Efficient Active Training for Deep LiDAR Odometry

Beibei Zhou; Hui Kong; Jianhui Guo; Zhenbo Song; Zhiyuan Zhang

REVIEW 5 major objections 5 minor 35 references

Efficient Active Training for Deep LiDAR Odometry

T0 review · 5 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a two-stage active selection strategy—initial diversity-focused selection plus iterative hard-sample discovery—lets a deep LiDAR odometry model match or beat full-dataset accuracy using only 52% of training…

desk verdict Parity claim at 52% is shaky because the full-data baseline looks undertrained; the two-stage active selection is novel but reproducibility is missing. read the letter →

arxiv 2509.03211 v1 pith:7Q6WTU5M submitted 2025-09-03 cs.RO cs.CV

Beibei Zhou , Zhiyuan Zhang , Zhenbo Song , Jianhui Guo , Hui Kong This is my paper · ORCID

classification cs.ROcs.CV

keywords LiDARodometryactivelearningtrainingsetselectionsequencediversityadverseweathersnowunsupervised

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

Deep LiDAR odometry models estimate vehicle motion from point clouds, and making them work in rain, snow, and dense traffic normally requires large, multi-condition training sets. This paper proposes ActiveLO-training, a two-stage procedure that decides which sequences deserve training time instead of using everything: it scores clear-weather sequences by how much the vehicle's trajectory varies, then iteratively adds the snowy sequences the current model finds hardest, as measured by reconstruction error and prediction inconsistency. The central experimental claim is that a model trained on about 52% of the available sequences matches or slightly beats the same model trained on all of them, with average translational drift of 1.36% versus 1.70% and rotation drift of 0.86°/100m versus 1.00°/100m across the Ford, CADC, and WADS test sets. If this holds, odometry models for new environments can be trained at roughly half the data and compute cost, and rare adverse-weather cases can be deliberately added instead of being left to chance.

What carries the argument

The carrying mechanism is a pair of selection scores. The first is trajectory diversity: per sequence the model computes node turn-angle standard deviation $\sigma_\theta$, edge-length standard deviation $\sigma_l$, edge-speed standard deviation $\sigma_v$, and outlier proportion, then forms $F_{\mathrm{Var}}(s)=\lambda_1\sigma_\theta+\lambda_2\sigma_l+\lambda_3\sigma_v$ and an importance term $F_{\mathrm{Impor}}(s)$ that rewards total turning and trajectory length. A linear program maximizes $F_{\mathrm{Var}}+F_{\mathrm{Impor}}$ subject to sampling at least one sequence from each bin of outlier proportion and average speed. The second score drives incremental selection: for each remaining sequence, the current model computes a point-to-plane scene reconstruction loss $F_{\mathrm{Recon}}$ plus a prediction inconsistency loss $F_{\mathrm{Incon}}$ obtained by perturbing the target frame with Gaussian rotations and translations and measuring the variance of the recovered pose; the top $h$ sequences by $F_{\mathrm{Recon}}+F_{\mathrm{Incon}}$ enter the training pool each round. These two scores convert the raw point-cloud pool into the 52%-sized training set that the experiments compare against full-data training.

What would settle it

Run the same HPPLO-Net training protocol at 52% with three ways of choosing sequences: ActiveLO as described, random subsets averaged over several seeds, and a generic diversity baseline such as farthest-point sampling in feature space. If either baseline matches or beats the reported 1.36% average translation error, the ITSS/AIS scores are not carrying the result. A more direct check is to correlate each sequence's ITSS + AIS score with the reduction in test error obtained by adding that sequence; a near-zero correlation would falsify the selection mechanism.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that dataset selection, not network architecture, is the lever for making deep LiDAR odometry robust to diverse weather. ActiveLO-training splits selection into two stages. Initial Training Set Selection (ITSS) converts each trajectory into nodes and edges, computes turn-angle, edge-length, speed, and outlier-proportion statistics, and solves a small linear program whose objective combines trajectory variability and trajectory importance to pick a compact initial set from clear-weather KITTI sequences. Active Incremental Selection (AIS) then iterates: train the current model, run it on the remaining pool, score each sequence by scene reconstruction loss (point-to-plane alignment error) plus prediction inconsistency loss (variance of poses recovered under small random pose augmentations), and add the highest-scoring sequences. In the main tables, the 52%-volume ActiveLO training set yields average translation error 1.36% and rotation error 0.86°/100m, versus 1.70% and 1.00°/100m for the full 100% set, and it beats random selection at every percentage bucket. The authors conclude that a small, deliberately chosen subset can carry the same odometry knowledge as the full pool.

Load-bearing premise

The entire active-selection argument rests on the assumption that a sequence's trajectory statistics—how much the vehicle turns, changes speed, and produces outliers—measure what makes it valuable for odometry training; if those statistics do not track actual odometry difficulty, ITSS is just a mildly informed heuristic.

Editorial extensions

If this is right

A 52%-volume ActiveLO subset produces lower average drift than the full dataset (1.36% vs 1.70% translation error), so the selected set is not merely equal but a better use of the data.
Even at 9% of the data, ActiveLO's ITSS initial set cuts average translation error from 7.74% for random selection to 3.37%, showing a small deliberately diverse starting set already yields a usable base model before snow data is added.
The SRL and PIL ablation (Table VI) shows each loss alone is weaker than the pair, because SRL tends to flag heavy-snow/dynamic-object scenes while PIL flags sparse scenes; keeping both is an explicit design decision supported by the experiments.
Because both stages use trajectory statistics and self-supervised losses rather than labels, the pipeline can rank unlabeled LiDAR logs, which is the form in which odometry training data usually accumulates.

Reading between the lines

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

An untested stress test: replace ITSS's handcrafted features with a generic diversity baseline such as farthest-point sampling over trajectory feature vectors; if that baseline matches the 52% result, the specific weighting in Eqs. (8)-(9) is not the active ingredient.
The reported experiments use one unsupervised backbone (HPPLO-Net); whether the selection logic transfers to supervised LiDAR odometry networks or to other sensors is left open by the paper.
A deployment-oriented comparison would track wall-clock time, including the per-iteration inference over the remaining pool; sequence-level counts in Table VII describe training load but not the full time budget.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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

Punchline: the headline result, matching full-data training with 52% of sequence volume, is not yet credible because the 100% baseline is under-specified and likely undertrained. That is the load-bearing issue. What is actually new: the two-stage selection pipeline (ITSS plus AIS) for deep LiDAR odometry. The specific combination is, as far as I know, novel. The ablations show both stages contribute, and the experiments cover KITTI, CADC, WADS, and Ford. Credit where due: the incremental selection via scene-reconstruction and prediction-inconsistency losses is a sensible active-learning approach, and the improvement over random selection is consistent across Tables I-IV. The use of HPPLO-Net as the backbone is fine, and self-citing that backbone is not itself a flaw. Soft spots: (1) The full baseline is described only as 50 epochs, with no learning curve, early stopping, hyperparameter search, or repeated trials. ActiveLO at 23%, 38%, and 52% all beat that baseline by large margins, which strongly suggests the 100% model is not converged; parity might vanish against a properly trained reference. (2) There are no error bars anywhere. The paper admits Random sometimes wins on individual sequences, so the headline 1.36-vs-1.70 gap is not known to be statistically significant. (3) Reproducibility is poor: no code or data, and the paper omits the node-selection rule, the number of linear-programming intervals, all weights (lambda1 through lambda5, alpha, beta, epsilon), and the augmentation variances. (4) Table VII's efficiency arithmetic is internally inconsistent: Eq. (19) with iter+1 gives 1230, not 1000, and Eq. (20) with iter=7 gives 364, not 336. The text seems to use iter=7 for training but iter=6 for inference. (5) The 9% random baseline protocol is ambiguous: does it use the same initial six sequences or a fully random draw? None of this kills the paper. The AIS stage selects 30 of the 36 sequences, so even if ITSS's hand-crafted features were arbitrary, the central parity claim could still hold. The core idea is coherent, the empirical trend is promising, and the missing details are fixable. Recommendation: send this to a serious peer review, but the authors must add a properly converged full-data baseline, report error bars, and release code or at least full hyperparameters. I would not cite it in its current form, but I would read the revised version.

Referee Report

5 major / 5 minor

Summary. The paper proposes ActiveLO-training, a two-stage active selection strategy for building a training set for a deep LiDAR odometry network (HPPLO-Net). The first stage, ITSS, represents each training trajectory as nodes and edges, computes hand-crafted features such as turning-angle variability, edge-length/speed variability, and outlier proportion, and solves a linear program to pick an initial diverse subset (6 of 69 sequences). The second stage, AIS, iteratively adds the 5 sequences with the highest scene-reconstruction loss and prediction-inconsistency loss, computed with the current model, until reaching 52% of the full training pool. Experiments on Ford, CADC, and WADS test sequences compare ActiveLO with random selection and classical LiDAR odometry baselines, reporting that ActiveLO at 52% achieves average translational drift 1.36% versus 1.70% for the full dataset, and consistently beats random selection. The authors claim this demonstrates training efficiency: full-dataset performance with only 52% of the sequence volume.

Significance. If the empirical claims hold, the paper addresses a relevant practical problem: reducing the training cost of deep LiDAR odometry while maintaining or improving generalization across weather conditions. The evaluation is broader than many active-learning papers for odometry, using held-out test sequences from three datasets and comparing against several classical baselines. The paper also provides ablations for both selection components and an explicit (though currently flawed) efficiency model. The central claim is falsifiable and the experimental protocol is, in principle, reproducible. However, the current evidence is not yet convincing because the full-data baseline may not be converged, the random-selection protocol is ambiguous, and no error bars or repeated trials are reported; these issues directly affect the headline 52% parity claim.

major comments (5)

[§IV-G, Eqs. (18)-(20), Table VII] The efficiency arithmetic is internally inconsistent. With Num(Sinit)=6, h=5, and iter=7, Eq. (19) evaluates to 6×15 + Σ_{itr=1}^{8} (6+5×itr)×5 = 1230, not 1000 as reported. Eq. (20) evaluates to (69−6) + Σ_{itr=1}^{7} (69−6−5×itr) = 63 + 301 = 364, not 336; the value 336 corresponds to using iter=6 in the sum, contradicting the stated iter=7. The total 1336 in Table VII is therefore also unexplained. Since training efficiency is one of the two claimed contributions, this arithmetic must be corrected and the definition of each term clarified.
[§IV-B, Tables I-IV] The random-selection protocol is ambiguous. The text says that all percentages above 9% 'incorporate the same initial set alongside incremental training data,' but the 9% rows in Tables I-IV report separate ActiveLO and Random results, implying different initial 6-sequence sets. If the 23%-52% ActiveLO results use the ITSS-selected initial set while random results use a different random initial set, the comparison conflates the quality of the initial selection with the quality of the incremental AIS selection. The authors should specify exactly which initial set is used for each row and, ideally, run the incremental process from the same initial set for both strategies to isolate the contribution of AIS.
[§IV-B, Table I, §IV-G, Eq. (18)] The full-dataset 100% baseline is described only as '50 epochs' with no convergence criterion, learning curve, or repeated trials. Table I shows that ActiveLO at 23%, 38%, and 52% all outperform the 100% baseline on average trel (1.68, 1.65, and 1.36 versus 1.70), which is more naturally read as evidence that the 100% model is undertrained or poorly regularized than as an indication that 52% of the data is sufficient. The central claim in the Abstract, that ActiveLO 'matches' full-dataset performance, is therefore not established. The authors should report the training curve of the full model, a convergence rule, and ideally multiple seeds, and then revise the claim accordingly.
[§IV-B, Tables I-IV, §IV-E] No error bars or statistical tests are provided for any of the main comparisons. The paper itself acknowledges that the Random strategy occasionally beats ActiveLO in about 23% of per-sequence results, and Table III shows several sequences (e.g., 54 and 56 at 52%) where random selection has lower trel than ActiveLO. Without repeated runs or a paired test over test sequences, the reported average gaps, such as the 1.36-vs-1.70 headline difference, cannot be distinguished from noise. The authors should add standard deviations or confidence intervals over at least several random-selection seeds and, if feasible, over model-training seeds.
[§III-A.2, Eqs. (8)-(9), Table V] The ITSS objective depends on hand-crafted features with weighting factors λ1 through λ5, but neither the chosen values nor any sensitivity analysis is reported. The only validation of ITSS is the four-sequence ablation in Table V, which compares three fixed random sets against one ITSS set. Given that the 9% ActiveLO row in Table I shows a very large improvement over 9% Random (trel 3.37 versus 7.74), the initial-selection component is load-bearing for the overall framework, and the claim that ITSS captures 'motion diversity' needs stronger support: report the λ values, test nearby values, compare against a standard diversity-based selection baseline, or show that the main result persists with a random initial set when AIS is used.

minor comments (5)

[Algorithm 1] Steps 3-4 contain typographical errors: 'Nodes and edges enk−1nk used to partition the trajectory nk' should refer to the sequence s, and Eq. references such as '(Eq.(1)˜(2)' are missing closing parentheses.
[§III-A.3, Eq. (10)] The objective function sums over 's=0' to '10', but the number of candidate sequences in the pool is not 10; the summation range should be over all sequences in Sge. Please clarify.
[Table VII] The column labeled 'epochs' actually reports total sequence-epoch iterations, not a number of epochs; the label is misleading and should be changed.
[§IV-B] The sentence 'The replicability of our results across multiple experiments further validates the reliability of ActiveLO-training' is not supported by any experimental detail in the paper; either describe the multiple experiments or remove the sentence.
[Tables II-IV] The distinction between 'Full A-LOAM' and 'A-LOAM' should be explained in the text; currently the reader must infer that 'Full A-LOAM' uses a different training set or preprocessing, and the tables do not state what 'Full' refers to.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the selection strategy is heuristic/empirical and the headline parity result is measured on held-out sequences, not derived from the selection criterion.

full rationale

The paper's central claim is that ActiveLO-training matches full-dataset LiDAR odometry performance using 52% of training sequences. This is an empirical result: Table I reports held-out test error on Ford, CADC, and WADS sequences, compared against a 100%-trained baseline and random-selection controls. The ITSS component computes trajectory variability and importance from dataset poses and selected initial sequences via a linear program; these hand-designed scores are selection heuristics, not fitted predictors of held-out performance. The AIS component selects sequences by the current model's scene-reconstruction and prediction-inconsistency losses, which is standard active learning and not a logical reduction of the final generalization claim. The only self-citation is HPPLO-Net [32] used as the backbone network; it is an independently published unsupervised odometry method and is not invoked as the justification for the selection criteria or the parity result. Issues such as the under-specified 50-epoch full baseline, absence of error bars, and the paper's own admission that Random occasionally wins in about 23% of per-sequence comparisons are reproducibility and significance concerns, not circularity. No equation in the paper is equivalent by construction to the claimed prediction, and no load-bearing argument reduces to a self-citation. Therefore the paper receives a circularity score of 0.

Assumptions & free parameters 7 free parameters · 6 assumptions · 0 invented entities

The framework rests on hand-picked features and weights (λ, ε, α, β), active-learning assumptions about which losses identify valuable samples, and a narrow baseline choice (random only). The values of all weighting hyperparameters are unreported, so the contribution of the selection mechanism cannot be separated from parameter tuning.

free parameters (7)

λ1, λ2, λ3 = not reported
Weights in trajectory variability FVar(s), Eq. (8). No values or sensitivity analysis provided; the selection outcome depends on them.
λ4, λ5 = not reported
Weights in trajectory importance FImpor(s), Eq. (9). No values or sensitivity analysis provided.
ε (outlier distance threshold) = not reported
Threshold in Eq. (6) defining outlier proportion used in sequence features.
α, β (overall loss weights) = not reported
Weights in Eq. (17) combining scene-reconstruction and prediction-inconsistency losses for AIS selection.
αx..αψ (augmentation variances) = not reported, stated in [0,1]
Variance scalars in Eq. (15) for sampling augmented poses; no values disclosed.
u, h, iter (selection budget) = 6, 5, 6 or 7 (inconsistent)
Initial set size, sequences added per iteration, and iteration count. Table VII reports training overhead consistent with iter=7 and inference overhead consistent with iter=6, while 52% of 69 implies iter=6.
Number of linear-programming intervals = not reported
The sequence pool is divided into intervals over outlier proportion and mean speed (Section III.A.3); the interval count and boundaries are unspecified.

assumptions (6)

standard math Linear programming (Eq. 10), nearest-neighbor outlier detection (Eq. 6), and point-to-plane loss (Eq. 11) are treated as standard and correct.
Used throughout Section III as background mathematical tools.
domain assumption Hand-crafted trajectory features (node angle std, edge length std, edge speed std, outlier proportion) weighted by λ1..λ5 capture the diversity that improves odometry generalization.
Invoked in Section III.A.2 to define ITSS selection; validated only by a 4-sequence ablation (Table V) with no sensitivity analysis or feature ablation.
domain assumption Sequences with high scene-reconstruction loss and prediction-inconsistency loss under the current model are the most valuable additions for improving generalization.
Invoked in Section III.B; standard active-learning premise, but no comparison against other acquisition functions.
ad hoc to paper The six degrees of freedom of the relative pose are independent for the purpose of sampling augmentation transformations.
Stated in Section III.B.2, Eq. (15); used to generate augmented samples for the inconsistency loss.
domain assumption Driving conditions (speed and direction) are consistent within each edge, so edge length and speed are meaningful motion features.
Stated in Section III.A.1.c; required for Eq. (3).
domain assumption The LiDAR frame rate r is known and constant across all datasets used.
Used in Eq. (3) to compute edge speeds; not verified for KITTI, CADC, or WADS.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Efficient Active Training for Deep LiDAR Odometry." pith.science (2026). https://pith.science/paper/7Q6WTU5M

@misc{pith2026250903211,
  author       = {Pith},
  title        = {Pith review of: Efficient Active Training for Deep LiDAR Odometry},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7Q6WTU5M}},
  note         = {Machine review of arXiv:2509.03211}
}

read the original abstract

Robust and efficient deep LiDAR odometry models are crucial for accurate localization and 3D reconstruction, but typically require extensive and diverse training data to adapt to diverse environments, leading to inefficiencies. To tackle this, we introduce an active training framework designed to selectively extract training data from diverse environments, thereby reducing the training load and enhancing model generalization. Our framework is based on two key strategies: Initial Training Set Selection (ITSS) and Active Incremental Selection (AIS). ITSS begins by breaking down motion sequences from general weather into nodes and edges for detailed trajectory analysis, prioritizing diverse sequences to form a rich initial training dataset for training the base model. For complex sequences that are difficult to analyze, especially under challenging snowy weather conditions, AIS uses scene reconstruction and prediction inconsistency to iteratively select training samples, refining the model to handle a wide range of real-world scenarios. Experiments across datasets and weather conditions validate our approach's effectiveness. Notably, our method matches the performance of full-dataset training with just 52\% of the sequence volume, demonstrating the training efficiency and robustness of our active training paradigm. By optimizing the training process, our approach sets the stage for more agile and reliable LiDAR odometry systems, capable of navigating diverse environmental conditions with greater precision.

Figures

Figures reproduced from arXiv: 2509.03211 by the authors.

**Figure 1.** The overview of ActiveLO-training process. We analyze the sequence diversity of general samples with complex motion states and [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗

**Figure 2.** Labeling results for nodes and edges of sequences 07 and 09. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** The details of AIS module: Generation Process of Augmented [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

Figures from the paper (5 more)

**Figure 4.** Figure 4: Overall Distribution of Translation and Rotation Errors for All [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]

**Figure 5.** Figure 5: The trajectory visualization on dynamic sequences. The results [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

**Figure 6.** Figure 6: The drift distribution of snowy sequences. [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

**Figure 7.** Figure 7: The trajectory visualization on snowy sequences. [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]

**Figure 8.** Figure 8: Scene visualization. (a)(b): Scenes extreme open with bliz [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

35 extracted references · 29 canonical work pages

[1]

Gaussian splatting slam,

H. Matsuki, R. Murai, P. H. Kelly, and A. J. Davison, “Gaussian splatting slam,” in IEEE Conf. Comput. Vis. Pattern Recog. , pp. 18 039–18 048

work page
[2]

Fast-lidar-slam: A robust and real-time factor graph for urban scenarios with unstable gps signals,

Z. Rui, W. Xu, and Z. Feng, “Fast-lidar-slam: A robust and real-time factor graph for urban scenarios with unstable gps signals,” IEEE Trans. Intell. Transport. Syst. , pp. 20 043–20 058, 2024

work page 2024
[3]

Dynamic semantic slam based on panoramic camera and lidar fusion for autonomous driving,

F. Li, C. Fu, J. Wang, and D. Sun, “Dynamic semantic slam based on panoramic camera and lidar fusion for autonomous driving,” IEEE Trans. Intell. Transport. Syst. , pp. 1–14, 2025

work page 2025
[4]

Perception and sensing for autonomous vehicles under adverse weather conditions: A survey,

Y . Zhang, A. Carballo, H. Yang, and K. Takeda, “Perception and sensing for autonomous vehicles under adverse weather conditions: A survey,” ISPRS J. Pho. Remote Sen. , pp. 146–177, 2023

work page 2023
[5]

Libre: The multiple 3d lidar dataset,

A. Carballo, J. Lambert, A. Monrroy, D. Wong, P. Narksri, Y . Kitsukawa, E. Takeuchi, S. Kato, and K. Takeda, “Libre: The multiple 3d lidar dataset,” in IEEE Intelligent Veh. Symp. IEEE, 2020, pp. 1094–1101

work page 2020
[6]

The perception system of intelligent ground vehicles in all weather conditions: A systematic literature review,

A. S. Mohammed, A. Amamou, F. K. Ayevide, S. Kelouwani, K. Ag- bossou, and N. Zioui, “The perception system of intelligent ground vehicles in all weather conditions: A systematic literature review,” Sensors, 2020

work page 2020
[7]

Robust object detection in challenging weather conditions,

H. Gupta, O. Kotlyar, H. Andreasson, and A. J. Lilienthal, “Robust object detection in challenging weather conditions,” in Winter Conf. Appl. Comput. Vis. , 2024, pp. 7523–7532

work page 2024
[8]

Multi-agent trajectory prediction by combining egocentric and allocentric views,

X. Jia, L. Sun, H. Zhao, M. Tomizuka, and W. Zhan, “Multi-agent trajectory prediction by combining egocentric and allocentric views,” in Conference on Robot Learning . PMLR, 2022, pp. 1434–1443

work page 2022

Show all 35 references

[9]

Hdgt: Heterogeneous driving graph transformer for multi-agent trajectory prediction via scene encoding,

X. Jia, P. Wu, L. Chen, Y . Liu, H. Li, and J. Yan, “Hdgt: Heterogeneous driving graph transformer for multi-agent trajectory prediction via scene encoding,” IEEE Trans. Pattern Anal. Mach. Intell. , 2023

2023
[10]

Motion transformer with global intention localization and local movement refinement,

S. Shi, L. Jiang, D. Dai, and B. Schiele, “Motion transformer with global intention localization and local movement refinement,” Adv. Neural Inform. Process. Syst. , pp. 6531–6543, 2022

2022
[11]

Pointpainting: Sequential fusion for 3d object detection,

S. V ora, A. H. Lang, B. Helou, and O. Beijbom, “Pointpainting: Sequential fusion for 3d object detection,” in IEEE Conf. Comput. Vis. Pattern Recog., 2020, pp. 4604–4612

2020
[12]

Activead: Planning-oriented active learning for end-to-end autonomous driving,

H. Lu, X. Jia, Y . Xie, W. Liao, X. Yang, and J. Yan, “Activead: Planning-oriented active learning for end-to-end autonomous driving,” arXiv preprint arXiv:2403.02877 , 2024

2024 arXiv
[13]

Deep learning- based robust positioning for all-weather autonomous driving,

Y . Almalioglu, M. Turan, N. Trigoni, and A. Markham, “Deep learning- based robust positioning for all-weather autonomous driving,” Nature Machine Intelligence, pp. 749–760, 2022

2022
[14]

Cnn for imu assisted odometry estimation using velodyne lidar,

M. Velas, M. Spanel, M. Hradis, and A. Herout, “Cnn for imu assisted odometry estimation using velodyne lidar,” in IEEE Int. Conf. Autonom. Rob. Syst. Compet. , 2018, pp. 71–77

2018
[15]

Deeppco: End-to-end point cloud odom- etry through deep parallel neural network,

W. Wang, M. R. U. Saputra, P. Zhao, P. Gusmao, B. Yang, C. Chen, A. Markham, and N. Trigoni, “Deeppco: End-to-end point cloud odom- etry through deep parallel neural network,” in IEEE/RSJ International Conference on Intelligent Robots and Systems , 2019, pp. 3248–3254

2019
[16]

Dmlo: Deep matching lidar odometry,

Z. Li and N. Wang, “Dmlo: Deep matching lidar odometry,” in IEEE/RSJ International Conference on Intelligent Robots and Systems , 2020, pp. 6010–6017

2020
[17]

Lodonet: A deep neural network with 2d keypoint matching for 3d lidar odometry estimation,

C. Zheng, Y . Lyu, M. Li, and Z. Zhang, “Lodonet: A deep neural network with 2d keypoint matching for 3d lidar odometry estimation,” in ACM Int. Conf. Multimedia , 2020, pp. 2391–2399

2020
[18]

Lidar odometry by deep learning-based feature points with two-step pose estimation,

T. Liu, Y . Wang, X. Niu, L. Chang, T. Zhang, and J. Liu, “Lidar odometry by deep learning-based feature points with two-step pose estimation,” Remote Sensing, p. 2764, 2022

2022
[19]

Lo- net: Deep real-time lidar odometry,

Q. Li, S. Chen, C. Wang, X. Li, C. Wen, M. Cheng, and J. Li, “Lo- net: Deep real-time lidar odometry,” in IEEE Conf. Comput. Vis. Pattern Recog., 2019, pp. 8473–8482

2019
[20]

Pwclo-net: Deep lidar odometry in 3d point clouds using hierarchical embedding mask optimization,

G. Wang, X. Wu, Z. Liu, and H. Wang, “Pwclo-net: Deep lidar odometry in 3d point clouds using hierarchical embedding mask optimization,” in IEEE Conf. Comput. Vis. Pattern Recog. , 2021, pp. 15 910–15 919

2021
[21]

Delo: deep evidential lidar odometry using partial optimal transport,

S. A. Ali, D. Aouada, G. Reis, and D. Stricker, “Delo: deep evidential lidar odometry using partial optimal transport,” in Int. Conf. Comput. Vis., 2023, pp. 4517–4526

2023
[22]

Efficient 3d deep lidar odometry,

G. Wang, X. Wu, S. Jiang, Z. Liu, and H. Wang, “Efficient 3d deep lidar odometry,” IEEE Trans. Pattern Anal. Mach. Intell. , 2022

2022
[23]

Unsupervised geometry-aware deep lidar odometry,

Y . Cho, G. Kim, and A. Kim, “Unsupervised geometry-aware deep lidar odometry,” in IEEE Int. Conf. Rob. Auto. , 2020, pp. 2145–2152

2020
[24]

Unsupervised learning of 3d scene flow with lidar odometry assistance,

G. Wang, Z. Feng, C. Jiang, J. Liu, and H. Wang, “Unsupervised learning of 3d scene flow with lidar odometry assistance,” IEEE Trans. Intell. Transport. Syst., pp. 4557–4567, 2025

2025
[25]

Deeplo: Geometry-aware deep lidar odometry,

Y . Cho, G. Kim, and A. Kim, “Deeplo: Geometry-aware deep lidar odometry,” arXiv preprint arXiv:1902.10562 , 2019

1902 arXiv
[26]

Selfvoxelo: Self-supervised lidar odometry with voxel-based deep neural networks,

Y . Xu, Z. Huang, K.-Y . Lin, X. Zhu, J. Shi, H. Bao, G. Zhang, and H. Li, “Selfvoxelo: Self-supervised lidar odometry with voxel-based deep neural networks,” in Conference on Robot Learning , 2021, pp. 115–125

2021
[27]

Robust self- supervised lidar odometry via representative structure discovery and 3d inherent error modeling,

Y . Xu, J. Lin, J. Shi, G. Zhang, X. Wang, and H. Li, “Robust self- supervised lidar odometry via representative structure discovery and 3d inherent error modeling,” IEEE Rob. Auto. L. , pp. 1651–1658, 2022

2022
[28]

Linear least-squares optimization for point-to-plane icp surface registration,

K. L. Low, “Linear least-squares optimization for point-to-plane icp surface registration,” Chapel Hill, 2004. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 11

2004
[29]

Self-supervised learning of lidar odometry for robotic applications,

J. Nubert, S. Khattak, and M. Hutter, “Self-supervised learning of lidar odometry for robotic applications,” in IEEE Int. Conf. Rob. Auto., 2021, pp. 9601–9607

2021
[30]

A 3d lidar odometry for ugvs using coarse-to-fine deep scene flow estimation,

C. Li, F. Yan, S. Wang, and Y . Zhuang, “A 3d lidar odometry for ugvs using coarse-to-fine deep scene flow estimation,” Transactions of the Institute of Measurement and Control , pp. 274–286, 2023

2023
[31]

Self- supervised learning of lidar odometry based on spherical projection,

X. Fu, C. Liu, C. Zhang, Z. Sun, Y . Song, Q. Xu, and X. Yuan, “Self- supervised learning of lidar odometry based on spherical projection,” Int. J. Adv. Robotic Sys. , vol. 19, no. 1, 2022

2022
[32]

Hpplo-net: Unsupervised lidar odometry using a hierarchical point-to-plane solver,

B. Zhou, Y . Tu, Z. Jin, C. Xu, and H. Kong, “Hpplo-net: Unsupervised lidar odometry using a hierarchical point-to-plane solver,” IEEE Trans. Intel. Veh., pp. 2727–2739, 2023

2023
[33]

A-loam: Advanced implementation of loam,

T. Qin, S. Cao, J. Behley, and C. Stachniss, “A-loam: Advanced implementation of loam,” https://github.com/HKUST-Aerial-Robotics/ A-LOAM, 2019, accessed: 2025-06-15

2019
[34]

Pyicp slam

J. Giseop Kim, “Pyicp slam.” in https://github.com/JustWon/ PyICP-SLAM, 2023

2023
[35]

V oxelized gicp for fast and accurate 3d point cloud registration,

K. Koide, M. Yokozuka, S. Oishi, and A. Banno, “V oxelized gicp for fast and accurate 3d point cloud registration,” in IEEE Int. Conf. Rob. Auto., 2021, pp. 11 054–11 059. Beibei Zhou received her Ph.D. degree at the PCA Lab, he Key Lab of Intelligent Perception and Systems fo...

2021

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.

[1] [1]

Gaussian splatting slam,

H. Matsuki, R. Murai, P. H. Kelly, and A. J. Davison, “Gaussian splatting slam,” in IEEE Conf. Comput. Vis. Pattern Recog. , pp. 18 039–18 048

work page

[2] [2]

Fast-lidar-slam: A robust and real-time factor graph for urban scenarios with unstable gps signals,

Z. Rui, W. Xu, and Z. Feng, “Fast-lidar-slam: A robust and real-time factor graph for urban scenarios with unstable gps signals,” IEEE Trans. Intell. Transport. Syst. , pp. 20 043–20 058, 2024

work page 2024

[3] [3]

Dynamic semantic slam based on panoramic camera and lidar fusion for autonomous driving,

F. Li, C. Fu, J. Wang, and D. Sun, “Dynamic semantic slam based on panoramic camera and lidar fusion for autonomous driving,” IEEE Trans. Intell. Transport. Syst. , pp. 1–14, 2025

work page 2025

[4] [4]

Perception and sensing for autonomous vehicles under adverse weather conditions: A survey,

Y . Zhang, A. Carballo, H. Yang, and K. Takeda, “Perception and sensing for autonomous vehicles under adverse weather conditions: A survey,” ISPRS J. Pho. Remote Sen. , pp. 146–177, 2023

work page 2023

[5] [5]

Libre: The multiple 3d lidar dataset,

A. Carballo, J. Lambert, A. Monrroy, D. Wong, P. Narksri, Y . Kitsukawa, E. Takeuchi, S. Kato, and K. Takeda, “Libre: The multiple 3d lidar dataset,” in IEEE Intelligent Veh. Symp. IEEE, 2020, pp. 1094–1101

work page 2020

[6] [6]

The perception system of intelligent ground vehicles in all weather conditions: A systematic literature review,

A. S. Mohammed, A. Amamou, F. K. Ayevide, S. Kelouwani, K. Ag- bossou, and N. Zioui, “The perception system of intelligent ground vehicles in all weather conditions: A systematic literature review,” Sensors, 2020

work page 2020

[7] [7]

Robust object detection in challenging weather conditions,

H. Gupta, O. Kotlyar, H. Andreasson, and A. J. Lilienthal, “Robust object detection in challenging weather conditions,” in Winter Conf. Appl. Comput. Vis. , 2024, pp. 7523–7532

work page 2024

[8] [8]

Multi-agent trajectory prediction by combining egocentric and allocentric views,

X. Jia, L. Sun, H. Zhao, M. Tomizuka, and W. Zhan, “Multi-agent trajectory prediction by combining egocentric and allocentric views,” in Conference on Robot Learning . PMLR, 2022, pp. 1434–1443

work page 2022