pith. sign in

arxiv: 2605.02809 · v1 · submitted 2026-05-04 · 💻 cs.RO

LiDAR Teach, Radar Repeat: Robust Cross-Modal Navigation in Degenerate and Varying Environments

Renxiang Xiao , Yichen Chen , Yuanfan Zhang , Qianyi Shao , Yushuai Chen , Yuxuan Han , Yunjiang Lou , Liang Hu This is my paper

Pith reviewed 2026-05-08 18:02 UTC · model grok-4.3

classification 💻 cs.RO
keywords cross-modal registrationteach-and-repeat navigationLiDAR4D radarlong-term autonomyrobot localizationenvironmental robustnesssensor alignment
0
0 comments X

The pith

LiDAR teaches paths that radar can repeat with centimeter accuracy despite weather and structural changes.

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

The paper introduces LTR², a cross-modal teach-and-repeat system in which a robot captures a precise path map with LiDAR under favorable conditions and then follows the same path using only radar when the scene degrades. A Cross-Modal Registration network registers the sparse, noisy forward-looking radar scans to the dense LiDAR map by combining Doppler motion estimates with the physical relationships between LiDAR intensity returns and radar power density. An error-driven fine-tuning loop then updates the network on the fly from its own localization mistakes, allowing continued performance over months without external ground truth. Experiments across three platforms and 40 km of travel over six months, including smoke and night conditions, show the approach maintains centimeter-level accuracy where single-modality or prior methods degrade.

Core claim

By teaching with dense omnidirectional 3D LiDAR and repeating with sparse forward-looking 4D radar, the CMR network aligns the modalities through Doppler-based motion priors and the physical laws of sensor returns; adaptive fine-tuning then sustains alignment across static changes, delivering state-of-the-art registration and centimeter-level navigation over long-term, large-scale deployments without ground-truth labels.

What carries the argument

The Cross-Modal Registration (CMR) network, which aligns sparse noisy forward-looking 4D radar with dense omnidirectional 3D LiDAR by jointly using Doppler-based motion priors and the physical laws of LiDAR intensity and radar power density.

If this is right

  • The CMR network achieves state-of-the-art accuracy on public cross-modal registration benchmarks.
  • Centimeter-level localization holds across 40+ km of travel and six months on multiple robot platforms.
  • Performance remains robust under nighttime smoke, weather degradation, and static structural changes.
  • Error-driven adaptation preserves accuracy without requiring ground-truth labels during deployment.

Where Pith is reading between the lines

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

  • The same registration principle could transfer to other teach-and-repeat pairings, such as camera teaching followed by radar repeat, to lower hardware costs for long-term autonomy.
  • In environments with rapidly moving objects the current physical-law alignment may need extra motion filtering to stay reliable.
  • Testing at higher speeds or in larger outdoor areas would reveal whether the Doppler priors continue to provide enough constraint for registration.

Load-bearing premise

Doppler motion priors together with the physical laws linking LiDAR intensity to radar power density remain sufficient to produce accurate alignments even as the environment slowly changes over months.

What would settle it

A long-term repeat run in which localization error grows beyond a few centimeters after heavy rain or major structural alteration when the fine-tuning loop is disabled.

Figures

Figures reproduced from arXiv: 2605.02809 by Liang Hu, Qianyi Shao, Renxiang Xiao, Yichen Chen, Yuanfan Zhang, Yunjiang Lou, Yushuai Chen, Yuxuan Han.

Figure 1
Figure 1. Figure 1: Demonstration trajectories of the LTR2 system. (a) The repeat trajectory is covered by an unexpected smoke, which was not seen during trajectory teaching. (b) A nav￾igation demonstration with a color-coded error map visual￾izing deviations between teach-and-repeat paths, achieving centimeter-level accuracy on average over multi-kilometer trajectories. cost, T&R has been widely adopted for navigation in div… view at source ↗
Figure 1
Figure 1. Figure 1: Such a heterogeneous sensing configuration at dif view at source ↗
Figure 2
Figure 2. Figure 2: System overview of the LTR2 pipeline. The system operates in four phases. In the pre-training phase, a Cross-Modal Registration (CMR) network is trained using synchronized LiDAR and 4D Radar frames. Then the trained CMR network is frozen and used in the teaching and repeating phases. In the teaching phase, the teach graph is constructed, whose nodes contain LiDAR-frame observations and edges store relative… view at source ↗
Figure 3
Figure 3. Figure 3: Iterative node selection. (a) Initial node selection based on successful cross-modal registration between radar and LiDAR frames and LiDAR co-collected in the teaching phase. (b) Iterative node addition, where successive cubic Hermite curves refine the trajectory and new nodes are inserted in regions requiring higher spatial fidelity. The magnified view indicates the need for high-fidelity path interpolati… view at source ↗
Figure 4
Figure 4. Figure 4: Target Node selection during the repeating phase. Green triangles represent radar frames during repeat. Blue triangles denote LiDAR-based nodes obtained during the teaching phase via LiDAR odometry. Target node change indicates a switch to the subsequent teach node for the latter radar frame. The black line denotes the fine estimate after CMR between the radar frame and the LiDAR frame in the target node, … view at source ↗
Figure 5
Figure 5. Figure 5: Gaussian distribution of ATE under different environment changes. The black line is the linear decision boundary that isolates the error distribution due to static changes (red) from the other three clusters. Representative scene images for every class are shown on the right. the samples accumulated in the library are used to fine-tune the CMR network in a self-supervised manner. The overall procedure is s… view at source ↗
Figure 6
Figure 6. Figure 6: Positive and negative samples collected in sliding windows. Positive samples are segments where the repeat trajectory remains consistent with the teach trajectory, while negative samples correspond to windows exhibiting abnormal drift. The sliding-window evaluation partitions the trajectory into these two sample types for subsequent fine-tuning. as the drift metric. To identify segments exhibiting abnormal… view at source ↗
Figure 7
Figure 7. Figure 7: Data preprocessing for 4D radar and LiDAR. (a) Raw radar with power is aligned with the intensity of LiDAR and followed by two separate cylindrical projections in the intensity and geometry. (b) Raw LiDAR is cropped using the 4D radar FOV mask and then transformed into the cylindrical projection view at source ↗
Figure 8
Figure 8. Figure 8: Cumulative distribution heatmaps of intensity ver￾sus distance for the original radar (left), the pseudo radar aligned to LiDAR intensity via (22) (middle), and the LiDAR signals (right). With the intensity alignment, the discrepancy in intensity distributions between pseudo radar and LiDAR is significantly less than that between the original radar and LiDAR. where PLiDAR is the received power density, Pt … view at source ↗
Figure 9
Figure 9. Figure 9: Overall structure of the CMR network. Preprocessed geometry and intensity images are fed into separate branches, and features are extracted by a Swin Transformer. The Doppler-guided association module exploits the coarse estimate to generate cross-modal correspondences. Then the cross-modal attention and progressive optimization refine alignment, and the final registration is achieved through the pose fusi… view at source ↗
Figure 10
Figure 10. Figure 10: The experimental platform. function and introduce two learnable parameters sq and st. For each layer, the training loss function is expressed as follows: L = || qˆ ||qˆ||−q||·exp(−sq)+sq+||ˆt−t||·exp(−st)+st, (31) where q and qˆ represent the ground truth and estimated rotation vector, t and ˆt are the ground truth and estimated translation vector. Our network provides registration results at four differe… view at source ↗
Figure 11
Figure 11. Figure 11: The point cloud overlay via registration in the VoD dataset. Grey and color points are LiDAR and radar point cloud, respectively. Zoomed color overlays highlight CMR network’s superior alignment accuracy over baseline methods and robustness to inaccurate pose priors. TABLE V: Difficulty Metrics for the Teach Paths Tag Range Elevation Pitch Min radius Challenges A 1325.8m 9.6m 8◦ 1.32m Long range B 516.7m … view at source ↗
Figure 12
Figure 12. Figure 12: Display of the real-world experimental trajectory on satellite map, which illustrates multiple teach-and-repeat trajectories across diverse environments within a university campus. localization on the map and trajectory tracking during the repeating phase; • LiDAR T&R navigation: Both GICP and learning-based Regformer [58] methods are implemented for point cloud registration; • Radar T&R navigation: Three… view at source ↗
Figure 13
Figure 13. Figure 13: Illustration of the repeat error in trajectory A. The color-coded path shows centimeter-level positioning errors between the teach-and-repeat trajectories. a ROS1-based driver is provided for our used 4D radar, we select CT-ICP [13] from the registration module of VT&R3 and replace it with our localization block for comparative experiments. Furthermore, our CMR network without the radar power LiDAR intens… view at source ↗
Figure 14
Figure 14. Figure 14: Screenshots of different navigation methods in four representative and challenging scenarios. The white trajectory indicates the teach path, while the colored arrows represent projections of run-time positions onto the current￾view image. Compared to representative baselines ADPGICP (Radar T&R), LIO-SAM (SLAM-based), CT-ICP (Cross-modal T&R), Regformer (LiDAR T&R), and our LTR2 system avoids trajectory re… view at source ↗
Figure 15
Figure 15. Figure 15: Visualization of storage requirements of different node selection strategies on Trajectory A. with increased node density. The larger EWAerror means the more accurate localization with less nodes. As shown in Tab. IX, our adaptive node selection method achieves the best trade-off between navigation accuracy and storage burden view at source ↗
Figure 16
Figure 16. Figure 16: Perception comparison of different sensors when repeating in a heavy smoke environment. LiDAR data experiences severe perception degradation, while 4D Radar maintains robust perception. The visual image is only for visualization, not used in navigation. Finetune1 and Finetune2 were recorded two months after their respective teaching phases. As illustrated in the upper row of view at source ↗
Figure 18
Figure 18. Figure 18: Tab. X indicates that LTR2 after finetuning provides a more accurate trajectory repeat, maintaining the reliability of navigation against static environment changes. A deeper insight into the adaptive finetuning is also provided in the lower rows of view at source ↗
Figure 18
Figure 18. Figure 18: Trajectory evaluation of fine-tuning. Larger lo￾calization errors occur against static environmental changes, and are mitigated after the fine-tuning. The red and blue shadowed regions denote where negative and positive samples are collected, respectively. biped robot, as shown in view at source ↗
Figure 19
Figure 19. Figure 19: Visualization of cross-platform navigation. strate that the proposed CMR network can support cross￾platform route reuse at the localization and route-following level, showing robustness to substantial differences in teach￾side sensor configuration and platform. The full cross-platform demonstration is also provided in the supplementary video. VI. CONCLUSION In this paper, we propose a novel omnidirectiona… view at source ↗
read the original abstract

Long-term autonomy requires robust navigation in environments subject to dynamic and static changes, as well as adverse weather conditions. Teach-and-Repeat (T\&R) navigation offers a reliable and cost-effective solution by avoiding the need for consistent global mapping; however, existing T\&R systems lack a systematic solution to tackle various environmental variations such as weather degradation, ephemeral dynamics, and structural changes. This work proposes LTR$^2$, the first cross-modal, cross-platform LiDAR-Teach-and-Radar-Repeat system that systematically addresses these challenges. LTR$^2$ leverages LiDAR during the teaching phase to capture precise structural information under normal conditions and utilizes 4D millimeter-wave radar during the repeating phase for robust operation under environmental degradations. To align sparse and noisy forward-looking 4D radar with dense and accurate omnidirectional 3D LiDAR data, we introduce a Cross-Modal Registration (CMR) network that jointly exploits Doppler-based motion priors and the physical laws governing LiDAR intensity and radar power density. Furthermore, we propose an adaptive fine-tuning strategy that incrementally updates the CMR network based on localization errors, enabling long-term adaptability to static environmental changes without ground-truth labels. We demonstrate that the proposed CMR network achieves state-of-the-art cross-modal registration performance on the open-access dataset. Then we validate LTR$^2$ across three robot platforms over a large-scale, long-term deployment (40+ km over 6 months), including challenging conditions such as nighttime smoke. Experimental results and ablation studies demonstrate centimeter-level accuracy and strong robustness against diverse environmental disturbances, significantly outperforming existing approaches.

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.

Referee Report

1 major / 2 minor

Summary. The manuscript proposes LTR², a cross-modal LiDAR-Teach-and-Radar-Repeat navigation system for long-term autonomy in dynamic and adverse environments. LiDAR is used in the teaching phase for precise mapping, while 4D radar handles the repeating phase for robustness. A Cross-Modal Registration (CMR) network aligns sparse noisy radar with dense LiDAR by exploiting Doppler-based motion priors and physical laws of LiDAR intensity and radar power density. An adaptive fine-tuning strategy incrementally updates the CMR network using localization errors without ground-truth labels. The system achieves SOTA cross-modal registration on an open dataset and demonstrates centimeter-level accuracy with strong robustness on a 40+ km, 6-month deployment across three platforms under conditions including smoke, outperforming baselines as supported by ablation studies.

Significance. If the results hold, the work provides a meaningful advance in robust teach-and-repeat navigation by enabling reliable cross-modal operation in varying and degenerate settings. Notable strengths include the ablation studies showing degradation when physical-law terms are removed and the fine-tuning loop reducing drift on extended traces, along with the scale of the real-world 6-month evaluation. These elements offer concrete evidence for practical long-term autonomy applications.

major comments (1)
  1. §5: The central performance claims of centimeter-level accuracy, SOTA registration, and outperformance on the 40 km deployment are load-bearing for the paper's contribution, yet the results lack reported error bars, standard deviations, or statistical significance tests for the localization and registration errors. This omission hinders evaluation of consistency and the reliability of gains over baselines.
minor comments (2)
  1. Abstract: The specific open-access dataset used to demonstrate SOTA registration performance should be named explicitly along with its citation to support reproducibility.
  2. §3–4: The description of the CMR network loss terms and the exact incremental update rule for fine-tuning could include additional equations or pseudocode for clarity on how Doppler priors and physical laws are implemented.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the positive evaluation and recommendation for minor revision. The feedback on statistical presentation is constructive, and we will strengthen the manuscript accordingly while preserving the core contributions.

read point-by-point responses

  1. Referee: §5: The central performance claims of centimeter-level accuracy, SOTA registration, and outperformance on the 40 km deployment are load-bearing for the paper's contribution, yet the results lack reported error bars, standard deviations, or statistical significance tests for the localization and registration errors. This omission hinders evaluation of consistency and the reliability of gains over baselines.

    Authors: We agree that explicit reporting of variability and statistical comparisons would improve the rigor of Section 5. In the revised manuscript we will add: (i) standard deviations alongside all mean errors for both the open cross-modal registration dataset and the 40+ km deployment results; (ii) error bars on all bar plots and trajectory-error figures; and (iii) paired statistical tests (Wilcoxon signed-rank or paired t-test, as appropriate to the data distribution) between LTR² and each baseline, with p-values reported. These additions will be computed from the multiple repeated traversals already present in our evaluation protocol. We believe the underlying performance gains remain robust, but the requested statistics will make that robustness transparent. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The derivation chain centers on the CMR network exploiting Doppler motion priors and physical intensity/power laws for cross-modal alignment, plus an error-driven incremental fine-tuning loop for long-term adaptation. These elements are defined and validated independently via architecture details, loss terms, ablations showing degradation when physical-law terms are removed, and quantitative results on open datasets plus 6-month field trials. No equations reduce reported accuracy or registration metrics to quantities defined by the same fitted parameters; the fine-tuning is driven by observed localization errors rather than the target performance metric itself. Any self-citations are non-load-bearing and do not substitute for the independent experimental evidence.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

The central performance claims rest on the effectiveness of the newly introduced CMR network and its adaptive update rule; these introduce learned parameters and a domain assumption about sensor physics alignment.

free parameters (1)
  • CMR network weights
    Neural network parameters are fitted during training and incremental fine-tuning on localization errors.
axioms (1)
  • domain assumption Physical laws relating LiDAR intensity to radar power density enable reliable cross-modal alignment when combined with Doppler motion priors.
    Invoked to justify the CMR network design in the abstract.
invented entities (1)
  • Cross-Modal Registration (CMR) network no independent evidence
    purpose: Aligns sparse noisy 4D radar data with dense 3D LiDAR maps using Doppler priors and sensor physics.
    Newly proposed component whose independent validation is limited to the paper's own experiments.

pith-pipeline@v0.9.0 · 5618 in / 1436 out tokens · 90427 ms · 2026-05-08T18:02:21.587289+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

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

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

63 extracted references · 63 canonical work pages

  1. [1]

    Visual Place Recognition: A Survey,

    S. Lowry, N. S ¨underhauf, P. Newman, J. J. Leonard, D. Cox, P. Corke, and M. J. Milford, “Visual Place Recognition: A Survey,”IEEE Trans- actions on Robotics, vol. 32, no. 1, pp. 1–19, 2015

    work page 2015

  2. [2]

    Experience-based Navigation for Long- Term Localisation,

    W. Churchill and P. Newman, “Experience-based Navigation for Long- Term Localisation,”The International Journal of Robotics Research, vol. 32, no. 14, pp. 1645–1661, 2013

    work page 2013

  3. [3]

    Special Issue on Long-Term Autonomy,

    T. Barfoot, J. Kelly, and G. Sibley, “Special Issue on Long-Term Autonomy,”The International Journal of Robotics Research, vol. 32, no. 14, pp. 1609–1610, 2013

    work page 2013

  4. [4]

    Radar teach and repeat: Architecture and initial field testing,

    X. Qiao, A. Krawciw, S. Lilge, and T. D. Barfoot, “Radar teach and repeat: Architecture and initial field testing,” in2025 IEEE International Conference on Robotics and Automation (ICRA), 2025, pp. 13 021– 13 027

    work page 2025

  5. [5]

    Teach-and-Repeat Path Following for an Autonomous Underwater Vehicle,

    P. King, A. Vardy, and A. L. Forrest, “Teach-and-Repeat Path Following for an Autonomous Underwater Vehicle,”Journal of Field Robotics, vol. 35, no. 5, pp. 748–763, 2018

    work page 2018

  6. [6]

    Kilometer-Scale Autonomous Navigation in Subarctic Forests: Challenges and Lessons Learned,

    D. Baril, S.-P. Desch ˆenes, O. Gamache, M. Vaidis, D. LaRocque, J. Laconte, V . Kubelka, P. Gigu`ere, and F. Pomerleau, “Kilometer-Scale Autonomous Navigation in Subarctic Forests: Challenges and Lessons Learned,”Field Robotics, vol. 2, pp. 1628–1660, 2021. [Online]. Available: https://api.semanticscholar.org/CorpusID:244714919

    work page 2021

  7. [7]

    Visual Teach and Repeat for Long-Range Rover Autonomy,

    P. Furgale and T. D. Barfoot, “Visual Teach and Repeat for Long-Range Rover Autonomy,”Journal of Field Robotics, vol. 27, no. 5, pp. 534– 560, 2010

    work page 2010

  8. [8]

    Cooperative Lunar Surface Exploration using Transfer Learning with Multi-Agent Visual Teach and Repeat,

    A. Akemoto and F. Zhu, “Cooperative Lunar Surface Exploration using Transfer Learning with Multi-Agent Visual Teach and Repeat,” in2023 IEEE Aerospace Conference. IEEE, 2023, pp. 1–9

    work page 2023

  9. [9]

    Lighting- invariant Visual Teach and Repeat Using Appearance-based LiDAR,

    C. McManus, P. Furgale, B. Stenning, and T. D. Barfoot, “Lighting- invariant Visual Teach and Repeat Using Appearance-based LiDAR,” Journal of Field Robotics, vol. 30, no. 2, pp. 254–287, 2013

    work page 2013

  10. [10]

    Lighting-invariant Adaptive Route Following Using Iterative Closest Point Matching,

    P. Kr ¨usi, B. B ¨ucheler, F. Pomerleau, U. Schwesinger, R. Siegwart, and P. Furgale, “Lighting-invariant Adaptive Route Following Using Iterative Closest Point Matching,”Journal of Field Robotics, vol. 32, no. 4, pp. 534–564, 2015

    work page 2015

  11. [11]

    Self-Supervised Robust Feature Matching Pipeline for Teach and Repeat Navigation,

    T. Rou ˇcek, A. S. Amjadi, Z. Rozsyp ´alek, G. Broughton, J. Blaha, K. Kusumam, and T. Krajn´ık, “Self-Supervised Robust Feature Matching Pipeline for Teach and Repeat Navigation,”Sensors, vol. 22, no. 8, p. 2836, 2022

    work page 2022

  12. [12]

    Robust and Long-term Monoc- ular Teach and Repeat Navigation using a Single-experience Map,

    L. Sun, M. Taher, C. Wild, C. Zhao, Y . Zhang, F. Majer, Z. Yan, T. Krajn´ık, T. Prescott, and T. Duckett, “Robust and Long-term Monoc- ular Teach and Repeat Navigation using a Single-experience Map,” in2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2021, pp. 2635–2642

    work page 2021

  13. [13]

    Are We Ready for Radar to Replace LiDAR in All-Weather Mapping and Localization?

    K. Burnett, Y . Wu, D. J. Yoon, A. P. Schoellig, and T. D. Barfoot, “Are We Ready for Radar to Replace LiDAR in All-Weather Mapping and Localization?”IEEE Robotics and Automation Letters, vol. 7, no. 4, pp. 10 328–10 335, 2022

    work page 2022

  14. [14]

    LiDAR-based Teach-and-Repeat of Mobile Robot Trajectories,

    C. Sprunk, G. D. Tipaldi, A. Cherubini, and W. Burgard, “LiDAR-based Teach-and-Repeat of Mobile Robot Trajectories,” in2013 IEEE/RSJ International Conference on Intelligent Robots and Systems. IEEE, 2013, pp. 3144–3149

    work page 2013

  15. [15]

    Cmrnext: Camera to lidar matching in the wild for localization and extrinsic calibration,

    D. Cattaneo and A. Valada, “Cmrnext: Camera to lidar matching in the wild for localization and extrinsic calibration,”IEEE Transactions on Robotics, 2025

    work page 2025

  16. [16]

    LiDAR-Level Localization with Radar? the CFEAR Approach to Accurate, Fast, and Robust Large-Scale Radar Odometry in Diverse Environments,

    D. Adolfsson, M. Magnusson, A. Alhashimi, A. J. Lilienthal, and H. Andreasson, “LiDAR-Level Localization with Radar? the CFEAR Approach to Accurate, Fast, and Robust Large-Scale Radar Odometry in Diverse Environments,”IEEE Transactions on Robotics, vol. 39, no. 2, pp. 1476–1495, 2022

    work page 2022

  17. [17]

    A New Wave in Robotics: Survey on Recent mmWave Radar Applications in Robotics,

    K. Harlow, H. Jang, T. D. Barfoot, A. Kim, and C. Heckman, “A New Wave in Robotics: Survey on Recent mmWave Radar Applications in Robotics,”IEEE Transactions on Robotics, vol. 40, pp. 4544–4560, 2024

    work page 2024

  18. [18]

    4DRT-SLAM: Robust SLAM in Smoke Environments using 4D Radar and Thermal Camera based on Dense Deep Learnt Features,

    J. Zhang, R. Xiao, H. Li, Y . Liu, X. Suo, C. Hong, Z. Lin, and D. Wang, “4DRT-SLAM: Robust SLAM in Smoke Environments using 4D Radar and Thermal Camera based on Dense Deep Learnt Features,” in2023 IEEE International Conference on Cybernetics and Intelligent Systems (CIS) and IEEE Conference on Robotics, Automation and Mechatronics (RAM). IEEE, 2023, pp. 19–24

    work page 2023

  19. [19]

    Adaptive Visual-Aided 4D Radar Odometry Through Transformer-Based Feature Fusion,

    Y . Zhang, R. Xiao, Z. Hong, L. Hu, and J. Liu, “Adaptive Visual-Aided 4D Radar Odometry Through Transformer-Based Feature Fusion,” in 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2024, pp. 12 529–12 535. 20 IEEE TRANSACTIONS ON ROBOTICS. PREPRINT VERSION. ACCEPTED MAY , 2026

    work page 2024

  20. [20]

    3d ego-Motion Estimation Using low-Cost mmWave Radars via Radar Velocity Factor for Pose- Graph SLAM,

    Y . S. Park, Y .-S. Shin, J. Kim, and A. Kim, “3d ego-Motion Estimation Using low-Cost mmWave Radars via Radar Velocity Factor for Pose- Graph SLAM,”IEEE Robotics and Automation Letters, vol. 6, no. 4, pp. 7691–7698, 2021

    work page 2021

  21. [21]

    Rad- gs: Radar-vision integration for 3d gaussian splatting slam in outdoor environments,

    R. Xiao, W. Liu, Y . Zhang, Y . Chen, J. Chen, Z. Wang, and L. Hu, “Rad- gs: Radar-vision integration for 3d gaussian splatting slam in outdoor environments,”IEEE Robotics and Automation Letters, vol. 10, no. 12, pp. 13 359–13 366, 2025

    work page 2025

  22. [22]

    Radar-on-LiDAR: Metric Radar Localization on Prior LiDAR Maps,

    H. Yin, Y . Wang, L. Tang, and R. Xiong, “Radar-on-LiDAR: Metric Radar Localization on Prior LiDAR Maps,” in2020 IEEE International Conference on Real-Time Computing and Robotics (RCAR). IEEE, 2020, pp. 1–7

    work page 2020

  23. [23]

    RaLL: End-to-End Radar Localization on LiDAR Map Using Differentiable Measurement Model,

    H. Yin, R. Chen, Y . Wang, and R. Xiong, “RaLL: End-to-End Radar Localization on LiDAR Map Using Differentiable Measurement Model,” IEEE Transactions on Intelligent Transportation Systems, vol. 23, no. 7, pp. 6737–6750, 2021

    work page 2021

  24. [24]

    Radar-to-LiDAR: Heteroge- neous Place Recognition via Joint Learning,

    H. Yin, X. Xu, Y . Wang, and R. Xiong, “Radar-to-LiDAR: Heteroge- neous Place Recognition via Joint Learning,”Frontiers in Robotics and AI, vol. 8, p. 661199, 2021

    work page 2021

  25. [25]

    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,” in2023 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2023, pp. 3976–3982

    work page 2023

  26. [26]

    FMCW Radar on LiDAR Map Localization in Structural Urban Environments,

    Y . Ma, H. Li, X. Zhao, Y . Gu, X. Lang, L. Li, and Y . Liu, “FMCW Radar on LiDAR Map Localization in Structural Urban Environments,” Journal of Field Robotics, vol. 41, no. 3, pp. 699–717, 2024

    work page 2024

  27. [27]

    Radar Localization and Mapping for Indoor Disaster Environments via Multi-modal Registration to Prior LiDAR Map,

    Y . S. Park, J. Kim, and A. Kim, “Radar Localization and Mapping for Indoor Disaster Environments via Multi-modal Registration to Prior LiDAR Map,” in2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2019, pp. 1307–1314

    work page 2019

  28. [28]

    RaLF: Flow-based Global and Metric Radar Localization in LiDAR Maps,

    A. Nayak, D. Cattaneo, and A. Valada, “RaLF: Flow-based Global and Metric Radar Localization in LiDAR Maps,” in2024 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2024, pp. 5097–5103

    work page 2024

  29. [29]

    Artificial Intelligence for Long-Term Robot Autonomy: A Survey,

    L. Kunze, N. Hawes, T. Duckett, M. Hanheide, and T. Krajn´ık, “Artificial Intelligence for Long-Term Robot Autonomy: A Survey,”IEEE Robotics and Automation Letters, vol. 3, no. 4, pp. 4023–4030, 2018

    work page 2018

  30. [30]

    Practice Makes Perfect? Managing and Leveraging Visual Experiences for Lifelong Navigation,

    W. Churchill and P. Newman, “Practice Makes Perfect? Managing and Leveraging Visual Experiences for Lifelong Navigation,” in2012 IEEE International Conference on Robotics and Automation. IEEE, 2012, pp. 4525–4532

    work page 2012

  31. [31]

    Navigation Without Localisation: Reliable Teach and Repeat Based on the Convergence Theorem,

    T. Krajn ´ık, F. Majer, L. Halodov ´a, and T. Vintr, “Navigation Without Localisation: Reliable Teach and Repeat Based on the Convergence Theorem,” in2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2018, pp. 1657–1664

    work page 2018

  32. [32]

    Self-Supervised Learning for Fusion of IR and RGB Images in Visual Teach and Repeat Navigation,

    X. Liu, Z. Rozsyp ´alek, and T. Krajn ´ık, “Self-Supervised Learning for Fusion of IR and RGB Images in Visual Teach and Repeat Navigation,” in2023 European Conference on Mobile Robots (ECMR). IEEE, 2023, pp. 1–7

    work page 2023

  33. [33]

    Predictive Data Acquisition for Lifelong Visual Teach, Repeat and Learn,

    T. Rou ˇcek, Z. Rozsyp´alek, J. Blaha, J. Ulrich, and T. Krajn´ık, “Predictive Data Acquisition for Lifelong Visual Teach, Repeat and Learn,”IEEE Robotics and Automation Letters, vol. 9, no. 11, pp. 10 042–10 049, 2024

    work page 2024

  34. [34]

    LiDAR Scan Registration Robust to Extreme Motions,

    S.-P. Desch ˆenes, D. Baril, V . Kubelka, P. Gigu `ere, and F. Pomerleau, “LiDAR Scan Registration Robust to Extreme Motions,” in2021 18th Conference on Robots and Vision (CRV). IEEE, 2021, pp. 17–24

    work page 2021

  35. [35]

    Precise Ego-Motion Estimation with Millimeter-Wave Radar under Diverse and Challenging Conditions,

    S. H. Cen and P. Newman, “Precise Ego-Motion Estimation with Millimeter-Wave Radar under Diverse and Challenging Conditions,” in2018 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2018, pp. 6045–6052

    work page 2018

  36. [36]

    Are doppler velocity measurements useful for spinning radar odometry?

    D. Lisus, K. Burnett, D. J. Yoon, R. Poulton, J. Marshall, and T. D. Barfoot, “Are doppler velocity measurements useful for spinning radar odometry?”IEEE Robotics and Automation Letters, vol. 10, no. 1, pp. 224–231, 2025

    work page 2025

  37. [37]

    1 Year, 1000 Km: The Oxford RobotCar Dataset,

    W. Maddern, G. Pascoe, C. Linegar, and P. Newman, “1 Year, 1000 Km: The Oxford RobotCar Dataset,”The International Journal of Robotics Research, vol. 36, no. 1, pp. 3–15, 2017

    work page 2017

  38. [38]

    Vision-based Place Recognition: How Low Can You Go?

    M. Milford, “Vision-based Place Recognition: How Low Can You Go?” The International Journal of Robotics Research, vol. 32, no. 7, pp. 766– 789, 2013

    work page 2013

  39. [39]

    Long-term navigation for autonomous robots based on spatio-temporal map prediction,

    Y . Wang, Y . Fan, J. Wang, and W. Chen, “Long-term navigation for autonomous robots based on spatio-temporal map prediction,”Robotics and Autonomous Systems, vol. 174, p. 104724, 2024

    work page 2024

  40. [40]

    Frequency map enhancement for long-term mobile robot autonomy in changing environments,

    T. Krajn ´ık, J. P. Fentanes, J. M. Santos, and T. Duckett, “Frequency map enhancement for long-term mobile robot autonomy in changing environments,”IEEE Transactions on Robotics, vol. 33, no. 4, pp. 964– 977, 2017

    work page 2017

  41. [41]

    Superpixel-based appearance change prediction for long-term navigation across seasons,

    P. Neubert, N. S ¨underhauf, and P. Protzel, “Superpixel-based appearance change prediction for long-term navigation across seasons,”Robotics and Autonomous Systems, vol. 69, pp. 1–13, 2015

    work page 2015

  42. [42]

    Ephemerality meets LiDAR- based Lifelong Mapping,

    H. Gil, D. Lee, G. Kim, and A. Kim, “Ephemerality meets LiDAR- based Lifelong Mapping,” inProceedings of the IEEE International Conference on Robotics and Automation (ICRA), Atlanta, GA, USA, May 2025

    work page 2025

  43. [43]

    Lifelong 3D Mapping Framework for Hand-held & Robot-mounted LiDAR Mapping Systems,

    L. Yang, S. M. Prakhya, S. Zhu, and Z. Liu, “Lifelong 3D Mapping Framework for Hand-held & Robot-mounted LiDAR Mapping Systems,” IEEE Robotics and Automation Letters, vol. 9, no. 11, pp. 9446–9453, 2024

    work page 2024

  44. [44]

    No More Potentially Dynamic Objects: Static Point Cloud Map Generation based on 3D Object Detection and Ground Projection,

    S. Woo, D. Jung, and S.-W. Kim, “No More Potentially Dynamic Objects: Static Point Cloud Map Generation based on 3D Object Detection and Ground Projection,”arXiv Preprint ArXiv:2407.01073, 2024

    work page arXiv 2024

  45. [45]

    Predictive and Adaptive Maps for Long-Term Visual Navigation in Changing Environments,

    L. Halodov ´a, E. Dvoˇrr´akov´a, F. Majer, T. Vintr, O. M. Mozos, F. Dayoub, and T. Krajn ´ık, “Predictive and Adaptive Maps for Long-Term Visual Navigation in Changing Environments,” in2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2019, pp. 7033– 7039

    work page 2019

  46. [46]

    Continual slam: Beyond lifelong simultaneous localization and mapping through con- tinual learning,

    N. V ¨odisch, D. Cattaneo, W. Burgard, and A. Valada, “Continual slam: Beyond lifelong simultaneous localization and mapping through con- tinual learning,” inThe International Symposium of Robotics Research. Springer, 2022, pp. 19–35

    work page 2022

  47. [47]

    VidLoc: A Deep Spatio-Temporal Model for 6-DoF Video-Clip Relocalization,

    R. Clark, S. Wang, A. Markham, N. Trigoni, and H. Wen, “VidLoc: A Deep Spatio-Temporal Model for 6-DoF Video-Clip Relocalization,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017, pp. 6856–6864

    work page 2017

  48. [48]

    Patch- NetVLAD: Multi-Scale Fusion of Locally-Global Descriptors for Place Recognition,

    S. Hausler, S. Garg, M. Xu, M. Milford, and T. Fischer, “Patch- NetVLAD: Multi-Scale Fusion of Locally-Global Descriptors for Place Recognition,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 14 141–14 152

    work page 2021

  49. [49]

    VLocNet++: Deep Multi- task Learning for Semantic Visual Localization and Odometry,

    N. Radwan, A. Valada, and W. Burgard, “VLocNet++: Deep Multi- task Learning for Semantic Visual Localization and Odometry,”IEEE Robotics and Automation Letters, vol. 3, no. 4, pp. 4407–4414, 2018

    work page 2018

  50. [50]

    Local maps are all you need: A review of topometric teach and repeat navigation,

    A. Krawciw and T. D. Barfoot, “Local maps are all you need: A review of topometric teach and repeat navigation,”Annual Review of Control, Robotics, and Autonomous Systems, vol. 9, 2025

    work page 2025

  51. [51]

    LIO- SAM: Tightly-coupled LiDAR Inertial Odometry via Smoothing and Mapping,

    T. Shan, B. Englot, D. Meyers, W. Wang, C. Ratti, and D. Rus, “LIO- SAM: Tightly-coupled LiDAR Inertial Odometry via Smoothing and Mapping,” in2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2020, pp. 5135–5142

    work page 2020

  52. [52]

    T. D. Barfoot,State Estimation for Robotics. Cambridge University Press, 2024

    work page 2024

  53. [53]

    An EKF Based Approach to Radar Inertial Odometry,

    C. Doer and G. F. Trommer, “An EKF Based Approach to Radar Inertial Odometry,” in2020 IEEE International Conference on Multisensor Fusion and Integration for Intelligent Systems (MFI). IEEE, 2020, pp. 152–159

    work page 2020

  54. [54]

    R. C. Coulteret al.,Implementation of the Pure Pursuit Path Tracking Algorithm. Carnegie Mellon University, The Robotics Institute, 1992

    work page 1992

  55. [55]

    M. I. Skolnik,Radar Handbook. New York: McGraw-Hill, 1970

    work page 1970

  56. [56]

    Fujii and T

    T. Fujii and T. Fukuchi, Eds.,Laser Remote Sensing. Boca Raton: CRC Press, 2005

    work page 2005

  57. [57]

    Swin Transformer: Hierarchical Vision Transformer using Shifted Win- dows,

    Z. Liu, Y . Lin, Y . Cao, H. Hu, Y . Wei, Z. Zhang, S. Lin, and B. Guo, “Swin Transformer: Hierarchical Vision Transformer using Shifted Win- dows,” in2021 IEEE/CVF International Conference on Computer Vision (ICCV), 2021, pp. 10 012–10 022

    work page 2021

  58. [58]

    RegFormer: An Efficient Projection-Aware Transformer Network for Large-Scale Point Cloud Registration,

    J. Liu, G. Wang, Z. Liu, C. Jiang, M. Pollefeys, and H. Wang, “RegFormer: An Efficient Projection-Aware Transformer Network for Large-Scale Point Cloud Registration,” in2023 IEEE/CVF International Conference on Computer Vision (ICCV), 2023, pp. 8417–8426

    work page 2023

  59. [59]

    PointPWC-Net: Cost V olume on Point Clouds for (Self-) Supervised Scene Flow Estimation,

    W. Wu, Z. Y . Wang, Z. Li, W. Liu, and L. Fuxin, “PointPWC-Net: Cost V olume on Point Clouds for (Self-) Supervised Scene Flow Estimation,” inComputer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part V 16. Springer, 2020, pp. 88–107

    work page 2020

  60. [60]

    Multi-class Road User Detection with 3+1D Radar in the View-of- Delft Dataset,

    A. Palffy, E. Pool, S. Baratam, J. F. P. Kooij, and D. M. Gavrila, “Multi-class Road User Detection with 3+1D Radar in the View-of- Delft Dataset,”IEEE Robotics and Automation Letters, vol. 7, no. 2, pp. 4961–4968, 2022

    work page 2022

  61. [61]

    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,” in2023 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2023, pp. 8333–8340

    work page 2023

  62. [62]

    Hidden Gems: 4D Radar Scene Flow Learning Using Cross-Modal Supervision,

    F. Ding, A. Palffy, D. M. Gavrila, and C. X. Lu, “Hidden Gems: 4D Radar Scene Flow Learning Using Cross-Modal Supervision,” in XIAOet al.: LIDAR TEACH, RADAR REPEAT: ROBUST CROSS-MODAL NA VIGATION IN DEGENERATE AND V ARYING ENVIRONMENTS 21 Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2023, pp. 9340–9349

    work page 2023

  63. [63]

    HRegNet: A Hierarchical Network for Efficient and Accurate Outdoor LiDAR Point Cloud Registration,

    F. Lu, G. Chen, Y . Liu, L. Zhang, S. Qu, S. Liu, R. Gu, and C. Jiang, “HRegNet: A Hierarchical Network for Efficient and Accurate Outdoor LiDAR Point Cloud Registration,”IEEE Transactions on Pattern Anal- ysis and Machine Intelligence, vol. 45, no. 10, pp. 11 884–11 897, 2023

    work page 2023