Design and Evaluation of Two Spherical Systems for Mobile 3D Mapping

Andreas N\"uchter; Fabian Arzberger; Marawan Khalil

REVIEW 3 major objections 5 minor 17 references

Design and Evaluation of Two Spherical Systems for Mobile 3D Mapping

T0 review · 3 major / 5 minor · reviewed 2026-08-04 · deepseek-v4-flash

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

Pith's one-line read This paper claims that off-the-shelf LiDAR-inertial odometry fails to produce globally consistent 3D maps when mounted on rolling spherical robots, because the high-dynamic multi-axis rotations exceed what standard motion models are designe

desk verdict A credible empirical warning that spherical rolling breaks LIO, but the quantitative evaluation is under-specified on the crucial map-to-ground-truth alignment. read the letter →

arxiv 2509.10032 v1 pith:GT23DRIJ submitted 2025-09-12 cs.RO

Marawan Khalil , Fabian Arzberger , Andreas Nüchter This is my paper

classification cs.RO

keywords sphericalrobotsmobilemappingLiDAR-inertialodometrySLAM3Dpointcloudrollinglocomotionpendulum-drivenrobotdriftevaluation

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

The paper builds two spherical robots for 3D mapping — one manually rolled, one driven by an internal pendulum — both carrying a solid-state LiDAR and running state-of-the-art LiDAR-inertial odometry (LIO) on compact onboard hardware. It then compares the produced point clouds to a high-precision terrestrial laser scan. The central finding is that spherical rolling motion causes aggressive rotation about all axes, which standard LIO motion models do not anticipate, leading to bent maps, accumulating drift, and sometimes unrecoverable failures. The best accuracy came from the manually rolled sphere with one of the LIO algorithms: a mean error of 9.60 cm and RMSE of 13.09 cm, which the authors attribute to the LiDAR being placed closer to the sphere's center of rotation.

What carries the argument

The central mechanism is the rolling contact between the spherical shell and the ground, which forces the internal sensor to undergo large rotations about all three principal axes simultaneously, with high angular velocity and frequent reversals. Standard LIO motion models assume more constrained rotational dynamics typical of wheeled, legged, or handheld platforms; the mismatch accumulates as drift and visible bending in the reconstructed point clouds. The evaluation uses point-to-point RMSE between each LIO map and a ground-truth terrestrial laser scan to quantify accuracy.

What would settle it

A reader could re-run the point-cloud comparison without any global registration step, using only raw, time-synchronized poses to compute RMSE. If the resulting errors are much larger than 13 cm or the maps show large discontinuities, the 'globally inconsistent' claim is confirmed; if the errors shrink to the reported values only after alignment, the claim is weakened.

Watch

Extended reading notes

Core claim

The paper establishes, on its own terms, that off-the-shelf LiDAR-inertial odometry is not reliable from a rolling spherical platform. The high dynamic, rotationally aggressive motion of the rolling shell violates the assumptions built into the algorithms' motion models, causing global map inconsistency and occasional unrecoverable drift. A secondary discovery is that the non-actuated sphere outperforms the pendulum-actuated one, which the authors explain by sensor placement: the LiDAR sits closer to the center of the smaller, simpler sphere, reducing the lever arm between the true center of rotation and the sensor. The reported error metrics support the conclusion that spherical mapping nee

Load-bearing premise

The accuracy claim rests on the assumption that the comparison to the ground-truth point cloud did not apply a global rigid registration that would hide large-scale drift; the paper does not state whether such a registration was applied before computing the RMSE in Eq. (1) in Section V-A.

Editorial extensions

If this is right

Spherical robots cannot simply reuse standard LIO pipelines; the motion models must incorporate the sphere's nonholonomic rolling dynamics to avoid global inconsistency.
Placing the LiDAR closer to the sphere's center of rotation is a practical design choice that materially improves mapping accuracy, as demonstrated by the better results from the non-actuated sphere.
The actuated, pendulum-driven sphere did not achieve better mapping than the manually rolled one, suggesting that actuation alone does not solve the problem unless the motion model is also corrected.
The reported 'bending' of point clouds points to a systematic rotational drift that could serve as a target for specialized filtering or full-state optimization.
The failed runs, where the algorithm could not recover after fast motion, indicate that a phase of aggressive rotation can permanently corrupt the estimator, not just degrade it temporarily.

Reading between the lines

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

The paper leaves implicit that its accuracy numbers depend on how the alignment to ground truth was performed; if the 3DTK processing included a global rigid registration before computing RMSE, the reported errors would reflect only local residuals and the 'globally inconsistent' claim would be under-supported.
If sensor placement near the center is indeed the key factor, then a sphere with the LiDAR mounted exactly at the geometric center, or with a counter-rotating gimbal, should produce near-pure rotational scans and substantially better LIO accuracy — a directly testable extension.
The bending artifact should be reproducible in simulation by feeding a rigid-body rolling trajectory into any LIO estimator while holding all other sensor parameters fixed; if the bending reappears in simulation, it confirms that the motion model, not sensor noise, is the cause.
A neighbouring problem this work connects to is robust pose estimation under highly dynamic rotation without relying on magnetometers — a scenario common in planetary exploration, where magnetic-field data is unavailable.

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.

Desk Editor's Note

read the letter

Quick take: they built two small spherical mapping robots, one human-rolled and one pendulum-actuated, and ran three off-the-shelf LIO algorithms (FAST-LIO2, FAST-LIVO2, DLIO) on a Pi 5. The hardware integration is real, and the qualitative finding—rolling dynamics produce bent maps and occasional unrecoverable drift—is credible and useful. But the numbers in Table III are less reliable than they look because the map-to-ground-truth alignment is unspecified, and the two platforms are not run under comparable conditions.

What's actually new: the first online LIO on an actuated spherical platform, as far as they know. The design is neat, with the LiDAR, IMU, Pi 5, and the pendulum control system all integrated into a 20 cm shell. Code is on GitHub, and they compare against Riegl TLS ground truth, which is more than most hardware papers bother to do. The paper is also honest about failures: they show DLIO completely failing on the non-actuated sphere and admit the actuated sphere's maps bend.

Soft spots: the evaluation is single-run per algorithm/platform, with no conventional wheeled baseline. The manual motion of the non-actuated sphere versus the teleoperated motion of the actuated sphere makes the comparison between the two platforms uncontrolled. The bigger issue is the alignment step. They say they used 3DTK to process the point clouds and then computed RMSE against the TLS map, but they never say whether they rigidly registered each SLAM map to ground truth first. If they did, the RMSE only measures local residuals after the global drift has been absorbed, so the abstract's claim of "globally inconsistent maps" rests on the figures, not on the table. They also don't report raw odometry drift, loop-closure residuals, or end-point error, which would be the direct evidence. Minor: DLIO is listed as failed for the non-actuated run, so the comparison is missing a cell.

None of this kills the paper. The core warning is plausible and the hardware contribution stands. But the quantitative claims need tightening: specify the registration procedure, report drift metrics that don't depend on alignment, and ideally add repeated runs and a wheeled baseline.

Who it's for: spherical robotics people and LIO developers. It deserves a serious referee, but the evaluation section should come back revised. I'd send it to review, not desk reject.

Referee Report

3 major / 5 minor

Summary. The paper describes two spherical 3D mapping robots, a non-actuated sphere rolled manually and a pendulum-actuated sphere controlled by an operator, both carrying a Livox Mid-360 LiDAR, a BNO-085 IMU, and a Raspberry Pi 5. The authors integrate FAST-LIO2, FAST-LIVO2 (LIO-only mode), and DLIO, and evaluate the resulting point-cloud maps against ground truth from a Riegl VZ-400 terrestrial laser scanner. The principal claim is that state-of-the-art LIO algorithms degrade on spherical platforms because of the high angular dynamics of rolling locomotion, producing globally inconsistent and sometimes unrecoverable drift. The non-actuated sphere with FAST-LIO2 is reported as the best configuration, with mean error 9.60 cm and RMSE 13.09 cm. The paper also contributes open-source ROS2 software and is, per the authors, the first self-actuated spherical robot performing online LIO.

Significance. If the central claim were conclusively established, the paper would be a useful systems contribution: it demonstrates that off-the-shelf LIO motion models, designed for wheeled or handheld platforms, are insufficient for spherical rolling dynamics, and it motivates motion-model redesign for spherical robots. The authors are to be credited for building two complete hardware prototypes, for running three modern LIO algorithms on resource-constrained onboard hardware, for releasing the source code, and for using a high-precision TLS as external ground truth. The qualitative evidence of bent planes and unrecoverable drift in Figs. 7 and 8 is visually plausible. However, the quantitative evaluation as currently presented does not support the global-drift claim because the map alignment procedure is undisclosed, the two platforms are moved differently, and every algorithm/platform combination is run only once. The paper's value currently rests more on the systems and observations than on a validated measurement of mapping accuracy.

major comments (3)

[Section V-A, Eq. (1), Table III] The manuscript states that 3DTK was used to process the point-clouds, but it never specifies whether a global rigid registration (e.g., ICP) was applied to align each SLAM map to the Riegl ground-truth map before computing the RMSE in Eq. (1). If a global registration was used, it would absorb large-scale drift and bending, and the RMSE would reflect only local residuals. In that case the abstract's claim of 'globally inconsistent maps' and the favorable 9.60 cm mean error / 13.09 cm RMSE for non-actuated FAST-LIO2 would not be supported by Table III. Please state the exact alignment pipeline, and report at least one drift metric that does not allow a global best-fit transform (e.g., end-point odometry error, trajectory error against ground truth, or map-to-ground-truth residual under the identity transform).
[Section V-A, Section V-B, Table III] The evaluation has multiple uncontrolled confounds that prevent attributing the results to spherical locomotion. The non-actuated sphere is moved manually by hand and foot, while the actuated sphere is teleoperated with a controller; the shell diameters, LiDAR offsets from the center, and path execution also differ. Moreover, each algorithm/platform combination is run only once, so Table III reports the statistics of a single point-cloud per configuration, not a statistically grounded comparison. Without repeated trials, a conventional wheeled/handheld baseline on the same sensor suite and paths, or at least a fixed/permitted trajectory, the conclusion that 'the performance of state-of-the-art LIO algorithms deteriorates due to the high dynamic movement introduced by the spherical locomotion' is not established by the reported numbers. Please add repeated runs and controlled comparisons.
[Section V-B, Table III, Fig. 7] There is an internal inconsistency in the handling of DLIO for the non-actuated sphere. Table III lists non-actuated DLIO as '-' (failed), yet Fig. 7a shows a 'DLIO point-cloud' from the non-actuated sphere, and Section V-B2 states that 'We were unable to obtain a satisfactory map using DLIO from the non-actuated sphere.' Please clarify what 'failed' means, whether this case is excluded from Table III, why Fig. 7a is shown, and how this failure supports the claim of 'unrecoverable drift.' The boundary between a hardware/software failure, a control failure, and an algorithm's inability to recover from fast angular motion should be stated explicitly.

minor comments (5)

[Section II heading] The heading 'STATE OF THEART' should be 'STATE OF THE ART.'
[Section I] Typo: 'UA Vs' should be 'UAVs.'
[Table I] The entry 'V oltage regulators' has a stray space; also consider using the same table formatting as Table II for readability.
[Section V-A] The text says the non-actuated and actuated robots were moved 'along similar paths,' but the paths are not quantified or shown. If path equivalence is important, provide trajectory overlays or a quantitative path similarity measure.
[Section IV-B] The PID gains for the pendulum pitch controller are not reported, which hampers reproducibility. Even approximate gains or a note that they were tuned empirically would help.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the central claim rests on external ground-truth evaluation and open-source LIO algorithms.

full rationale

This is an empirical system-evaluation paper, not a derivation. The central claim—that state-of-the-art LIO algorithms deteriorate under spherical-locomotion dynamics—is supported by comparing point clouds produced by three external, published LIO algorithms (FAST-LIO2, FAST-LIVO2 in LIO mode, DLIO) against an independent Riegl VZ-400 TLS ground truth. No parameter is fitted to the ground-truth map and then renamed a prediction. The authors' previous works [4, 6, 11] are cited for hardware inspiration and as related spherical-SLAM systems, but none is load-bearing for the reported deterioration; the external algorithms and the Riegl ground truth provide independent content. The only notable methodological concern is that the paper says 3DTK was used to process the point clouds without stating whether a rigid registration (e.g., ICP) was applied before computing Eq. (1); if such registration was used, the RMSE might understate global drift. That is a validity/measurement limitation, not circularity, because it does not make the result equivalent to its input by construction, nor does it involve a fitted parameter being presented as a prediction. No circular step can be quoted from the paper, so the circularity score is 0.

Assumptions & free parameters 1 free parameters · 5 assumptions · 0 invented entities

The mapping evaluation rests on several unstated premises about ground-truth accuracy, the alignment process, sensor calibration, environment stability, and LIO parameter settings. None of these are demonstrated in the paper, and each could affect the reported RMSE values and the central conclusion that spherical motion is the cause of LIO degradation.

free parameters (1)

PID gains for pendulum pitch control = not reported
The actuated sphere's pitch control is a discrete PID controller (Section IV-B); the gains are chosen by hand but not listed. They determine how aggressively the sphere responds to input and therefore influence the motion dynamics experienced by the LIO algorithms. The paper does not show that the results are robust to these gains.

assumptions (5)

domain assumption The Riegl VZ-400 TLS map is a sufficiently accurate ground truth for the evaluated indoor areas.
Used as the reference map in Section V-A and Fig. 5; no separate accuracy or uncertainty estimate for the TLS map is provided.
domain assumption The 3DTK point-cloud alignment does not mask the global drift of the SLAM maps.
The RMSE in Eq. (1) is computed after 3DTK processing; if a global ICP registration was applied, the comparison could remove exactly the global inconsistency the paper claims. The alignment procedure is not specified.
domain assumption The LiDAR-IMU extrinsics and synchronization are accurate enough for the LIO algorithms.
FAST-LIO2, FAST-LIVO2, and DLIO rely on precise relative poses between the Livox Mid-360 and the BNO-085 IMU; the paper does not report calibration values or a calibration procedure.
domain assumption The mapping environment was static between the TLS scan and each robot run.
The comparisons assume identical geometry; the paper does not mention any handling of dynamic objects or changes between runs.
domain assumption The default or reasonably tuned parameters of the LIO algorithms are valid for the spherical platform.
The paper does not describe parameter choices; if the algorithms were not tuned, the observed failures could be due to misconfiguration rather than spherical motion.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Design and Evaluation of Two Spherical Systems for Mobile 3D Mapping." pith.science (2026). https://pith.science/paper/GT23DRIJ

@misc{pith2026250910032,
  author       = {Pith},
  title        = {Pith review of: Design and Evaluation of Two Spherical Systems for Mobile 3D Mapping},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GT23DRIJ}},
  note         = {Machine review of arXiv:2509.10032}
}

read the original abstract

Spherical robots offer unique advantages for mapping applications in hazardous or confined environments, thanks to their protective shells and omnidirectional mobility. This work presents two complementary spherical mapping systems: a lightweight, non-actuated design and an actuated variant featuring internal pendulum-driven locomotion. Both systems are equipped with a Livox Mid-360 solid-state LiDAR sensor and run LiDAR-Inertial Odometry (LIO) algorithms on resource-constrained hardware. We assess the mapping accuracy of these systems by comparing the resulting 3D point-clouds from the LIO algorithms to a ground truth map. The results indicate that the performance of state-of-the-art LIO algorithms deteriorates due to the high dynamic movement introduced by the spherical locomotion, leading to globally inconsistent maps and sometimes unrecoverable drift.

Figures

Figures reproduced from arXiv: 2509.10032 by the authors.

**Figure 1.** (Left:) 20 cm diameter actuated sphere. (Right:) 16 cm diameter non-actuated sphere. A video where both spheres are moving is available at https://youtube.com/shorts/ lxTF85HK-zY. [4]–[6]. However, the rapid evolution of LiDAR–inertial algorithms and modern hardware platforms presents new opportunities for spherical SLAM, while the integration of actuation with mapping and localization remains largely unexplored. … view at source ↗

**Figure 2.** Schematic model and design of the non-actuated [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 4.** State chart of the Sphere System showing the control and mapping subsystems. [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗

Figures from the paper (4 more)

**Figure 7.** Figure 7: Example failed cases from the non-actuated sphere [PITH_FULL_IMAGE:figures/full_fig_p005_7.png]

**Figure 5.** Figure 5: Evaluation setup, showing the Riegl VZ-400 terrestrial [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]

**Figure 6.** Figure 6: Bird’s-eye view of a cross section of resulting point [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]

**Figure 9.** Figure 9: Point-cloud Results and Error Distribution Analysis. A fly through video of the point-clouds is available at https: [PITH_FULL_IMAGE:figures/full_fig_p006_9.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

17 extracted references · 1 canonical work pages

[1]

Roboball: An all-terrain spherical robot with a pressurized shell,

M. Oevermann, D. Pravecek, G. Jibrail, R. Jangale, and R. O. Ambrose, “Roboball: An all-terrain spherical robot with a pressurized shell,” in2024 IEEE International Conference on Robotics and Automation (ICRA), 2024, pp. 13 502–13 508

2024
[2]

Spherical robot: A novel robot for exploration in harsh unknown environments,

W. Ren, Y . Wang, H. Liu, S. Jin, Y . Wang, Y . Liu, Z. Zhang, T. Hu, and G. Li, “Spherical robot: A novel robot for exploration in harsh unknown environments,”IET Cyber-Systems and Robotics, vol. 5, no. 4, p. e12099, 2023. [Online]. Available: https://ietresearch. onlinelibrary.wiley.com/doi/abs/10.1049/csy2.12099

work page doi:10.1049/csy2.12099 2023
[3]

Impedance esti- mation and motion control of a pendulum-driven spherical robot,

H. Kolbari, A. Ahmadi, M. Bahrami, and F. Janati, “Impedance esti- mation and motion control of a pendulum-driven spherical robot,” pp. 6–11, 2018

2018
[4]

Delta- and kalman- filter designs for multi-sensor pose estimation on spherical mobile mapping systems,

F. Arzberger, T. Schubert, F. Wiecha, J. Zevering, J. Rothe, D. Borrmann, S. Montenegro, and A. Nüchter, “Delta- and kalman- filter designs for multi-sensor pose estimation on spherical mobile mapping systems,”Robotics and Autonomous Systems, vol. 184, p. 104852, 2025. [Online]. Available: https://www.sciencedirect.com/ science/article/pii/S0921889024002367

2025
[5]

Daedalus- descent and exploration in deep autonomy of lava underground struc- tures,

A. P. Rossi, F. Maurelli, V . Unnithan, H. Dreger, K. Mathewos, N. Prad- han, D.-A. Corbeanu, R. Pozzobon, A. Bredenbecket al., “Daedalus- descent and exploration in deep autonomy of lava underground struc- tures,” 2021

2021
[6]

Arzberger, A

F. Arzberger, A. Bredenbeck, J. Zevering, D. Borrmann, and A. Nüchter,Towards spherical robots for mobile mapping in human made environments, 2021, vol. 1. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S2667393221000041

2021
[7]

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,” 2021. [Online]. Available: https: //arxiv.org/abs/2107.06829

arXiv 2021
[8]

Fast-livo2: Fast, direct lidar-inertial-visual odometry,

C. Zheng, W. Xu, Z. Zou, T. Hua, C. Yuan, D. He, B. Zhou, Z. Liu, J. Lin, F. Zhu, Y . Ren, R. Wang, F. Meng, and F. Zhang, “Fast-livo2: Fast, direct lidar-inertial-visual odometry,” 2024. [Online]. Available: https://arxiv.org/abs/2408.14035

arXiv 2024

Show all 17 references

[9]

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,”
[10]

Zevering, A

J. Zevering, A. Bredenbeck, F. Arzberger, D. Borrmann, and A. Nüchter, L.U.N.A. - A Laser-Mapping Unidirectional Navigation Actuator, 03 2021, pp. 85–94

2021
[11]

Delta filter - robust visual-inertial pose estimation in real-time: A multi-trajectory filter on a spherical mobile mapping system,

F. Arzberger, F. Wiecha, J. Zevering, J. Rothe, D. Borrmann, S. Montene- gro, and A. Nüchter, “Delta filter - robust visual-inertial pose estimation in real-time: A multi-trajectory filter on a spherical mobile mapping system,” in2023 European Conference on Mobile Robots (ECMR...

2023
[12]

Controlling the locomotion of spherical robots or why bb-8 works,

P. Akella, O. O’Reilly, and K. Sreenath, “Controlling the locomotion of spherical robots or why bb-8 works,”Journal of Mechanisms and Robotics, vol. 11, 12 2018

2018
[13]

The concept of rod-driven locomotion for spherical lunar exploration robots,

J. Zevering, D. Borrmann, A. Bredenbeck, and A. Nuchter, “The concept of rod-driven locomotion for spherical lunar exploration robots,” 10 2022, pp. 5656–5663

2022
[14]

Neural network control for the linear motion of a spherical mobile robot,

Y . Cai, Q. Zhan, and X. Xi, “Neural network control for the linear motion of a spherical mobile robot,”International Journal of Advanced Robotic Systems, vol. 8, 01 2011

2011
[15]

3d mapping sphere ros2,

M. Khalil and JMUWRobotics, “3d mapping sphere ros2,” https:// github.com/JMUWRobotics/3D_Mapping_Sphere_ROS2, 2025, gitHub repository

2025
[16]

3DTK - The 3D Toolkit,

JMUWRobotics, “3DTK - The 3D Toolkit,” https://github.com/ JMUWRobotics/3DTK, 2025, gitHub repository

2025
[2023]

Available: https://arxiv.org/abs/2203.03749

[Online]. Available: https://arxiv.org/abs/2203.03749

arXiv

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.

[1] [1]

Roboball: An all-terrain spherical robot with a pressurized shell,

M. Oevermann, D. Pravecek, G. Jibrail, R. Jangale, and R. O. Ambrose, “Roboball: An all-terrain spherical robot with a pressurized shell,” in2024 IEEE International Conference on Robotics and Automation (ICRA), 2024, pp. 13 502–13 508

2024

[2] [2]

Spherical robot: A novel robot for exploration in harsh unknown environments,

W. Ren, Y . Wang, H. Liu, S. Jin, Y . Wang, Y . Liu, Z. Zhang, T. Hu, and G. Li, “Spherical robot: A novel robot for exploration in harsh unknown environments,”IET Cyber-Systems and Robotics, vol. 5, no. 4, p. e12099, 2023. [Online]. Available: https://ietresearch. onlinelibrary.wiley.com/doi/abs/10.1049/csy2.12099

work page doi:10.1049/csy2.12099 2023

[3] [3]

Impedance esti- mation and motion control of a pendulum-driven spherical robot,

H. Kolbari, A. Ahmadi, M. Bahrami, and F. Janati, “Impedance esti- mation and motion control of a pendulum-driven spherical robot,” pp. 6–11, 2018

2018

[4] [4]

Delta- and kalman- filter designs for multi-sensor pose estimation on spherical mobile mapping systems,

F. Arzberger, T. Schubert, F. Wiecha, J. Zevering, J. Rothe, D. Borrmann, S. Montenegro, and A. Nüchter, “Delta- and kalman- filter designs for multi-sensor pose estimation on spherical mobile mapping systems,”Robotics and Autonomous Systems, vol. 184, p. 104852, 2025. [Online]. Available: https://www.sciencedirect.com/ science/article/pii/S0921889024002367

2025

[5] [5]

Daedalus- descent and exploration in deep autonomy of lava underground struc- tures,

A. P. Rossi, F. Maurelli, V . Unnithan, H. Dreger, K. Mathewos, N. Prad- han, D.-A. Corbeanu, R. Pozzobon, A. Bredenbecket al., “Daedalus- descent and exploration in deep autonomy of lava underground struc- tures,” 2021

2021

[6] [6]

Arzberger, A

F. Arzberger, A. Bredenbeck, J. Zevering, D. Borrmann, and A. Nüchter,Towards spherical robots for mobile mapping in human made environments, 2021, vol. 1. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S2667393221000041

2021

[7] [7]

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,” 2021. [Online]. Available: https: //arxiv.org/abs/2107.06829

arXiv 2021

[8] [8]

Fast-livo2: Fast, direct lidar-inertial-visual odometry,

C. Zheng, W. Xu, Z. Zou, T. Hua, C. Yuan, D. He, B. Zhou, Z. Liu, J. Lin, F. Zhu, Y . Ren, R. Wang, F. Meng, and F. Zhang, “Fast-livo2: Fast, direct lidar-inertial-visual odometry,” 2024. [Online]. Available: https://arxiv.org/abs/2408.14035

arXiv 2024