Pith. sign in

REVIEW 3 major objections 5 minor 47 references

LiDAR-Based Vehicle Detection and Tracking for Autonomous Racing

T0 review · 3 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read This paper claims that a non-learned LiDAR pipeline—range-image segmentation, map-assisted pose estimation, and variable-step multi-target tracking—enables fully autonomous overtaking maneuvers at speeds above 275 km/h with an average…

desk verdict A real race-proven perception pipeline with a genuinely novel segmentation layer, but with an inverted core equation and an evaluation that needs stronger validation before the strong claims are accepted. read the letter →

arxiv 2501.14502 v1 pith:KFSRP7SD submitted 2025-01-24 cs.RO cs.CV

Marcello Cellina , Matteo Corno , Sergio Matteo Savaresi This is my paper · ORCID
classification cs.ROcs.CV
keywords LiDARperceptionpointcloudsegmentationrangeimagevehicleposeestimationmulti-targettrackingautonomousracingextendedKalmanfilterIndyChallenge
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 tries to prove that a purely algorithmic, label-free LiDAR perception pipeline can support competitive autonomous racing at speeds far beyond what is typical for road autonomy. Its central assertion is that a carefully designed sequence of three modules—a range-image-based point cloud segmentation that handles non-uniform, time-varying scan patterns without losing information, a vehicle pose estimator that blends a track-map heading assumption with a variance-based rectangle fit, and a variable-step multi-target tracker—delivers reliable opponent detection and tracking in real race conditions. If true, this matters because it shows that high-speed head-to-head maneuvers can be executed safely without deep learning, labeled datasets, or expensive ground-truth infrastructure, and at a latency below the LiDAR refresh rate. The evidence comes from official races where the system enabled fully autonomous overtakes at speeds above 275 km/h, with opponent detection out to 90 m ahead and 85 m behind.

What carries the argument

The load-bearing mechanism is a range image defined on the azimuth–layer plane, paired with an elevation image, so that every original 3D point is preserved even when the scan pattern is non-uniform and time-varying; this representation turns point-cloud segmentation into a 2D image-segmentation problem. The non-ground range image reparation (NGRIR) step fills ground-labeled pixels with a horizontal mean filter only when the symmetric depth difference is below a threshold, reconnecting clusters that a racecar's black suspension and flat surfaces would otherwise split. The heading estimator fuses a map-based reference $\hat{\psi}_{\text{REF}}$ (opponent assumed parallel to the track centerline) with a variance-minimizing L-shape fit $\hat{\psi}_{\text{VAR}}$, choosing the one whose fitted rectangle encloses more points, which lets the system track normally parallel cars while still recovering the true pose of a spinning car. The tracker is a variable-step Extended Kalman Filter with a constant-velocity turn-rate model, global-nearest-neighbour data association via Mahalanobis distance solved by the Munkres algorithm, and an M/N confirm/delete track-management logic; because the measurement vector contains only position, a vehicle in a spin does not corrupt the filter with a misleading heading.

What would settle it

Inject a controlled heading bias (for example, 1$^\circ$ to 5$^\circ$) into the ego pose used by the map-projection and heading-reference modules during a replayed race sequence and measure the false-positive rate in the $[-70,-90]$ m band and the opponent heading error; the paper's stated dependence on precise localization predicts both should degrade with the injected bias.

Watch

Extended reading notes

Core claim

The paper's central discovery is that the difficult perception problems of autonomous racing can be solved by exploiting the structure of the problem rather than learned representations. The segmentation step constructs a range image on the azimuth–layer plane plus an elevation image, so that unstructured and time-varying scans from MEMS LiDARs are processed losslessly; a non-ground range image reparation step then bridges disconnected components caused by black, non-reflective bodywork, and the three sensors are processed in parallel and merged. For pose, instead of relying on a rectangular shape model that racecars violate, the algorithm estimates heading from two independent sources—the track centerline direction and a variance-minimizing rectangle fit—and selects the one that contains more points inside its fitted rectangle, then constrains the box to the a-priori known vehicle dimensions. The tracking module uses a variable-step Extended Kalman Filter with a constant-velocity turn-rate model, Mahalanobis-distance gating with Munkres assignment, and an M/N track lifecycle, deliberately excluding heading from the measurement vector so that spinning opponents remain trackable. The combined pipeline is claimed to run at 38 Hz with an average processing delay of 26 ms, to detect opponents up to 90 m ahead and 85 m behind, and to have enabled fully autonomous overtaking maneuvers at speeds exceeding 275 km/h in official competition.

Load-bearing premise

The pipeline assumes that the ego vehicle has precise absolute localization and a map of the race track, and it leans on that map both to discard off-track clusters and to produce the primary heading estimate for opponents; if that localization or map is inaccurate, false positives appear at long range and the heading estimate for any car not running parallel to the centerline becomes unreliable.

Editorial extensions

If this is right

  • Opponent vehicles are detected up to 90 m ahead and 85 m behind the ego car, with a 50% detection probability at 80 m ahead, and the track-management logic keeps tracks alive through the rear-wing blind spot where raw detections drop.
  • The full perception loop runs at 38 Hz with an average processing delay of 26 ms, half the 50 ms sensor scan period, which is lower than the reported state of the art for this application.
  • Segmentation and pose estimation require no labeled data or trained networks, making the pipeline immediately deployable in domains where racing edge cases (debris, smoke, spinning vehicles) defeat learned detectors.
  • Because the three LiDAR point clouds are segmented in parallel on separate CPU cores, the approach scales to additional sensors as long as cores are available.
  • The modular architecture allows future neural-network components to replace the detection or pose modules without altering the tracking and data-association layers.

Reading between the lines

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

  • The map-based heading prior could be adapted to non-oval tracks with changing curvature, but the assumption that the opponent travels parallel to the centerline breaks during intentional side-by-side maneuvers or drifting; a curvature-aware prior or a learned fusion weight might mitigate that, which the paper does not test.
  • The false positives concentrated at [-70,-90] m are attributed to ego-heading sideslip projecting off-track objects into the map; fusing the GNSS course angle with an IMU yaw-rate bias estimate, or using the dual-estimator logic for the ego state, is a natural extension that the paper does not evaluate.
  • The NGRIR reparation mechanism is not specific to racing: any robot using a scanning LiDAR with low-reflectivity or elongated objects should benefit from reconnecting ground-separated components in range-image space, which could be tested on public road datasets.
  • A quantitative sensitivity analysis of tracking accuracy versus EGO localization error (e.g., adding synthetic pose noise) would directly test the paper's stated dependence on precise absolute localization, since the map is used both for cluster rejection and heading reference.
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.

Referee Report

3 major / 5 minor

Summary. The paper presents the LiDAR-based vehicle detection and tracking pipeline deployed on Team PoliMOVE's autonomous racecar for the Indy Autonomous Challenge. The pipeline consists of (i) a range-image-based segmentation with ground removal and a repair step (NGRIR), (ii) a vehicle pose estimator that fuses a map/centerline-based heading with an L-shape variance-minimization estimator, and (iii) a variable-step EKF multi-target tracker with GNN association and M/N track management. The system is evaluated on data from the 2023 IAC @ CES race, including 15 overtake maneuvers at speeds up to 278 km/h. The paper reports a 26 ms average processing delay, detection range up to 90 m ahead and 85 m behind, and successful autonomous overtaking at velocities exceeding 275 km/h.

Significance. If the reported results hold, the paper is a valuable real-world demonstration of a classical, non-learned perception stack operating at the extreme speeds and latencies of autonomous racing. Its strengths include evaluation on genuine race data with synchronized RTK GNSS ground truth, component-wise comparisons against literature benchmarks, and a detailed parameter table for the deployed configuration. The main algorithmic contribution, a range-image segmentation designed for unstructured, time-varying scan patterns, is a reasonable engineering extension of prior range-image methods, though the novelty is incremental relative to [19] and [26]. However, the paper does not release code or data, and the reproducibility of the central results is currently limited by the undocumented ground-truth labeling threshold and the inconsistent angle definition in Eq. (1). The system-level claim of enabling autonomous overtaking is credible given the race record, but the published quantitative evidence is component-level rather than an end-to-end controlled comparison.

major comments (3)
  1. [Section III.A, Eq. (1)] The definition of a_{i,j} is inconsistent with the stated angle between the x-y plane and the vector connecting two consecutive points with the same azimuth. For points represented by (r cos e, r sin e) in a vertical plane, the tangent of the angle with the x-y plane is (vertical difference)/(horizontal difference), i.e., (r_i sin e_i - r_{i+1} sin e_{i+1})/(r_i cos e_i - r_{i+1} cos e_{i+1}). Equation (1) uses the reciprocal ratio, so a flat ground (vertical difference near zero) would produce computed angles near 90° and would be incorrectly labeled as non-ground under the stated threshold thgnd = 20°. Since ground removal is the first stage of the proposed segmentation pipeline, this inconsistency is load-bearing; the equation must be corrected or the text must clarify the reference axis.
  2. [Section V.A.1] The ground-removal evaluation relies on an offline automatic labeling algorithm that fits planes, computes normal versors, and then imposes "a suitable threshold over the components of the normal versor" to label ground points. The threshold value is not reported anywhere, including Table I. This is a free parameter of the evaluation, and the reported TPR/PPV/F1 comparison against the benchmark of [47] cannot be reproduced or properly interpreted without it. The authors should report the threshold, analyze its sensitivity, and ideally validate the automatic labeler against a set of manually labeled scans. Without this, the claim that the proposed method outperforms the benchmark in ground removal is not fully supported.
  3. [Section V.A.2] The clustering evaluation uses two custom metrics based on an over-dimensioned bounding box of the single opponent vehicle derived from synchronized GNSS positions. The first metric counts clusters containing at least one vehicle point, and the second counts clusters containing both vehicle and non-vehicle points, with the latter identified heuristically by cluster size. This metric does not provide a per-point segmentation ground truth, and a mixed cluster that includes both the car and a wall would still be counted as a "car cluster" by the first metric. The authors should either provide a small manually labeled validation set or discuss the biases of these metrics more thoroughly; otherwise the quantitative comparison in Figure 10 is difficult to interpret.
minor comments (5)
  1. [Section III.A, Eq. (3)] The set definition for N(r_{i,j}) is ambiguous: "∃r_{i,k}" should likely be "∃r_{i,j+k}", and the symmetric-pair condition "|r_{j-k} − r_{j+k}| < thr" should be written with both row and column indices. Please correct the notation.
  2. [Section III.B, after Eq. (5)] The symbol ψ_HAT appears without definition; it should be ψ_VAR.
  3. [Section V.B] The statement "detection range up to 90 m ahead and 85 m behind" is based on Figure 11, but at 90 m the detection probability appears to be low in the figure. Please define "detection range" explicitly, for example as the distance at which the true positive rate exceeds a specified threshold.
  4. [Section VI] The conclusion states "our method demonstrates superior performance over current algorithmic approaches" without reporting an end-to-end comparison against a full detection-and-tracking baseline. The component-wise comparisons cover ground removal and clustering; tracking and pose estimation are evaluated in absolute terms against GNSS. This statement should be softened or supported by a full-system comparative experiment.
  5. [Throughout] There are several typos: "restrictive" for "restrictive" (Section II.D), "partecipate" for "participate" (Section III.B), "empoly" for "employ" (Section V.A.2), "consists om" for "consists of" (Section V.A.2), "Reparied" for "Repaired" (Section III.A), and "Sawitsky-Golay" in Table I. A careful proofread is needed.

Circularity Check

0 steps flagged · score 2.0 of 10

No circularity in the derivation chain: results are measured against external GNSS ground truth, and the only self-citations ([14], [39]) are non-load-bearing literature context; Eq. (1)'s printed angle formula is internally inconsistent, a correctness defect, not a circular reduction.

full rationale

The paper's derivation chain is audited for reduction-to-input. The central measurable outputs — opponent detection probability vs. range, tracking state errors, 26 ms average processing delay — are measured quantities, verified against synchronized RTK GNSS ground truth of EGO and opponent vehicles ('the timestamped RTK GNSS pose history of the two cars can be synchronized and smoothed in order to build a ground truth'), not against values fitted inside the paper. Segmentation and tracking parameters (Table I: thgnd=20°, thseg=2.5°, Q, R) are fixed operating constants; no parameter is optimized on a subset of data and then reported as a prediction of a closely related quantity. Pose estimation is not self-referential: ψ_REF is an open-loop map/centerline assumption that the paper itself states 'clearly fails under non-nominal conditions like a vehicle spinning without control or stopped sideways', and ψ_VAR is the classical variance-minimization L-shape fit of [34]; fusion rule (5) is an interior-point-count heuristic. Benchmarks are external ([47] for ground removal, [23] for PCL Euclidean clustering). Self-citations are minor and non-load-bearing: [14] (co-author Cellina) supports only the factual claim that the RACECAR dataset has no labeled version; [39] (co-authors Corno, Savaresi) supports only the existence of shape-free ETT techniques. Neither is an unverified premise on which a result depends; no uniqueness theorem is imported, no ansatz is smuggled through self-citation, and no known result is merely renamed as an organizing contribution. One flagged defect, Section III-A Eq. (1): under the standard elevation convention (z = r sin e) the printed arctangent is the reciprocal of the tangent of the stated angle with the x-y plane, so a literal implementation with thgnd=20° would classify flat track as non-ground; since the system demonstrably performed real overtakes, the published equation is likely inconsistent with the code — an auditability/correctness problem, but not a circular one, because the evaluation does not derive from that equation. Verdict: no significant circularity; the two minor non-load-bearing self-citations justify a score of 2 within the 0-2 band.

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

The central operational claims rest on external infrastructure (precise ego localization and track map) and a set of hand-tuned thresholds (Table I). The evaluation additionally depends on an unvalidated automatic ground-truth labeler and single-opponent assumptions. No new physical entities are introduced.

free parameters (7)
  • thseg (clustering angle threshold) = 2.5°
    Chosen after inspecting the performance curves in Fig. 10 on the evaluation dataset to reduce over-segmentation while avoiding mixed car-wall clusters; directly affects segmentation and detection.
  • thgnd (ground segmentation angle threshold) = 20.0°
    Hand-set threshold on the smoothed angle image to label ground points; affects detection range and false positives.
  • NGRIR window ws and depth threshold thr = ws=9, thr=5.0 m
    Hand-tuned parameters for repairing the non-ground range image; control merging of separated vehicle components.
  • thmrg (cluster merging distance) = 1.80 m
    Threshold for merging clusters across adjacent LiDAR sensors; influences whether a single opponent is split into multiple detections.
  • EKF covariance matrices Q_diag and R_diag = Q_diag={0.005, 0.005, 0.5, 0.005, 0.0005}, R_diag={5.0, 5.0}
    Tuned to favor smooth velocity estimation over position accuracy, as stated in Section V-C.
  • Track management N, Mc, Me = N=20, Mc=6, Me=5
    Thresholds for tentative/confirmed/dead tracks; affect persistence and false-track rejection.
  • Automatic GT labeling normal threshold
    The ground-truth labeler for ground-removal evaluation uses a 'suitable threshold' on normal vector components; this threshold is not specified, so the evaluation is not fully reproducible.
assumptions (7)
  • domain assumption The size of the vehicles to be tracked is a-priori known
    Used in Section III-B to impose W=W_EGO and L=L_EGO in rectangle fitting; valid because all IAC cars are Dallara AV21, but limits generality.
  • domain assumption Precise EGO vehicle absolute localization is available, together with a map of the operative environment
    Stated in Section III; used for map-based cluster filtering and for the psi_REF heading estimator.
  • domain assumption Opponents travel approximately parallel to the track centerline
    Underpins the map-based heading estimator psi_REF in Section III-B; the paper acknowledges it fails for spinning or sideways vehicles.
  • domain assumption Motion distortion in LiDAR scans is negligible because relative speed between vehicles is small
    Stated in Section III before III-A; this ignores distortion from stationary obstacles or large speed differentials.
  • domain assumption The Constant Velocity Turn Rate (CVTR) model describes opponent motion
    The EKF in Section III-C uses this model; it is standard but may be inaccurate during abrupt maneuvers.
  • domain assumption For clustering evaluation, only a single opponent is present on track with no other obstacles
    Used to build vehicle bounding-box ground truth from GNSS trajectories in Section V-A2; the paper acknowledges this assumption.
  • ad hoc to paper The offline automatic ground-truth labeler (plane fitting + normal threshold + K-NN cleaning) produces correct labels
    Used in Section V-A1 to compute TPR/PPV/F1; the threshold is not reported and the labeler is not validated against manual labels.

how reviews work

0 comments
Cite this review

Pith. "Pith review of LiDAR-Based Vehicle Detection and Tracking for Autonomous Racing." pith.science (2026). https://pith.science/paper/KFSRP7SD

@misc{pith2026250114502,
  author       = {Pith},
  title        = {Pith review of: LiDAR-Based Vehicle Detection and Tracking for Autonomous Racing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KFSRP7SD}},
  note         = {Machine review of arXiv:2501.14502}
}
read the original abstract

Autonomous racing provides a controlled environment for testing the software and hardware of autonomous vehicles operating at their performance limits. Competitive interactions between multiple autonomous racecars however introduce challenging and potentially dangerous scenarios. Accurate and consistent vehicle detection and tracking is crucial for overtaking maneuvers, and low-latency sensor processing is essential to respond quickly to hazardous situations. This paper presents the LiDAR-based perception algorithms deployed on Team PoliMOVE's autonomous racecar, which won multiple competitions in the Indy Autonomous Challenge series. Our Vehicle Detection and Tracking pipeline is composed of a novel fast Point Cloud Segmentation technique and a specific Vehicle Pose Estimation methodology, together with a variable-step Multi-Target Tracking algorithm. Experimental results demonstrate the algorithm's performance, robustness, computational efficiency, and suitability for autonomous racing applications, enabling fully autonomous overtaking maneuvers at velocities exceeding 275 km/h.

Figures

Figures reproduced from arXiv: 2501.14502 by the authors.

Figure 1
Figure 1. Team PoliMOVE’s Dallara AV21 “MinerVa” defending from an [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Representation of the main research problems in Vehicle Detection [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Block scheme of the LiDAR-Based Tracking algorithm [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: The five steps of the Point Cloud segmentation algorithm and the [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Input and Segmented Point Clouds from the three sensors capturing an [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Graphical comparison of the results of the two rectangle fitting ˆˆ [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: LiDAR sensors mounting positions and horizontal FOV coverage of [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: Violin plots of the computation time of the main modules of the [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]
Figure 9
Figure 9. Figure 9: Violin plots of the True Positive Rate (TPR), Positive Perceived Value [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]
Figure 10
Figure 10. Figure 10: Clustering performance computed as average and standard deviations [PITH_FULL_IMAGE:figures/full_fig_p010_10.png]
Figure 11
Figure 11. Figure 11: LiDAR detection (green) and Target Tracking confirmation (purple) [PITH_FULL_IMAGE:figures/full_fig_p010_11.png]
Figure 12
Figure 12. Figure 12: Violin plots of the raw measurements and tracked state estimation error computed with respect to the ground truth. The [PITH_FULL_IMAGE:figures/full_fig_p011_12.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

47 extracted references · 37 canonical work pages

  1. [19]

    Fast range image-based segmentation of sparse 3d laser scans for online operation,

    I. Bogoslavskyi and C. Stachniss, “Fast range image-based segmentation of sparse 3d laser scans for online operation,” in 2016 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) . IEEE, 2016, pp. 163–169

    work page 2016

  2. [26]

    Efficient online segmentation for sparse 3d laser scans,

    I. Bogoslavskyi and C. Stachniss, “Efficient online segmentation for sparse 3d laser scans,” PFG–Journal of Photogrammetry, Remote Sens- ing and Geoinformation Science , vol. 85, pp. 41–52, 2017

    work page 2017

  3. [47]

    Real-time object classification in 3d point clouds using point feature histograms,

    M. Himmelsbach, T. Luettel, and H.-J. Wuensche, “Real-time object classification in 3d point clouds using point feature histograms,” in 2009 IEEE/RSJ International Conference on Intelligent Robots and Systems . IEEE, 2009, pp. 994–1000

    work page 2009

  4. [1]

    Indy au- tonomous challenge-autonomous race cars at the handling limits,

    A. Wischnewski, M. Geisslinger, J. Betz, T. Betz, F. Fent, A. Heilmeier, L. Hermansdorfer, T. Herrmann, S. Huch, P. Karle et al. , “Indy au- tonomous challenge-autonomous race cars at the handling limits,” in 12th International Munich Chassis Symposium 2021: chassis. tech plus . Springer, 2022, pp. 163–182

    2021

  5. [2]

    A review of tracking and trajectory prediction methods for autonomous driving,

    F. Leon and M. Gavrilescu, “A review of tracking and trajectory prediction methods for autonomous driving,” Mathematics, vol. 9, no. 6, p. 660, 2021

    work page 2021

  6. [3]

    Multi-modal 3d object detection in autonomous driving: a survey,

    Y . Wang, Q. Mao, H. Zhu, J. Deng, Y . Zhang, J. Ji, H. Li, and Y . Zhang, “Multi-modal 3d object detection in autonomous driving: a survey,” International Journal of Computer Vision , vol. 131, no. 8, pp. 2122– 2152, 2023

    work page 2023

  7. [4]

    3d object detection for autonomous driving: A comprehensive survey,

    J. Mao, S. Shi, X. Wang, and H. Li, “3d object detection for autonomous driving: A comprehensive survey,” International Journal of Computer Vision, pp. 1–55, 2023

    work page 2023

  8. [5]

    Model-free vehicle tracking and state estimation in point cloud sequences,

    Z. Pang, Z. Li, and N. Wang, “Model-free vehicle tracking and state estimation in point cloud sequences,” in 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) . IEEE, 2021, pp. 8075–8082

    work page 2021

Show all 47 references
  1. [6]

    Ditnet: End-to-end 3d object detection and track id assignment in spatio-temporal world,

    S. Wang, P. Cai, L. Wang, and M. Liu, “Ditnet: End-to-end 3d object detection and track id assignment in spatio-temporal world,” IEEE Robotics and Automation Letters , vol. 6, no. 2, pp. 3397–3404, 2021

    2021

  2. [7]

    3d-siamrpn: An end-to-end learning method for real-time 3d single object tracking using raw point cloud,

    Z. Fang, S. Zhou, Y . Cui, and S. Scherer, “3d-siamrpn: An end-to-end learning method for real-time 3d single object tracking using raw point cloud,” IEEE Sensors Journal , vol. 21, no. 4, pp. 4995–5011, 2020

    2020

  3. [8]

    Pointpillars: Fast encoders for object detection from point clouds,

    A. H. Lang, S. V ora, H. Caesar, L. Zhou, J. Yang, and O. Beijbom, “Pointpillars: Fast encoders for object detection from point clouds,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 12 697–12 705

    2019

  4. [9]

    Pv-rcnn: Point-voxel feature set abstraction for 3d object detection,

    S. Shi, C. Guo, L. Jiang, Z. Wang, J. Shi, X. Wang, and H. Li, “Pv-rcnn: Point-voxel feature set abstraction for 3d object detection,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 10 529–10 538

    2020

  5. [10]

    A survey on deep-learning-based lidar 3d object detection for autonomous driving,

    S. Y . Alaba and J. E. Ball, “A survey on deep-learning-based lidar 3d object detection for autonomous driving,” Sensors, vol. 22, no. 24, p. 9577, 2022

    2022

  6. [11]

    Vision meets robotics: The kitti dataset,

    A. Geiger, P. Lenz, C. Stiller, and R. Urtasun, “Vision meets robotics: The kitti dataset,” The International Journal of Robotics Research , vol. 32, no. 11, pp. 1231–1237, 2013

    2013

  7. [12]

    nuscenes: A multimodal dataset for autonomous driving,

    H. Caesar, V . Bankiti, A. H. Lang, S. V ora, V . E. Liong, Q. Xu, A. Krishnan, Y . Pan, G. Baldan, and O. Beijbom, “nuscenes: A multimodal dataset for autonomous driving,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 11 621–11 631

    2020

  8. [13]

    Scalability in perception for autonomous driving: Waymo open dataset,

    P. Sun, H. Kretzschmar, X. Dotiwalla, A. Chouard, V . Patnaik, P. Tsui, J. Guo, Y . Zhou, Y . Chai, B. Caine et al. , “Scalability in perception for autonomous driving: Waymo open dataset,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2...

    2020

  9. [14]

    Racecar–the dataset for high- speed autonomous racing,

    A. Kulkarni, J. Chrosniak, E. Ducote, F. Sauerbeck, A. Saba, U. Chir- imar, J. Link, M. Cellina, and M. Behl, “Racecar–the dataset for high- speed autonomous racing,” in 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) . IEEE, 2023, pp. 11 458– 11 463

    2023

  10. [15]

    Learn to see fast: Lessons learned from autonomous racing on how to develop perception systems,

    F. Sauerbeck, S. Huch, F. Fent, P. Karle, D. Kulmer, and J. Betz, “Learn to see fast: Lessons learned from autonomous racing on how to develop perception systems,” IEEE Access, 2023

    2023

  11. [16]

    Model based vehicle detection and tracking for autonomous urban driving,

    A. Petrovskaya and S. Thrun, “Model based vehicle detection and tracking for autonomous urban driving,” Autonomous Robots , vol. 26, no. 2-3, pp. 123–139, 2009

    2009

  12. [17]

    Detection, classification and tracking of moving objects in a 3d environment,

    A. Azim and O. Aycard, “Detection, classification and tracking of moving objects in a 3d environment,” in 2012 IEEE Intelligent Vehicles Symposium. IEEE, 2012, pp. 802–807

    2012

  13. [18]

    Fast segmentation of 3d point clouds for ground vehicles,

    M. Himmelsbach, F. V . Hundelshausen, and H.-J. Wuensche, “Fast segmentation of 3d point clouds for ground vehicles,” in 2010 IEEE Intelligent Vehicles Symposium. IEEE, 2010, pp. 560–565

    2010

  14. [20]

    An autonomous system for head-to-head race: Design, implementation and analysis; team kaist at the indy autonomous challenge,

    C. Jung, A. Finazzi, H. Seong, D. Lee, S. Lee, B. Kim, G. Gang, S. Han, and D. H. Shim, “An autonomous system for head-to-head race: Design, implementation and analysis; team kaist at the indy autonomous challenge,” arXiv preprint arXiv:2303.09463 , 2023

    2023 arXiv

  15. [21]

    Tum autonomous motorsport: An autonomous racing software for the indy autonomous challenge,

    J. Betz, T. Betz, F. Fent, M. Geisslinger, A. Heilmeier, L. Hermansdorfer, T. Herrmann, S. Huch, P. Karle, M. Lienkamp et al., “Tum autonomous motorsport: An autonomous racing software for the indy autonomous challenge,” Journal of Field Robotics, vol. 40, no. 4, pp. 783–809, 2023

    2023

  16. [22]

    Sloped terrain segmentation for autonomous drive using sparse 3d point cloud,

    S. Cho, J. Kim, W. Ikram, K. Cho, Y .-S. Jeong, K. Um, and S. Sim, “Sloped terrain segmentation for autonomous drive using sparse 3d point cloud,” The Scientific World Journal , vol. 2014, 2014

    2014

  17. [23]

    3d is here: Point cloud library (pcl),

    R. B. Rusu and S. Cousins, “3d is here: Point cloud library (pcl),” in 2011 IEEE international conference on robotics and automation. IEEE, 2011, pp. 1–4

    2011

  18. [24]

    On the segmentation of 3d lidar point clouds,

    B. Douillard, J. Underwood, N. Kuntz, V . Vlaskine, A. Quadros, P. Mor- ton, and A. Frenkel, “On the segmentation of 3d lidar point clouds,” in 2011 IEEE International Conference on Robotics and Automation . IEEE, 2011, pp. 2798–2805

    2011

  19. [25]

    Fast segmentation of 3d point clouds: A paradigm on lidar data for autonomous vehicle applications,

    D. Zermas, I. Izzat, and N. Papanikolopoulos, “Fast segmentation of 3d point clouds: A paradigm on lidar data for autonomous vehicle applications,” in 2017 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2017, pp. 5067–5073

    2017

  20. [27]

    Rangenet++: Fast and accurate lidar semantic segmentation,

    A. Milioto, I. Vizzo, J. Behley, and C. Stachniss, “Rangenet++: Fast and accurate lidar semantic segmentation,” in 2019 IEEE/RSJ international conference on intelligent robots and systems (IROS) . IEEE, 2019, pp. 4213–4220

    2019

  21. [28]

    Cylindrical and asymmetrical 3d convolution networks for lidar segmentation,

    X. Zhu, H. Zhou, T. Wang, F. Hong, Y . Ma, W. Li, H. Li, and D. Lin, “Cylindrical and asymmetrical 3d convolution networks for lidar segmentation,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2021, pp. 9939–9948

    2021

  22. [29]

    Algorithms for the reduction of the number of points required to represent a digitized line or its caricature,

    D. H. Douglas and T. K. Peucker, “Algorithms for the reduction of the number of points required to represent a digitized line or its caricature,” Cartographica: the international journal for geographic information and geovisualization, vol. 10, no. 2, pp. 112–122, 1973

    1973

  23. [30]

    Object detection and tracking using multi-layer laser for autonomous urban driving,

    Y . Ye, L. Fu, and B. Li, “Object detection and tracking using multi-layer laser for autonomous urban driving,” in 2016 IEEE 19th international conference on intelligent transportation systems (ITSC) . IEEE, 2016, pp. 259–264

    2016

  24. [31]

    Dynamic multi-lidar based multiple object detection and tracking,

    M. Sualeh and G.-W. Kim, “Dynamic multi-lidar based multiple object detection and tracking,” Sensors, vol. 19, no. 6, p. 1474, 2019

    2019

  25. [32]

    Efficient l-shape fitting of laser scanner data for vehicle pose estimation,

    X. Shen, S. Pendleton, and M. H. Ang, “Efficient l-shape fitting of laser scanner data for vehicle pose estimation,” in2015 IEEE 7th International Conference on Cybernetics and Intelligent Systems (CIS) and IEEE Conference on Robotics, Automation and Mechatronics (RAM) . IEEE,...

    2015

  26. [33]

    L-shape fitting-based vehicle pose estimation and tracking using 3d-lidar,

    C. Zhao, C. Fu, J. M. Dolan, and J. Wang, “L-shape fitting-based vehicle pose estimation and tracking using 3d-lidar,” IEEE Transactions on Intelligent Vehicles, vol. 6, no. 4, pp. 787–798, 2021

    2021

  27. [34]

    Efficient l-shape fitting for vehicle detection using laser scanners,

    X. Zhang, W. Xu, C. Dong, and J. M. Dolan, “Efficient l-shape fitting for vehicle detection using laser scanners,” in 2017 IEEE Intelligent Vehicles Symposium (IV). IEEE, 2017, pp. 54–59

    2017

  28. [35]

    L-shape model switching- based precise motion tracking of moving vehicles using laser scanners,

    D. Kim, K. Jo, M. Lee, and M. Sunwoo, “L-shape model switching- based precise motion tracking of moving vehicles using laser scanners,” IEEE Transactions on Intelligent Transportation Systems, vol. 19, no. 2, pp. 598–612, 2017. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUG...

    2017

  29. [36]

    Lidar-based object track- ing and shape estimation using polylines and free-space information,

    S. Kraemer, C. Stiller, and M. E. Bouzouraa, “Lidar-based object track- ing and shape estimation using polylines and free-space information,” in 2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2018, pp. 4515–4522

    2018

  30. [37]

    Design and analysis of modern tracking systems(book),

    S. Blackman and R. Popoli, “Design and analysis of modern tracking systems(book),” Norwood, MA: Artech House, 1999. , 1999

    1999

  31. [38]

    Multitarget tracking,

    B.-n. V o, M. Mallick, Y . Bar-Shalom, S. Coraluppi, R. Osborne, R. Mahler, and B.-t. V o, “Multitarget tracking,” Wiley encyclopedia of electrical and electronics engineering , no. 2015, 2015

    2015

  32. [39]

    Design of an automated street crossing management module for a delivery robot,

    R. Pieroni, M. Corno, F. Parravicini, and S. M. Savaresi, “Design of an automated street crossing management module for a delivery robot,” Control Engineering Practice , vol. 153, p. 106095, 2024

    2024

  33. [40]

    A multiple model phd approach to tracking of cars under an assumed rectangular shape,

    K. Granstr ¨om, S. Reuter, D. Meissner, and A. Scheel, “A multiple model phd approach to tracking of cars under an assumed rectangular shape,” in 17th International Conference on Information Fusion (FUSION). IEEE, 2014, pp. 1–8

    2014

  34. [41]

    Extended object tracking in curvilinear road coordinates for autonomous driving,

    P. Dahal, S. Mentasti, S. Arrigoni, F. Braghin, M. Matteucci, and F. Cheli, “Extended object tracking in curvilinear road coordinates for autonomous driving,” IEEE Transactions on Intelligent Vehicles , 2022

    2022

  35. [42]

    Multi- modal sensor fusion and object tracking for autonomous racing,

    P. Karle, F. Fent, S. Huch, F. Sauerbeck, and M. Lienkamp, “Multi- modal sensor fusion and object tracking for autonomous racing,” IEEE Transactions on Intelligent Vehicles, 2023

    2023

  36. [43]

    Tracking and behavior reasoning of moving vehicles based on roadway geometry constraints,

    K. Jo, M. Lee, J. Kim, and M. Sunwoo, “Tracking and behavior reasoning of moving vehicles based on roadway geometry constraints,” IEEE transactions on intelligent transportation systems , vol. 18, no. 2, pp. 460–476, 2016

    2016

  37. [44]

    Accessed: 2023-06-29

    dSPACE AUTERA Technical Details. Accessed: 2023-06-29. [Online]. Available: https://www.dspace.com/en/inc/home/products/hw/autera.cfm

    2023

  38. [45]

    Robot operating system 2: Design, architecture, and uses in the wild,

    S. Macenski, T. Foote, B. Gerkey, C. Lalancette, and W. Woodall, “Robot operating system 2: Design, architecture, and uses in the wild,” Science Robotics, vol. 7, no. 66, p. eabm6074, 2022

    2022

  39. [46]

    Accessed: 2023-06-29

    Luminar Hydra Specs. Accessed: 2023-06-29. [Online]. Avail- able: https://autonomoustuff.com/-/media/Images/Hexagon/Hexagon% 20Core/autonomousstuff/pdf/Luminar Hydra Specs

    2023

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.