulting uncertainty-aware map plus explicit geometric constraints to improve odometry accuracy and efficiency. Extensive real-world experiments are claimed to demonstrate substantial gains over baselines.

Significance. If the uncertainty propagation and fusion steps are shown to be unbiased, the approach could strengthen continuous-time radar-inertial pipelines in sparse, high-noise regimes by providing principled down-weighting and map quality, with potential impact on robust localization for autonomous systems.

major comments (2)

[Method (uncertainty propagation and weighting)] The central mechanism—propagation of control-point covariances to arbitrary timestamps followed by fusion with per-point measurement covariance to produce observation weights—is load-bearing for all claimed accuracy gains. The manuscript must specify whether the full cross-covariance blocks between control points are retained during propagation or whether only local Jacobians are used, because the latter (common in continuous-time SLAM) can systematically bias the resulting weights when combined with heteroscedastic radar noise.
[Method (point projection and weighting)] The abstract asserts that the continuous model enables 'dynamic evaluation of observation confidence' without introducing bias, yet no validation against ground-truth posterior covariances or Monte-Carlo sampling is described. If the fusion treats the projected pose uncertainty as a scalar variance rather than the full 3-D covariance in the measurement residual, the adaptive down-weighting can over- or under-penalize points, undermining the superiority claims.

minor comments (2)

[Abstract and Introduction] Clarify the exact form of the 'explicit geometrical constraints' (e.g., which planes, lines, or coplanarity conditions) and how they are incorporated into the optimization; this is mentioned in the abstract but not detailed in the high-level description.
[Experiments] The experiments section should include an ablation isolating the contribution of the uncertainty-aware mapping versus the geometric constraints alone, with quantitative metrics (e.g., ATE, RPE) reported for each component.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments on the uncertainty propagation and fusion mechanisms. These points help clarify key aspects of our approach. We address each major comment below and will revise the manuscript to improve explicitness and address the concerns raised.

read point-by-point responses

Referee: [Method (uncertainty propagation and weighting)] The central mechanism—propagation of control-point covariances to arbitrary timestamps followed by fusion with per-point measurement covariance to produce observation weights—is load-bearing for all claimed accuracy gains. The manuscript must specify whether the full cross-covariance blocks between control points are retained during propagation or whether only local Jacobians are used, because the latter (common in continuous-time SLAM) can systematically bias the resulting weights when combined with heteroscedastic radar noise.

Authors: We agree that explicit specification is necessary. Our continuous trajectory model propagates the full joint covariance matrix of the control points (including all cross-covariance blocks) to arbitrary timestamps via the trajectory representation. This avoids the bias that would arise from local Jacobians alone. The fusion step then combines this full propagated pose covariance with the heteroscedastic radar measurement covariance. We will revise the method section to state this explicitly, add the relevant propagation equations, and note why full covariances are retained to ensure unbiased weighting. revision: yes
Referee: [Method (point projection and weighting)] The abstract asserts that the continuous model enables 'dynamic evaluation of observation confidence' without introducing bias, yet no validation against ground-truth posterior covariances or Monte-Carlo sampling is described. If the fusion treats the projected pose uncertainty as a scalar variance rather than the full 3-D covariance in the measurement residual, the adaptive down-weighting can over- or under-penalize points, undermining the superiority claims.

Authors: We thank the referee for this observation. The fusion uses the full 3-D projected pose covariance (transformed into the measurement space via the appropriate Jacobian) together with the per-point measurement covariance to compute the observation weight; it is not reduced to a scalar variance. This is derived in the uncertainty modeling section. While the current experiments rely on real-world odometry accuracy to support the claims rather than explicit Monte-Carlo sampling against ground-truth posteriors, we acknowledge the value of such validation. We will add a clarifying paragraph on the full-covariance usage and include a brief theoretical discussion or small-scale validation to address potential bias concerns. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper's central mechanism propagates control-point uncertainties through a continuous trajectory model to obtain pose covariances at arbitrary times, then fuses them with per-measurement heteroscedastic noise for adaptive weighting and mapping. No equations or descriptions in the abstract reduce this propagation or weighting to a fitted parameter renamed as a prediction, a self-definitional loop, or a load-bearing self-citation whose validity is assumed without external grounding. The claimed accuracy gains rest on the geometric constraints and uncertainty-aware fusion being applied to real datasets, which are independent of the derivation itself. This is the normal case of a self-contained technical contribution.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract supplies no explicit free parameters, axioms, or invented entities; all modeling details remain unspecified.

pith-pipeline@v0.9.0 · 5469 in / 977 out tokens · 28419 ms · 2026-05-13T20:26:22.582735+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We employ the continuous trajectory model to estimate the pose uncertainty at any arbitrary timestamp by propagating uncertainties of the control points. ... ΣΣΣt(t)=∑mk(u)²ΣΣΣti−3+k ... ΣΣΣWp=ΣΣΣt(t)+R(t)[Ip]×ΣΣΣR(t)[Ip]⊤×R(t)⊤+R(t)ΣΣΣIpR(t)⊤
IndisputableMonolith/Foundation/AlexanderDuality.lean alexander_duality_circle_linking unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Localizability-constrained IEKF update ... ΛΛΛ=L⊤L ... δx projected onto null space of under-constrained axes

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

24 extracted references · 24 canonical work pages

[1]

A new wave in robotics: Survey on recent mmwave radar applications in robotics,

K. Harlow, H. Jang, T. D. Barfoot, A. Kim, and C. Heck- man, “A new wave in robotics: Survey on recent mmwave radar applications in robotics,”IEEE Trans. Robot., vol. 40, pp. 4544–4560, 2024

work page 2024
[2]

Lidar odometry survey: recent advancements and remaining challenges,

D. Lee, M. Jung, W. Yang, and A. Kim, “Lidar odometry survey: recent advancements and remaining challenges,” Intl. Service Robot., vol. 17, no. 2, pp. 95–118, 2024

work page 2024
[3]

Kiss-icp: In defense of point-to-point icp–simple, accurate, and robust registra- tion if done the right way,

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 registra- tion if done the right way,”IEEE Robot. and Automat. Lett., vol. 8, no. 2, pp. 1029–1036, 2023

work page 2023
[4]

Ground-optimized 4d radar-inertial odometry via continuous velocity integra- tion using gaussian process,

W. Yang, H. Jang, and A. Kim, “Ground-optimized 4d radar-inertial odometry via continuous velocity integra- tion using gaussian process,” inProc. IEEE Intl. Conf. on Robot. and Automat., 2025, pp. 3815–3821

work page 2025
[5]

4d iriom: 4d imaging radar inertial odometry and mapping,

Y . Zhuang, B. Wang, J. Huai, and M. Li, “4d iriom: 4d imaging radar inertial odometry and mapping,”IEEE Robot. and Automat. Lett., vol. 8, no. 6, pp. 3246–3253, 2023

work page 2023
[6]

Radar-only odometry and mapping for autonomous vehicles,

D. C. Herraez, M. Zeller, L. Chang, I. Vizzo, M. Hei- dingsfeld, and C. Stachniss, “Radar-only odometry and mapping for autonomous vehicles,” inProc. IEEE Intl. Conf. on Robot. and Automat., 2024, pp. 10 275–10 282

work page 2024
[7]

Rai-slam: Radar-inertial slam for autonomous vehicles,

D. C. Herraez, M. Zeller, D. Wang, J. Behley, M. Heid- ingsfeld, and C. Stachniss, “Rai-slam: Radar-inertial slam for autonomous vehicles,”IEEE Robot. and Automat. Lett., vol. 10, no. 6, pp. 5257–5264, 2025

work page 2025
[8]

Asynchronous multiple lidar-inertial odometry using point-wise inter-lidar un- certainty propagation,

M. Jung, S. Jung, and A. Kim, “Asynchronous multiple lidar-inertial odometry using point-wise inter-lidar un- certainty propagation,”IEEE Robot. and Automat. Lett., vol. 8, no. 7, pp. 4211–4218, 2023

work page 2023
[9]

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,” inProc. IEEE Intl. Conf. on Robot. and Automat., 2023, pp. 8333–8340

work page 2023
[10]

Efficient and probabilistic adaptive voxel mapping for accurate online lidar odometry,

C. Yuan, W. Xu, X. Liu, X. Hong, and F. Zhang, “Efficient and probabilistic adaptive voxel mapping for accurate online lidar odometry,”IEEE Robot. and Au- tomat. Lett., vol. 7, no. 3, pp. 8518–8525, 2022

work page 2022
[11]

Incorporating point uncertainty in radar slam,

Y . Xu, Q. Huang, S. Shen, and H. Yin, “Incorporating point uncertainty in radar slam,”IEEE Robot. and Au- tomat. Lett., vol. 10, no. 3, pp. 2168–2175, 2025

work page 2025
[12]

4d radar-based pose graph slam with ego-velocity pre-integration factor,

X. Li, H. Zhang, and W. Chen, “4d radar-based pose graph slam with ego-velocity pre-integration factor,” IEEE Robot. and Automat. Lett., vol. 8, no. 8, pp. 5124– 5131, 2023

work page 2023
[13]

Efear- 4d: Ego-velocity filtering for efficient and accurate 4d radar odometry,

X. Wu, Y . Chen, Z. Li, Z. Hong, and L. Hu, “Efear- 4d: Ego-velocity filtering for efficient and accurate 4d radar odometry,”IEEE Robot. and Automat. Lett., vol. 9, no. 11, pp. 9828–9835, 2024

work page 2024
[14]

Ground-aware automotive radar odometry,

D. C. Herraez, F. Kaschner, M. Zeller, D. Muhle, J. Behley, M. Heidingsfeld, D. Cremers, and C. Stach- niss, “Ground-aware automotive radar odometry,” in Proc. IEEE Intl. Conf. on Robot. and Automat., 2025

work page 2025
[15]

isam: In- cremental smoothing and mapping,

M. Kaess, A. Ranganathan, and F. Dellaert, “isam: In- cremental smoothing and mapping,”IEEE Trans. Robot., vol. 24, no. 6, pp. 1365–1378, 2008

work page 2008
[16]

Log-lio2: A lidar-inertial odometry with efficient uncertainty analysis,

K. Huang, J. Zhao, J. Lin, Z. Zhu, S. Song, C. Ye, and T. Feng, “Log-lio2: A lidar-inertial odometry with efficient uncertainty analysis,”IEEE Robot. and Automat. Lett., vol. 9, no. 10, pp. 8226–8233, 2024

work page 2024
[17]

Resple: Recursive spline estimation for lidar-based odometry,

Z. Cao, W. Talbot, and K. Li, “Resple: Recursive spline estimation for lidar-based odometry,”IEEE Robot. and Automat. Lett., vol. 10, no. 10, pp. 10 666–10 673, 2025

work page 2025
[18]

Co-ral: Complementary radar-leg odometry with 4-dof optimization and rolling contact,

S. Jung, W. Yang, and A. Kim, “Co-ral: Complementary radar-leg odometry with 4-dof optimization and rolling contact,” inProc. IEEE/RSJ Intl. Conf. on Intell. Robots and Sys., 2024, pp. 13 289–13 296

work page 2024
[19]

A review of point cloud registration algorithms for mobile robotics,

F. Pomerleau, F. Colas, and R. Siegwart, “A review of point cloud registration algorithms for mobile robotics,” F oundations and Trends® in Robotics, vol. 4, pp. 1–104, 05 2015

work page 2015
[20]

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 2053
[21]

Hercules: Heterogeneous radar dataset in complex urban environment for multi-session radar slam,

H. Kim, M. Jung, C. Noh, S. Jung, H. Song, W. Yang, H. Jang, and A. Kim, “Hercules: Heterogeneous radar dataset in complex urban environment for multi-session radar slam,” inProc. IEEE Intl. Conf. on Robot. and Automat., 2025

work page 2025
[22]

Snail radar: A large-scale diverse benchmark for evaluating 4d-radar-based slam,

J. Huai, B. Wang, Y . Zhuang, Y . Chen, Q. Li, and Y . Han, “Snail radar: A large-scale diverse benchmark for evaluating 4d-radar-based slam,”Intl. J. of Robot. Research, p. 02783649251329048, 2024

work page 2024
[23]

Ekf- based radar-inertial odometry with online temporal cali- bration,

C. Kim, G. Bae, W. Shin, S. Wang, and H. Oh, “Ekf- based radar-inertial odometry with online temporal cali- bration,”IEEE Robot. and Automat. Lett., vol. 10, no. 7, pp. 7230–7237, 2025

work page 2025
[24]

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

work page 2017

[1] [1]

A new wave in robotics: Survey on recent mmwave radar applications in robotics,

K. Harlow, H. Jang, T. D. Barfoot, A. Kim, and C. Heck- man, “A new wave in robotics: Survey on recent mmwave radar applications in robotics,”IEEE Trans. Robot., vol. 40, pp. 4544–4560, 2024

work page 2024

[2] [2]

Lidar odometry survey: recent advancements and remaining challenges,

D. Lee, M. Jung, W. Yang, and A. Kim, “Lidar odometry survey: recent advancements and remaining challenges,” Intl. Service Robot., vol. 17, no. 2, pp. 95–118, 2024

work page 2024

[3] [3]

Kiss-icp: In defense of point-to-point icp–simple, accurate, and robust registra- tion if done the right way,

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 registra- tion if done the right way,”IEEE Robot. and Automat. Lett., vol. 8, no. 2, pp. 1029–1036, 2023

work page 2023

[4] [4]

Ground-optimized 4d radar-inertial odometry via continuous velocity integra- tion using gaussian process,

W. Yang, H. Jang, and A. Kim, “Ground-optimized 4d radar-inertial odometry via continuous velocity integra- tion using gaussian process,” inProc. IEEE Intl. Conf. on Robot. and Automat., 2025, pp. 3815–3821

work page 2025

[5] [5]

4d iriom: 4d imaging radar inertial odometry and mapping,

Y . Zhuang, B. Wang, J. Huai, and M. Li, “4d iriom: 4d imaging radar inertial odometry and mapping,”IEEE Robot. and Automat. Lett., vol. 8, no. 6, pp. 3246–3253, 2023

work page 2023

[6] [6]

Radar-only odometry and mapping for autonomous vehicles,

D. C. Herraez, M. Zeller, L. Chang, I. Vizzo, M. Hei- dingsfeld, and C. Stachniss, “Radar-only odometry and mapping for autonomous vehicles,” inProc. IEEE Intl. Conf. on Robot. and Automat., 2024, pp. 10 275–10 282

work page 2024

[7] [7]

Rai-slam: Radar-inertial slam for autonomous vehicles,

D. C. Herraez, M. Zeller, D. Wang, J. Behley, M. Heid- ingsfeld, and C. Stachniss, “Rai-slam: Radar-inertial slam for autonomous vehicles,”IEEE Robot. and Automat. Lett., vol. 10, no. 6, pp. 5257–5264, 2025

work page 2025

[8] [8]

Asynchronous multiple lidar-inertial odometry using point-wise inter-lidar un- certainty propagation,

M. Jung, S. Jung, and A. Kim, “Asynchronous multiple lidar-inertial odometry using point-wise inter-lidar un- certainty propagation,”IEEE Robot. and Automat. Lett., vol. 8, no. 7, pp. 4211–4218, 2023

work page 2023

[9] [9]

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,” inProc. IEEE Intl. Conf. on Robot. and Automat., 2023, pp. 8333–8340

work page 2023

[10] [10]

Efficient and probabilistic adaptive voxel mapping for accurate online lidar odometry,

C. Yuan, W. Xu, X. Liu, X. Hong, and F. Zhang, “Efficient and probabilistic adaptive voxel mapping for accurate online lidar odometry,”IEEE Robot. and Au- tomat. Lett., vol. 7, no. 3, pp. 8518–8525, 2022

work page 2022

[11] [11]

Incorporating point uncertainty in radar slam,

Y . Xu, Q. Huang, S. Shen, and H. Yin, “Incorporating point uncertainty in radar slam,”IEEE Robot. and Au- tomat. Lett., vol. 10, no. 3, pp. 2168–2175, 2025

work page 2025

[12] [12]

4d radar-based pose graph slam with ego-velocity pre-integration factor,

X. Li, H. Zhang, and W. Chen, “4d radar-based pose graph slam with ego-velocity pre-integration factor,” IEEE Robot. and Automat. Lett., vol. 8, no. 8, pp. 5124– 5131, 2023

work page 2023

[13] [13]

Efear- 4d: Ego-velocity filtering for efficient and accurate 4d radar odometry,

X. Wu, Y . Chen, Z. Li, Z. Hong, and L. Hu, “Efear- 4d: Ego-velocity filtering for efficient and accurate 4d radar odometry,”IEEE Robot. and Automat. Lett., vol. 9, no. 11, pp. 9828–9835, 2024

work page 2024

[14] [14]

Ground-aware automotive radar odometry,

D. C. Herraez, F. Kaschner, M. Zeller, D. Muhle, J. Behley, M. Heidingsfeld, D. Cremers, and C. Stach- niss, “Ground-aware automotive radar odometry,” in Proc. IEEE Intl. Conf. on Robot. and Automat., 2025

work page 2025

[15] [15]

isam: In- cremental smoothing and mapping,

M. Kaess, A. Ranganathan, and F. Dellaert, “isam: In- cremental smoothing and mapping,”IEEE Trans. Robot., vol. 24, no. 6, pp. 1365–1378, 2008

work page 2008

[16] [16]

Log-lio2: A lidar-inertial odometry with efficient uncertainty analysis,

K. Huang, J. Zhao, J. Lin, Z. Zhu, S. Song, C. Ye, and T. Feng, “Log-lio2: A lidar-inertial odometry with efficient uncertainty analysis,”IEEE Robot. and Automat. Lett., vol. 9, no. 10, pp. 8226–8233, 2024

work page 2024

[17] [17]

Resple: Recursive spline estimation for lidar-based odometry,

Z. Cao, W. Talbot, and K. Li, “Resple: Recursive spline estimation for lidar-based odometry,”IEEE Robot. and Automat. Lett., vol. 10, no. 10, pp. 10 666–10 673, 2025

work page 2025

[18] [18]

Co-ral: Complementary radar-leg odometry with 4-dof optimization and rolling contact,

S. Jung, W. Yang, and A. Kim, “Co-ral: Complementary radar-leg odometry with 4-dof optimization and rolling contact,” inProc. IEEE/RSJ Intl. Conf. on Intell. Robots and Sys., 2024, pp. 13 289–13 296

work page 2024

[19] [19]

A review of point cloud registration algorithms for mobile robotics,

F. Pomerleau, F. Colas, and R. Siegwart, “A review of point cloud registration algorithms for mobile robotics,” F oundations and Trends® in Robotics, vol. 4, pp. 1–104, 05 2015

work page 2015

[20] [20]

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 2053

[21] [21]

Hercules: Heterogeneous radar dataset in complex urban environment for multi-session radar slam,

H. Kim, M. Jung, C. Noh, S. Jung, H. Song, W. Yang, H. Jang, and A. Kim, “Hercules: Heterogeneous radar dataset in complex urban environment for multi-session radar slam,” inProc. IEEE Intl. Conf. on Robot. and Automat., 2025

work page 2025

[22] [22]

Snail radar: A large-scale diverse benchmark for evaluating 4d-radar-based slam,

J. Huai, B. Wang, Y . Zhuang, Y . Chen, Q. Li, and Y . Han, “Snail radar: A large-scale diverse benchmark for evaluating 4d-radar-based slam,”Intl. J. of Robot. Research, p. 02783649251329048, 2024

work page 2024

[23] [23]

Ekf- based radar-inertial odometry with online temporal cali- bration,

C. Kim, G. Bae, W. Shin, S. Wang, and H. Oh, “Ekf- based radar-inertial odometry with online temporal cali- bration,”IEEE Robot. and Automat. Lett., vol. 10, no. 7, pp. 7230–7237, 2025

work page 2025

[24] [24]

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

work page 2017