Raymoval: Raycasting-based Dynamic Object Removal for Static 3D Mapping

Daebeom Kim; Hyun Myung; Kevin Christiansen Marsim; Seoyeon Jang; Seungjae Lee

arxiv: 2605.08937 · v1 · submitted 2026-05-09 · 💻 cs.RO

Raymoval: Raycasting-based Dynamic Object Removal for Static 3D Mapping

Daebeom Kim , Seungjae Lee , Seoyeon Jang , Kevin Christiansen Marsim , Hyun Myung This is my paper

Pith reviewed 2026-05-12 01:45 UTC · model grok-4.3

classification 💻 cs.RO

keywords dynamic object removalstatic 3D mappingraycastingazimuth-elevation gridmap consistencyrobot navigationSemanticKITTI

0 comments

The pith

Raycasting on angular bins removes dynamic objects from static 3D maps with less over-removal.

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

Static 3D maps give robots a consistent geometric reference for navigation, yet moving objects leave unwanted traces and erode surfaces. The paper shows that projecting each scan onto an azimuth-elevation grid and comparing the minimum range in every viewing direction against the map's first-hit distance, computed by raycasting, can label dynamic points. A raycast consistency test then distinguishes true motion from static structure, and a final spatial validation step refines the labels. The resulting maps contain fewer residual dynamics and preserve more static geometry than prior removal techniques, as measured on SemanticKITTI and a custom dataset. Readers care because reliable static maps underpin long-term autonomy without repeated surface loss or ghosting artifacts.

Core claim

Each scan is projected onto an azimuth-elevation grid; for every bin the minimum range is compared with the map's first-hit distance obtained by raycasting. A raycast consistency test separates dynamic from static points, after which spatial consistency validation refines the labels, yielding static maps that exhibit lower residual dynamics and reduced over-removal.

What carries the argument

The raycasting-based removal module that projects scans to an azimuth-elevation grid, compares bin-wise minimum ranges to the map's first-hit distance, and applies consistency tests to label and excise dynamic points.

If this is right

Static maps retain fewer residual dynamics from moving objects.
Over-removal of static surfaces is measurably reduced relative to existing methods.
Quantitative and qualitative gains appear on both SemanticKITTI and a custom challenging dataset.
Improved map consistency directly supports persistent reference use in robot navigation.

Where Pith is reading between the lines

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

Cleaner static maps could reduce drift accumulation in repeated traversals by supplying a more stable prior.
The grid-based comparison might extend naturally to online mapping pipelines where the map updates incrementally.
Handling of very slow objects would require tighter integration of the consistency test with velocity estimates.

Load-bearing premise

The bin-wise minimum range comparison together with the raycast consistency test can separate dynamic points from static points without systematic over-removal or failure to catch slow-moving objects.

What would settle it

If the output maps still contain visible residual traces of moving objects or show measurable loss of static surfaces on the SemanticKITTI sequences or the custom dataset, the separation logic would be shown to be insufficient.

Figures

Figures reproduced from arXiv: 2605.08937 by Daebeom Kim, Hyun Myung, Kevin Christiansen Marsim, Seoyeon Jang, Seungjae Lee.

**Figure 1.** Figure 1: Overview of Raymoval pipeline. (a) Each scan is projected to an azimuth– elevation grid to compute the raycasting, which stores the first-hit distance per viewing direction. (b) A range-scaled consistency check compares scan ranges and raycasting map distances along the same visibility, determining dynamic points while keeping erase cases to prevent over-removal. (c) A spatial consistency validation proces… view at source ↗

**Figure 2.** Figure 2: Visualization of azimuth–elevation representation. The raycasting stores the first-hit distance from the sensor origin along the bin’s viewing direction to the nearest occupied point. We parameterize the scan directions by azimuth α and elevation β as shown in [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Figure 3: A picture of the excavator platform that acquired our custom dataset. A rigid sensors suite (SOSLab ML-X LiDAR, StereoLabs Zed-X camera, and Xsens MTI-30 IMU) is mounted on the excavator body, and an additional downward-facing SOSLab ML-X LiDAR is mounted on the boom to detect the ground. Data were acquired in active construction sites while the machine was moving and excavating. To evaluate the dynamic re… view at source ↗

**Figure 4.** Figure 4: (a)-(e) Qualitative comparison with the state-of-the-art dynamic removal approaches on 00, 01, 02, 05, 07 sequences in the SemanticKITTI dataset (from top to bottom). Green, red, and blue points indicate true positive (TP), false positive (FP), and false negative (FN), respectively (best viewed in color). (a) ERASOR [1] (b) 4DMOS [9] (c) Raymoval (d) Ground truth [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗

**Figure 5.** Figure 5: (a)-(c) Qualitative comparison with the state-of-the-art dynamic removal approaches on our challenging custom dataset. Red points indicate dynamic points (best viewed in color) [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗

read the original abstract

Static mapping is fundamental to robot navigation, providing a persistent geometric prior and a consistent reference for long-term autonomy. However, dynamic objects leave residual traces and cause surface loss, which reduces map consistency. We propose a raycasting-based module for dynamic object removal in static 3D mapping. Each scan is projected onto an azimuth-elevation grid, and for every viewing direction we compare the bin-wise minimum range with the map's first-hit distance computed by raycasting. Furthermore, we apply a raycast consistency test that separates dynamic from static points. Finally, a spatial consistency validation step refines labels, producing static maps with lower residual dynamics and reduced over-removal. We evaluate our approach quantitatively and qualitatively on SemanticKITTI and a challenging custom dataset, and show consistent static mapping results.

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

Raymoval gives a clean raycasting filter for pulling dynamic points out of maps, but the bin comparison plus consistency steps look vulnerable to slow movers and any noise already in the map.

read the letter

The paper's main contribution is a straightforward pipeline: project the scan to an azimuth-elevation grid, compare each bin's minimum range against the map's raycast first-hit distance, run a consistency test, then apply spatial validation to refine the labels. This produces maps with fewer dynamic leftovers and less accidental removal of static surfaces. The specific two-stage check after the range test is not a straight copy of the cited prior work, so there is a modest technical increment here. They evaluate on SemanticKITTI and a custom dataset with both numbers and visuals, which is the right way to show practical effect for mapping tasks. That part is useful and worth having on record. The approach stays procedural rather than learned, so no obvious circularity or heavy parameter fitting on the test data. The soft spots are exactly where the stress test points. Slow-moving objects can easily produce ranges that sit inside the same bin as static structure in a single scan, and any leftover error in the map's first-hit distances will flip the comparison the wrong way. The abstract gives no derivation or sensitivity numbers for the consistency thresholds, so it is hard to judge how much the reported gains depend on careful tuning or favorable scenes. If those cases are common in real deployment, the headline improvements shrink. This is for robotics researchers who maintain long-term static maps for navigation or autonomy. A mapping or SLAM person could implement the grid projection and test it quickly on their own data. It is not a foundational shift, but the method is concrete enough that a referee can verify the claims and suggest fixes for the edge cases. I would send it to review rather than desk reject.

Referee Report

3 major / 2 minor

Summary. The paper proposes Raymoval, a raycasting-based module for dynamic object removal in static 3D mapping. Each scan is projected onto an azimuth-elevation grid; for every bin the minimum range is compared against the map's first-hit distance obtained by raycasting the current static map. A raycast consistency test then labels points as dynamic or static, followed by a spatial consistency validation step to refine the labels. The resulting static maps are claimed to exhibit lower residual dynamics and reduced over-removal. Quantitative and qualitative results are reported on SemanticKITTI and a custom dataset.

Significance. If the bin-wise comparison and consistency tests reliably separate dynamic from static points without systematic bias, the approach would provide a lightweight, geometry-driven addition to static mapping pipelines that could improve long-term map consistency for robot navigation. The evaluation on SemanticKITTI plus a custom dataset indicates practical relevance, but the absence of detailed ablations or robustness analysis limits the assessed impact.

major comments (3)

[§3] §3 (Method): The central separation logic—bin-wise minimum-range comparison against the map's raycast first-hit distance, followed by the raycast consistency test—lacks any derivation, threshold values, or pseudocode. Without these, it is impossible to verify whether the test avoids the failure modes of slow-moving objects (whose ranges can be statistically indistinguishable from static structure within one scan) or noisy first-hit distances from prior map errors.
[§4] §4 (Experiments): The quantitative claims of “lower residual dynamics and reduced over-removal” are presented without error bars, statistical significance tests, or ablations that isolate the contribution of the consistency test versus the spatial validation step. This makes it difficult to assess whether the reported gains are robust or dataset-specific.
[§3.3] §3.3 (Spatial validation): The spatial consistency validation is described only at a high level; no details are given on the neighborhood size, distance metric, or how it interacts with the preceding raycast test. This step is load-bearing for the “reduced over-removal” claim yet remains underspecified.

minor comments (2)

[Abstract] The abstract and introduction use the phrase “raycast consistency test” without a forward reference to the exact subsection or equation that defines it.
[Figures] Figure captions should explicitly state the color coding for dynamic (removed) versus static points and the viewpoint used for the qualitative results.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive review and the recommendation for major revision. The comments highlight important areas for improving clarity, reproducibility, and rigor in the presentation of Raymoval. We address each major comment below and commit to incorporating the suggested enhancements in the revised manuscript.

read point-by-point responses

Referee: [§3] §3 (Method): The central separation logic—bin-wise minimum-range comparison against the map's raycast first-hit distance, followed by the raycast consistency test—lacks any derivation, threshold values, or pseudocode. Without these, it is impossible to verify whether the test avoids the failure modes of slow-moving objects (whose ranges can be statistically indistinguishable from static structure within one scan) or noisy first-hit distances from prior map errors.

Authors: We agree that §3 would benefit from explicit details to support reproducibility and to address potential edge cases. The bin-wise minimum-range comparison detects dynamic points by identifying observations closer than the raycasted first-hit distance from the current static map. The subsequent raycast consistency test evaluates whether a point deviates from the expected static surface along the ray, providing separation even when single-scan ranges overlap. In the revised manuscript, we will add a derivation of the logic, specify all threshold values (including the range-difference threshold), include pseudocode for the full pipeline, and discuss mitigation of slow-moving objects and map noise via the consistency checks and spatial validation. revision: yes
Referee: [§4] §4 (Experiments): The quantitative claims of “lower residual dynamics and reduced over-removal” are presented without error bars, statistical significance tests, or ablations that isolate the contribution of the consistency test versus the spatial validation step. This makes it difficult to assess whether the reported gains are robust or dataset-specific.

Authors: We acknowledge that the experimental section would be strengthened by additional statistical analysis and component-wise ablations. The current results compare Raymoval against baselines on SemanticKITTI and the custom dataset using metrics for residual dynamics and over-removal. In the revision, we will include error bars on all quantitative plots, apply statistical significance tests (such as paired t-tests across sequences), and add ablation studies that separately evaluate the raycast consistency test and the spatial validation step. These changes will clarify the robustness and contribution of each component. revision: yes
Referee: [§3.3] §3.3 (Spatial validation): The spatial consistency validation is described only at a high level; no details are given on the neighborhood size, distance metric, or how it interacts with the preceding raycast test. This step is load-bearing for the “reduced over-removal” claim yet remains underspecified.

Authors: We agree that §3.3 requires more precise specification, as the spatial validation is key to minimizing erroneous removal of static points. This step performs a local neighborhood check in the projected grid to confirm that candidate dynamic labels form coherent clusters rather than isolated outliers. In the revised manuscript, we will detail the neighborhood size (a fixed 5×5 bin window in azimuth-elevation space), the distance metric (3D Euclidean distance to neighboring points), and the interaction rule: points that pass the raycast test but fail spatial consistency are re-labeled static. This will directly substantiate the reduced over-removal results. revision: yes

Circularity Check

0 steps flagged

No circularity: procedural geometric pipeline with no fitted predictions or self-referential reductions

full rationale

The described method is a sequence of explicit geometric operations—azimuth-elevation binning, minimum-range comparison against raycast first-hit distance, an unspecified consistency test, and spatial validation—none of which are shown to reduce by construction to parameters or labels fitted on the evaluation data. No equations equate outputs to inputs, no self-citation chains justify core claims, and the separation logic is presented as a direct application of raycasting principles rather than a derived or renamed result. The abstract and pipeline description remain self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The approach assumes that raycasting from the sensor pose accurately predicts first-hit distances in the static map and that dynamic objects produce detectable range discrepancies without requiring additional motion priors or multi-frame tracking.

axioms (2)

domain assumption Raycasting through the current static map yields reliable first-hit distances for comparison with new scans.
Invoked when the method compares bin-wise minimum range against the map's first-hit distance.
domain assumption Dynamic points are consistently closer than the static map prediction in at least one viewing direction.
Underlying the raycast consistency test that separates dynamic from static points.

pith-pipeline@v0.9.0 · 5445 in / 1443 out tokens · 27540 ms · 2026-05-12T01:45:52.772406+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

19 extracted references · 19 canonical work pages

[1]

IEEE Robot

Lim, H., Hwang, S., Myung, H.: ERASOR: Egocentric ratio of pseudo occupancy- based dynamic object removal for static 3D point cloud map building. IEEE Robot. Autom. Lett. 6(2) (2021) 2272–2279

work page 2021
[2]

Sung, C., Jeon, S., Lim, H., Myung, H.: What if there was no revisit? Large-scale graph-based SLAM with traffic sign detection in an HD map using LiDAR inertial odometry. Intell. Serv. Robot. 15(2) (2022) 161–170

work page 2022
[3]

IEEE Robot

Chen, X., Mersch, B., Nunes, L., Marcuzzi, R., Vizzo, I., Behley, J., Stachniss, C.: Automatic labeling to generate training data for online LiDAR-based moving object segmentation. IEEE Robot. Autom. Lett. 7(3) (2022) 6107–6114

work page 2022
[4]

In: Proc

Lim, H., Nunes, L., Mersch, B., Chen, X., Behley, J., Myung, H., Stachniss, C.: Instance-aware robust 3D mapping of the static world in dynamic scenes. In: Proc. Robotics: Science and Systems (RSS) (2023)

work page 2023
[5]

In: Proc

Maddern, W., Pascoe, G., Newman, P.: Leveraging experience for large-scale LiDAR localisation in changing cities. In: Proc. IEEE Int. Conf. Robot. Autom. (ICRA) (2015) 1684–1691

work page 2015
[6]

Wei, C., Wang, R., Wu, T., Chen, T., Fang, Y., Fu, H.: Plane-based scan registration with moving vehicles exclusion. Robot. Auton. Syst. 83 (2016) 261–274 12 D. Kim et al

work page 2016
[7]

Hornung, A., Wurm, K.M., Bennewitz, M., Stachniss, C., Burgard, W.: OctoMap: An efficient probabilistic 3D mapping framework based on octrees. Auton. Robots 34 (2013) 189–206

work page 2013
[8]

In: Proc

Kim, G., Kim, A.: Remove, then revert: Static point cloud map construction using multiresolution range images. In: Proc. IEEE/RSJ Int. Conf. Intell. Robots Syst. (IROS) (2020) 10758–10765

work page 2020
[9]

IEEE Robot

Mersch,B.,Chen,X.,Vizzo,I.,Nunes,L.,Behley,J.,Stachniss,C.:Recedingmoving object segmentation in 3D LiDAR data using sparse 4D convolutions. IEEE Robot. Autom. Lett. 7(3) (2022) 7503–7510

work page 2022
[10]

Jang, S., Oh, M., Yu, B., Nahrendra, I. M. A., Lee, S., Lim, H., and Myung, H.: TOSS:Real-timetrackingandmovingobjectsegmentationforstaticscenemapping. In: Proc. Int. Conf. Robot Intell. Technol. Appl. (RiTA) (2023)

work page 2023
[11]

IEEE Robot

Schauer, J., Nüchter, A.: The Peopleremover—Removing dynamic objects from 3D point cloud data by traversing a voxel occupancy grid. IEEE Robot. Autom. Lett. 3(3) (2018) 1679–1686

work page 2018
[12]

In: Proc

Lim, H., Jang, S., Mersch, B., Behley, J., Myung, H., Stachniss, C.: HeLiMOS: A dataset for moving object segmentation in 3D point clouds from heterogeneous LiDAR sensors. In: Proc. IEEE/RSJ Int. Conf. Intell. Robots Syst. (IROS) (2024) 14087-14094

work page 2024
[13]

Xiao, W., Vallet, B., Brédif, M., Paparoditis, N.: Street environment change detec- tion from mobile laser scanning point clouds. ISPRS J. Photogramm. Remote Sens. 107 (2015) 38–49

work page 2015
[14]

In: Proc

Hahnel, D., Schulz, D., Burgard, W.: Map building with mobile robots in populated environments. In: Proc. IEEE/RSJ Int. Conf. Intell. Robots Syst. (IROS) (2002) 496–501

work page 2002
[15]

IEEE Robot

Nunes, L., Chen, X., Marcuzzi, R., Osep, A., Leal-Taix’e, L., Stachniss, C., Behley, J.: Unsupervised class-agnostic instance segmentation of 3D LiDAR data for au- tonomous vehicles. IEEE Robot. Autom. Lett. 7(4) (2022) 8713–8720

work page 2022
[16]

IEEE Robot

Chen, X., Li, S., Mersch, B., Wiesmann, L., Gall, J., Behley, J., Stachniss, C.: Mov- ing object segmentation in 3D LiDAR data: A learning-based approach exploiting sequential data. IEEE Robot. Autom. Lett. 6 (2021) 6529–6536

work page 2021
[17]

In: Proc

Behley,J.,Garbade,M.,Milioto,A.,Quenzel,J.,Behnke,S.,Stachniss,C.,Gall,J.: SemanticKITTI: A dataset for semantic scene understanding of LiDAR sequences. In: Proc. IEEE/CVF Int. Conf. on Computer Vision (ICCV) (2019) 9297–9307

work page 2019
[18]

IEEE Robot

Marsim, K.C., Oh, M., Yu, B., Lee, S., Nahrendra, I.M.A., Lim, H., Myung, H.: LVI-Q: Robust LiDAR-visual-inertial-kinematic odometry for quadruped robots us- ing tightly-coupled and efficient alternating optimization. IEEE Robot. Autom. Lett. 10(10) (2025) 10050–10057

work page 2025
[19]

M., Park, J., Kim, J., Lee, J., Myung, H.: TRAVEL: Traversable ground and above-ground object segmentation using graph representation of 3D LiDAR scans

Oh, M., Jung, E., Lim, H., Song, W., Hu, S., Lee, E. M., Park, J., Kim, J., Lee, J., Myung, H.: TRAVEL: Traversable ground and above-ground object segmentation using graph representation of 3D LiDAR scans. IEEE Robot. Autom. Lett. 7(3) (2022) 7255–7262

work page 2022

[1] [1]

IEEE Robot

Lim, H., Hwang, S., Myung, H.: ERASOR: Egocentric ratio of pseudo occupancy- based dynamic object removal for static 3D point cloud map building. IEEE Robot. Autom. Lett. 6(2) (2021) 2272–2279

work page 2021

[2] [2]

Sung, C., Jeon, S., Lim, H., Myung, H.: What if there was no revisit? Large-scale graph-based SLAM with traffic sign detection in an HD map using LiDAR inertial odometry. Intell. Serv. Robot. 15(2) (2022) 161–170

work page 2022

[3] [3]

IEEE Robot

Chen, X., Mersch, B., Nunes, L., Marcuzzi, R., Vizzo, I., Behley, J., Stachniss, C.: Automatic labeling to generate training data for online LiDAR-based moving object segmentation. IEEE Robot. Autom. Lett. 7(3) (2022) 6107–6114

work page 2022

[4] [4]

In: Proc

Lim, H., Nunes, L., Mersch, B., Chen, X., Behley, J., Myung, H., Stachniss, C.: Instance-aware robust 3D mapping of the static world in dynamic scenes. In: Proc. Robotics: Science and Systems (RSS) (2023)

work page 2023

[5] [5]

In: Proc

Maddern, W., Pascoe, G., Newman, P.: Leveraging experience for large-scale LiDAR localisation in changing cities. In: Proc. IEEE Int. Conf. Robot. Autom. (ICRA) (2015) 1684–1691

work page 2015

[6] [6]

Wei, C., Wang, R., Wu, T., Chen, T., Fang, Y., Fu, H.: Plane-based scan registration with moving vehicles exclusion. Robot. Auton. Syst. 83 (2016) 261–274 12 D. Kim et al

work page 2016

[7] [7]

Hornung, A., Wurm, K.M., Bennewitz, M., Stachniss, C., Burgard, W.: OctoMap: An efficient probabilistic 3D mapping framework based on octrees. Auton. Robots 34 (2013) 189–206

work page 2013

[8] [8]

In: Proc

Kim, G., Kim, A.: Remove, then revert: Static point cloud map construction using multiresolution range images. In: Proc. IEEE/RSJ Int. Conf. Intell. Robots Syst. (IROS) (2020) 10758–10765

work page 2020

[9] [9]

IEEE Robot

Mersch,B.,Chen,X.,Vizzo,I.,Nunes,L.,Behley,J.,Stachniss,C.:Recedingmoving object segmentation in 3D LiDAR data using sparse 4D convolutions. IEEE Robot. Autom. Lett. 7(3) (2022) 7503–7510

work page 2022

[10] [10]

Jang, S., Oh, M., Yu, B., Nahrendra, I. M. A., Lee, S., Lim, H., and Myung, H.: TOSS:Real-timetrackingandmovingobjectsegmentationforstaticscenemapping. In: Proc. Int. Conf. Robot Intell. Technol. Appl. (RiTA) (2023)

work page 2023

[11] [11]

IEEE Robot

Schauer, J., Nüchter, A.: The Peopleremover—Removing dynamic objects from 3D point cloud data by traversing a voxel occupancy grid. IEEE Robot. Autom. Lett. 3(3) (2018) 1679–1686

work page 2018

[12] [12]

In: Proc

Lim, H., Jang, S., Mersch, B., Behley, J., Myung, H., Stachniss, C.: HeLiMOS: A dataset for moving object segmentation in 3D point clouds from heterogeneous LiDAR sensors. In: Proc. IEEE/RSJ Int. Conf. Intell. Robots Syst. (IROS) (2024) 14087-14094

work page 2024

[13] [13]

Xiao, W., Vallet, B., Brédif, M., Paparoditis, N.: Street environment change detec- tion from mobile laser scanning point clouds. ISPRS J. Photogramm. Remote Sens. 107 (2015) 38–49

work page 2015

[14] [14]

In: Proc

Hahnel, D., Schulz, D., Burgard, W.: Map building with mobile robots in populated environments. In: Proc. IEEE/RSJ Int. Conf. Intell. Robots Syst. (IROS) (2002) 496–501

work page 2002

[15] [15]

IEEE Robot

Nunes, L., Chen, X., Marcuzzi, R., Osep, A., Leal-Taix’e, L., Stachniss, C., Behley, J.: Unsupervised class-agnostic instance segmentation of 3D LiDAR data for au- tonomous vehicles. IEEE Robot. Autom. Lett. 7(4) (2022) 8713–8720

work page 2022

[16] [16]

IEEE Robot

Chen, X., Li, S., Mersch, B., Wiesmann, L., Gall, J., Behley, J., Stachniss, C.: Mov- ing object segmentation in 3D LiDAR data: A learning-based approach exploiting sequential data. IEEE Robot. Autom. Lett. 6 (2021) 6529–6536

work page 2021

[17] [17]

In: Proc

Behley,J.,Garbade,M.,Milioto,A.,Quenzel,J.,Behnke,S.,Stachniss,C.,Gall,J.: SemanticKITTI: A dataset for semantic scene understanding of LiDAR sequences. In: Proc. IEEE/CVF Int. Conf. on Computer Vision (ICCV) (2019) 9297–9307

work page 2019

[18] [18]

IEEE Robot

Marsim, K.C., Oh, M., Yu, B., Lee, S., Nahrendra, I.M.A., Lim, H., Myung, H.: LVI-Q: Robust LiDAR-visual-inertial-kinematic odometry for quadruped robots us- ing tightly-coupled and efficient alternating optimization. IEEE Robot. Autom. Lett. 10(10) (2025) 10050–10057

work page 2025

[19] [19]

M., Park, J., Kim, J., Lee, J., Myung, H.: TRAVEL: Traversable ground and above-ground object segmentation using graph representation of 3D LiDAR scans

Oh, M., Jung, E., Lim, H., Song, W., Hu, S., Lee, E. M., Park, J., Kim, J., Lee, J., Myung, H.: TRAVEL: Traversable ground and above-ground object segmentation using graph representation of 3D LiDAR scans. IEEE Robot. Autom. Lett. 7(3) (2022) 7255–7262

work page 2022