Pith. sign in

REVIEW 5 major objections 5 minor 29 references

Online Dynamic SLAM with Incremental Smoothing and Mapping

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

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

Pith's one-line read Dynamic SLAM can be solved online by anchoring object points in an embedded frame and solving each object's graph in parallel.

desk verdict A genuinely useful Hybrid factor-graph formulation for Dynamic SLAM with sparse Bayes trees, undercut by an abstract that overstates the accuracy of the only online variant. read the letter →

arxiv 2509.08197 v1 pith:XDDMXAVA submitted 2025-09-10 cs.RO

Jesse Morris , Yiduo Wang , Viorela Ila This is my paper · ORCID
classification cs.RO
keywords DynamicSLAMincrementalsmoothingandmappingiSAM2factorgraphsBayestreeobjectmotionestimationonlinerigid-body
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

This paper sets out to make Dynamic SLAM—jointly estimating a moving camera, a static scene, and the rigid motions of moving objects—fast enough to run online. Its central proposal is a Hybrid factor-graph representation that anchors each dynamic object's points in a fixed embedded object frame while representing motion as a world-frame transform, so that dynamic points become leaf nodes and the graph stays sparse under incremental solving. On top of that, the Parallel-Hybrid architecture cuts the graph into one static and several per-object graphs, conditioned on the camera pose, and solves each with its own iSAM2 instance. The authors report that this runs at 1–5 Hz, about five times faster than the state-of-the-art baseline, with camera-pose and object-motion accuracy equal to or better than the baseline on most tested sequences, while noting a small accuracy cost from decoupling.

What carries the argument

The central object is the Hybrid factor graph: object points stored as static variables in an embedded object frame {L_e}, motions represented as world-centric transforms W_e H_k, and a ternary object smoothing factor r_{δH} written in the body frame. Its job is to make each dynamic point a leaf node so variable elimination under iSAM2 produces small Bayes-tree cliques, while still enforcing rigid-body kinematics. The Parallel-Hybrid architecture then cuts this graph along the camera–object edges, giving each dynamic object its own factor graph conditioned by a camera-pose prior, so multiple iSAM2 instances solve independently in parallel.

What would settle it

Track a sequence where the camera is stationary and a single large object moves across the field of view, then compare Parallel-Hybrid's recovered camera pose against the batch-joint Hybrid solution. If the object's motion measurably changes the camera pose estimate, the one-way conditioning assumption is violated.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that Dynamic SLAM is not inherently batch-bound: its factor graph can be arranged so that incremental solvers keep small cliques. The Hybrid formulation expresses all object points in a fixed embedded frame {L_e} and all motions as relative world-frame transforms W_e H_k; each dynamic point appears once, as a leaf, and the object smoothing factor is written in the body frame to avoid distance-dependent distortion. The authors further show that the joint graph's remaining bottleneck is the camera–object connection, and that cutting it—conditioning each object graph on the static camera pose prior—lets independent iSAM2 solvers run in parallel. The re

Load-bearing premise

That feeding each dynamic-object graph only the static camera pose prior, with no feedback from object motion back into the camera estimate, gives up negligible accuracy.

Editorial extensions

If this is right

  • The Hybrid formulation lets incremental solvers (iSAM2) update Dynamic SLAM graphs without the memory failures the baseline suffers on multi-object sequences.
  • Parallel-Hybrid achieves 1–5 Hz optimisation and a roughly 5× speed-up over the batch world-centric baseline, making Dynamic SLAM usable for online robot operation.
  • Solving the full joint problem remains more accurate than decoupling, so accuracy and speed trade off directly against each other.
  • Problem structure in Dynamic SLAM depends not only on ego-motion but on each object's motion relative to the camera, since longer visible tracklets enlarge cliques.
  • The embedded-frame anchor accumulates object structure over time, so object pose and velocity at every frame can be read out directly from the motion variables.

Reading between the lines

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

  • The authors' Bayes-tree analysis suggests a natural extension: when objects are tracked continuously for long stretches, eliminating landmark variables last via Schur-complement-style updates could further cut computation—the paper flags this as future work.
  • Because the accuracy gap between joint and decoupled solving is attributed to lost feedback from objects to camera, a bidirectional variant that occasionally reintroduces object-derived constraints on the camera pose could recover joint accuracy while keeping parallel speed.
  • The observed dependence of clique size on object–camera relative rotation implies an active-perception strategy: a robot that keeps object faces visible and relative rotation moderate can keep its incremental updates cheap.
  • The embedded-frame formulation is a natural host for class-specific motion models or learned pose priors, since the smoothing residual already lives in the body frame.
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

5 major / 5 minor

Summary. The paper proposes a factor-graph formulation for Dynamic SLAM called Hybrid, which represents object points in an embedded object frame while estimating object motions as world-centric relative transformations. The authors argue this improves sparsity and makes the problem amenable to incremental inference with iSAM2. They also introduce Parallel-Hybrid, an architecture that decouples the static and per-object dynamic factor graphs, with unidirectional information flow from the static to the dynamic subproblems, and evaluate the approach on KITTI, Outdoor Cluster, OMD, TartanAir, and VIODE. The headline claims are that the method matches or exceeds state-of-the-art accuracy, is the first application of incremental optimisation to Dynamic SLAM, achieves a 5x speed-up, and runs online at 1-5 Hz.

Significance. If the claims held, the paper would make a useful contribution: it identifies a structural reason why naive Dynamic SLAM factor graphs are expensive under incremental solvers, proposes a representation that reduces clique sizes, and provides an extensive empirical study on multiple datasets. The Bayes-tree analysis in Section VI-C.1 is a valuable diagnostic and the authors are transparent about the failure modes of the baseline and of their own iHybrid variant. The formulation itself appears internally consistent, and the open-source implementation in DynoSAM/GTSAM is a practical asset. However, the significance is reduced by the gap between the headline accuracy/speed claims and the evidence: the only online variant, Parallel-Hybrid, is consistently less accurate than joint estimation, and the claimed 5x speed-up is measured against a baseline that fails on most sequences. The evaluation also compares only against the authors' own DynoSAM baseline, not against other published Dynamic SLAM systems.

major comments (5)
  1. [Abstract and Section VI-B] The abstract and introduction claim accuracy 'equal to or better than state-of-the-art' and 'minimal to no accuracy loss', but Section VI-B and Table I show that Parallel-Hybrid, the only variant that completes all sequences and runs online, performs consistently worse than the jointly solved methods, regardless of solver. For example, Table I reports Parallel-Hybrid MEr differences as large as -1.29 deg on KITTI 02 and ATE differences of -0.30 m and -0.23 m on OMD S4U and Outdoor Cluster L2. Since the online architecture explicitly discards information flow from dynamic objects back to the camera, the accuracy claims need to be either restricted to the batch/semi-batch Hybrid variant or accompanied by a clear statement that the online variant trades accuracy for speed.
  2. [Section VI-C.2 and Table II] The '5x speed-up over existing methods' claim is not supported by the reported comparison. Table II shows that iBaseline fails on nearly all sequences at lambda_rs = 10 and on many sequences even at lambda_rs = 1; the 5x factor is therefore computed on a very small, non-representative subset of sequences where the baseline happens to complete. Moreover, the table shows Parallel-Hybrid is only about 2x faster than iHybrid on a number of sequences, not 5x. The paper should report per-sequence speedups and explicitly state the denominators used to compute the 5x figure, or soften the claim to 'up to 5x on sequences where the baseline completes'.
  3. [Section VI-C.2 and Table II] The statement that Parallel-Hybrid 'achieves an optimisation frequency of between 1 Hz to 5 Hz' is inconsistent with Table II. Many reported average update times are above 1000 ms (e.g., KITTI 00-06 at both lambda_rs settings), corresponding to less than 1 Hz. The 1-5 Hz range appears to hold for only a subset of sequences. The authors should either report the distribution of update times per sequence or revise the frequency claim to reflect the Table II data.
  4. [Section VI-A and Table I] The only baseline is the authors' own DynoSAM/world-centric implementation. The abstract's 'state-of-the-art' comparison is therefore not demonstrated against independent published Dynamic SLAM systems such as DynaSLAM II, VDO-SLAM, ClusterVO, or MVO. This is not a fatal flaw, because the Hybrid-vs-world-centric comparison is informative for isolating formulation effects, but the 'state-of-the-art' phrasing should be changed to 'baseline' or the authors should add comparisons to external systems.
  5. [Abstract and Section II-B] The claim that this is 'the first application of incremental optimisation techniques to Dynamic SLAM' is stated too strongly. Several dynamic SLAM systems use incremental back-ends (e.g., fixed-lag smoothers or iSAM2-style updates) even if they do not emphasize Bayes-tree structure. The novelty claim should be narrowed to the specific factor-graph design for incremental Dynamic SLAM, or supported by a literature review that explicitly accounts for prior incremental back-ends in dynamic SLAM.
minor comments (5)
  1. [Throughout] Typographical errors and wording: 'In additional to' in Section I, 'charaterise' in Section VI-C, 'homogenous' in Section III-A, and 'notated' in the Fig. 3 caption. These should be corrected.
  2. [Table II] The table formatting is very difficult to parse because the three method columns and the two lambda_rs values are not visually separated. Please add clear column headers and delimiters for each method and parameter setting.
  3. [Section VI-D] The real-world timing claim reports 'iSAM2 updates under 150 ms' but does not include frontend or tracking time. The text should clarify that this is back-end-only update time, not end-to-end online performance.
  4. [Section VI-C.1] The failure of iBaseline is attributed to memory exhaustion from large cliques. Reporting memory usage and the exact sequence length at failure would strengthen the scalability argument.
  5. [Equation (9)] The formula for W_kHe and the surrounding notation are dense and would benefit from a short derivation or a reference to the cited Chirikjian et al. work, especially because the distinction from the inverse of W_eHk is emphasized in the text.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity; the central results are empirical measurements of a proposed factor-graph design, with only minor self-citation in the baseline choice.

full rationale

The paper's core claims are empirical: the Hybrid formulation produces smaller Bayes-tree cliques, iHybrid runs faster than an incremental baseline, and Parallel-Hybrid achieves 1–5 Hz while trading away some accuracy. These are measured outcomes, not quantities derived from fitted parameters or from definitions that presuppose the conclusion. The factor-graph residuals (Eq. 7 and Eq. 11) are defined from first principles using rigid-body motion and constant-motion models; the claimed sparsity is then validated by direct Bayes-tree analysis (Section VI-C.1), not by construction. The 5× speedup is a reported timing comparison against iBaseline, not a prediction forced by a fit. The main self-citation is the use of DynoSAM [2] as the baseline, which is the authors' own prior work. This is a self-reference in the experimental comparison, but it is not load-bearing in a circular sense: the comparison still measures real algorithm behavior, and the paper's conclusions would stand or fall on those measurements regardless of who wrote the baseline. The paper also explicitly acknowledges limitations that are correctness concerns rather than circularity: Parallel-Hybrid 'performs consistently worse than the methods solved jointly' (Section VI-B) and has unidirectional information flow (Section V). These are honest trade-offs, not disguised inputs. No step in the derivation reduces to its own input by construction, so circularity is minimal.

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

The central claim depends on standard robotics assumptions (rigid bodies, constant motion), the correctness of the iSAM2 algorithm, and the quality of the reused DynoSAM frontend. No new physical entities or ad hoc free parameters are introduced. The key unverified dependency is the arbitrary invariance of the embedded frame, imported from the authors' own prior work.

assumptions (5)
  • domain assumption Dynamic objects move as rigid bodies, so all points on an object share a single SE(3) motion.
    Assumed throughout the formulation (Eqs. 2, 3, 7); non-rigid objects are only tested qualitatively in Section VI-D.
  • domain assumption A constant motion model between consecutive frames is a valid prior for object motion.
    The object smoothing factor (Eq. 11) encodes this; it is analogous to a constant velocity assumption and may be violated by sudden object accelerations.
  • standard math The iSAM2 Bayes tree correctly and efficiently solves the linearized incremental nonlinear least squares problem.
    The entire incremental efficiency argument rests on the cited iSAM2 theory (Section III-C).
  • domain assumption The DynoSAM frontend provides reliable frame-to-frame correspondences, object segmentation, and static/dynamic classification.
    The paper states it uses the DynoSAM frontend for all measurements (Section III-B); errors in this frontend propagate into all evaluated results.
  • domain assumption The choice of the embedded object frame {L_e} does not affect the estimation result.
    The paper relies on prior work [2, 7, 18] for this invariance (Section IV-A); it is not proven in this paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Online Dynamic SLAM with Incremental Smoothing and Mapping." pith.science (2026). https://pith.science/paper/XDDMXAVA

@misc{pith2026250908197,
  author       = {Pith},
  title        = {Pith review of: Online Dynamic SLAM with Incremental Smoothing and Mapping},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XDDMXAVA}},
  note         = {Machine review of arXiv:2509.08197}
}
read the original abstract

Dynamic SLAM methods jointly estimate for the static and dynamic scene components, however existing approaches, while accurate, are computationally expensive and unsuitable for online applications. In this work, we present the first application of incremental optimisation techniques to Dynamic SLAM. We introduce a novel factor-graph formulation and system architecture designed to take advantage of existing incremental optimisation methods and support online estimation. On multiple datasets, we demonstrate that our method achieves equal to or better than state-of-the-art in camera pose and object motion accuracy. We further analyse the structural properties of our approach to demonstrate its scalability and provide insight regarding the challenges of solving Dynamic SLAM incrementally. Finally, we show that our formulation results in problem structure well-suited to incremental solvers, while our system architecture further enhances performance, achieving a 5x speed-up over existing methods.

Figures

Figures reproduced from arXiv: 2509.08197 by the authors.

Figure 1
Figure 1. Output of our proposed Dynamic SLAM system recorded in indoor environments with multiple moving objects. Our proposed method employs a novel Hybrid representation for dynamic objects and motions, allowing the camera pose (red trajectory) and static scene (black point cloud) as well as the motion, structure and pose of dynamic objects (uniquely coloured) to be estimated jointly and incrementally using iSAM2 [4]. We s… view at source ↗
Figure 3
Figure 3. Our Hybrid Dynamic SLAM representation combines the benefits of an object-centric point representation and a world￾centric representation for motions. Here we show a object with 5 points seen at three consecutive frames. The first frame, k −2, is used to define the objects embedded frame {Le} which is moved through time and space by the estimation object motion W eHk. For simplicity, we notate only one point Lem on … view at source ↗
Figure 2
Figure 2. A possible Bayes Tree generated using the world-centric formulation proposed in [7] from a scene containing 3 objects observed over 4 frames. Notice that following the iSAM2 algorithm creates large cliques which contains variables from all objects (highlighted in red), thus leading to inefficient computation [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Full Hybrid Dynamic SLAM factor-graph showing static points (blue) and camera poses (grey) at three consecutive frames. Two dynamic objects are shown, each with one point (orange), observed in all frames and connected by the proposed hybrid motion factors (ϕH). The ter…
Figure 5
Figure 5. Figure 5: Parallel-Hybrid approach decouples the static (a) and dynamic (b) components from the Hybrid formulation [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Bayes Tree evaluation results on a portion of KITTI 20. To demonstrate the effect of variable re-ordering and the natural scalability of the Hybrid compared to the Baseline we set λrs to 10 and 1; this parameter limits relinearisation to every n steps, thereby delaying…
Figure 7
Figure 7. Figure 7: Evaluation results of Parallel-Hybrid on omd-s4u: show￾ing accumulated iSAM2 update time per object, average clique size, number of landmark variables involved in the update and total number of (landmark) variables in the per-object state estimate. ble throughout, thus…
Figure 8
Figure 8. Figure 8: Per frame iSAM2 update time on selected sequences: (a) KITTI 20 and (b) VIODE Parking Lot (Mid). The failure points are highlighted (red). Grey lines indicate the number of objects tracked at each frame. will continue to explore incremental methods and problem formulat…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

29 extracted references · 20 canonical work pages

  1. [1]

    Multimotion Visual Odometry (MVO),

    K. M. Judd and J. D. Gammell, “Multimotion Visual Odometry (MVO),”Intl. J. of Robotics Research, 2024

    2024

  2. [2]

    Dynosam: Open-source smoothing and mapping framework for dynamic slam,

    J. Morris, Y . Wang, M. Kliniewski, and V . Ila, “Dynosam: Open-source smoothing and mapping framework for dynamic slam,”arXiv preprint arXiv:2501.11893, 2025

    arXiv 2025

  3. [3]

    isam: Incremental smoothing and mapping,

    M. Kaess, A. Ranganathan, and F. Dellaert, “isam: Incremental smoothing and mapping,”IEEE Transactions on Robotics, vol. 24, no. 6, pp. 1365–1378, 2008

    work page 2008

  4. [4]

    isam2: Incremental smoothing and mapping using the bayes tree,

    M. Kaess, H. Johannsson, R. Roberts, V . Ila, J. J. Leonard, and F. Dellaert, “isam2: Incremental smoothing and mapping using the bayes tree,”Intl. J. of Robotics Research, vol. 31, no. 2, pp. 216–235, 2012

    work page 2012

  5. [5]

    Factor graphs for robot perception,

    F. Dellaert, M. Kaess,et al., “Factor graphs for robot perception,” Foundations and Trends® in Robotics, vol. 6, no. 1-2, pp. 1–139, 2017

    work page 2017

  6. [6]

    Dynaslam ii: Tightly-coupled multi-object tracking and slam,

    B. Bescos, C. Campos, J. D. Tard ´os, and J. Neira, “Dynaslam ii: Tightly-coupled multi-object tracking and slam,”IEEE Robotics and Automation Letters, vol. 6, no. 3, pp. 5191–5198, 2021

    2021

  7. [7]

    The importance of coordinate frames in dynamic slam,

    J. Morris, Y . Wang, and V . Ila, “The importance of coordinate frames in dynamic slam,” inProc. of the IEEE Intl. Conf. on Robotics and Automation (ICRA), 2024

    2024

  8. [8]

    Dynaslam: Tracking, mapping, and inpainting in dynamic scenes,

    B. Bescos, J. M. F ´acil, J. Civera, and J. Neira, “Dynaslam: Tracking, mapping, and inpainting in dynamic scenes,”IEEE Robotics and Automation Letters, vol. 3, no. 4, pp. 4076–4083, 2018

    work page 2018

Show all 29 references
  1. [9]

    ORB-SLAM3: An accurate open-source library for visual, visual- inertial and multi-map SLAM,

    C. Campos, R. Elvira, J. J. G ´omez, J. M. M. Montiel, and J. D. Tard´os, “ORB-SLAM3: An accurate open-source library for visual, visual- inertial and multi-map SLAM,”IEEE Trans. Robotics, vol. 37, no. 6, pp. 1874–1890, 2021

    2021

  2. [10]

    Dynavins++: Robust visual-inertial state estimator in dynamic environments by adaptive truncated least squares and stable state recovery,

    S. Song, H. Lim, A. J. Lee, and H. Myung, “Dynavins++: Robust visual-inertial state estimator in dynamic environments by adaptive truncated least squares and stable state recovery,”IEEE Robotics and Automation Letters, 2024

    2024

  3. [11]

    VDO-SLAM: A Visual Dynamic Object-aware SLAM System,

    J. Zhang, M. Henein, R. Mahony, and V . Ila, “VDO-SLAM: A Visual Dynamic Object-aware SLAM System,”arXiv preprint arXiv:2005.11052, 2020

    2005 arXiv

  4. [12]

    Dynameshslam: A mesh-based dynamic visual slammot method,

    Y . Liu, C. Guo, Y . Luo, and Y . Wang, “Dynameshslam: A mesh-based dynamic visual slammot method,”IEEE Robotics and Automation Letters, vol. 9, no. 6, pp. 5791–5798, 2024

    2024

  5. [13]

    Dynaquadric: Dynamic quadric slam for quadric initialization, map- ping, and tracking,

    R. Tian, Y . Zhang, L. Yang, J. Zhang, S. Coleman, and D. Kerr, “Dynaquadric: Dynamic quadric slam for quadric initialization, map- ping, and tracking,”IEEE Trans. on Intelligent Transportation Systems, vol. 25, no. 11, pp. 17 234–17 246, 2024

    2024

  6. [14]

    Clusterslam: A slam backend for simultaneous rigid body clustering and motion estimation,

    J. Huang, S. Yang, Z. Zhao, Y . Lai, and S. Hu, “Clusterslam: A slam backend for simultaneous rigid body clustering and motion estimation,” inProc. of the Intl. Conf. on Computer Vision (ICCV), 2019, pp. 5874–5883

    2019

  7. [15]

    Clustervo: Clustering moving instances and estimating visual odometry for self and sur- roundings,

    J. Huang, S. Yang, T.-J. Mu, and S.-M. Hu, “Clustervo: Clustering moving instances and estimating visual odometry for self and sur- roundings,” inProc. of the IEEE/CVF Intl. Conf. Computer Vision and Pattern Recognition, 2020, pp. 2168–2177

    2020

  8. [16]

    Dot: Dynamic object tracking for visual slam,

    I. Ballester, A. Font ´an, J. Civera, K. H. Strobl, and R. Triebel, “Dot: Dynamic object tracking for visual slam,” inProc. of the IEEE Intl. Conf. on Robotics and Automation (ICRA), 2021, pp. 11 705–11 711

    2021

  9. [17]

    Dynamic SLAM: The Need for Speed,

    M. Henein, J. Zhang, R. Mahony, and V . Ila, “Dynamic SLAM: The Need for Speed,” inProc. of the IEEE Intl. Conf. on Robotics and Automation (ICRA), 2020, pp. 2123–2129

    2020

  10. [18]

    Pose changes from a different point of view,

    G. S. Chirikjian, R. Mahony, S. Ruan, and J. Trumpf, “Pose changes from a different point of view,” inProc. of the ASME Intl. Design Engineering Technical Conf. (IDETC). ASME, 2017

    2017

  11. [19]

    Kimera: an open- source library for real-time metric-semantic localization and mapping,

    A. Rosinol, M. Abate, Y . Chang, and L. Carlone, “Kimera: an open- source library for real-time metric-semantic localization and mapping,” inProc. of the IEEE Intl. Conf. on Robotics and Automation (ICRA), 2020

    2020

  12. [20]

    Svo: Semidirect visual odometry for monocular and multicamera systems,

    C. Forster, Z. Zhang, M. Gassner, M. Werlberger, and D. Scaramuzza, “Svo: Semidirect visual odometry for monocular and multicamera systems,”IEEE Trans. Robotics, vol. 33, no. 2, pp. 249–265, 2017

    2017

  13. [21]

    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,” inProc. of the IEEE/RSJ Intl. Conf. on Intelligent Robots and Systems (IROS), 2020, pp. 5135–5142

    2020

  14. [22]

    The bayes tree: An algorithmic foundation for probabilistic robot mapping,

    M. Kaess, V . Ila, R. Roberts, and F. Dellaert, “The bayes tree: An algorithmic foundation for probabilistic robot mapping,” inAlgorith- mic Foundations of Robotics IX: Selected Contributions of the Ninth International Workshop on the Algorithmic Foundations of Robotics. Sprin...

    2010

  15. [23]

    Mr-isam2: Incremental smoothing and mapping with multi-root bayes tree for multi-robot slam,

    Y . Zhang, M. Hsiao, J. Dong, J. Engel, and F. Dellaert, “Mr-isam2: Incremental smoothing and mapping with multi-root bayes tree for multi-robot slam,” inIROS. IEEE, 2021, pp. 8671–8678

    2021

  16. [24]

    Twistslam: Constrained slam in dynamic environment,

    M. Gonzalez, E. Marchand, A. Kacete, and J. Royan, “Twistslam: Constrained slam in dynamic environment,”IEEE Robotics and Au- tomation Letters, vol. 7, no. 3, pp. 6846–6853, 2022

    2022

  17. [25]

    Vision meets robotics: The KITTI dataset,

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

    2013

  18. [26]

    The Oxford Multimotion Dataset: Multiple SE(3) Motions with Ground Truth,

    K. M. Judd and J. D. Gammell, “The Oxford Multimotion Dataset: Multiple SE(3) Motions with Ground Truth,”IEEE Robotics and Automation Letters, vol. 4, no. 2, pp. 800–807, 2019

    2019

  19. [27]

    Airdos: Dynamic slam benefits from articulated objects,

    Y . Qiu, C. Wang, W. Wang, M. Henein, and S. Scherer, “Airdos: Dynamic slam benefits from articulated objects,” inProc. of the IEEE Intl. Conf. on Robotics and Automation (ICRA), 2022, pp. 8047–8053

    2022

  20. [28]

    Viode: A simulated dataset to address the challenges of visual-inertial odometry in dynamic environments,

    K. Minoda, F. Schilling, V . W¨uest, D. Floreano, and T. Yairi, “Viode: A simulated dataset to address the challenges of visual-inertial odometry in dynamic environments,”IEEE Robotics and Automation Letters, vol. 6, no. 2, pp. 1343–1350, 2021

    2021

  21. [29]

    borglab/gtsam,

    F. Dellaert and GTSAM Contributors, “borglab/gtsam,” May 2022. [Online]. Available: https://github.com/borglab/gtsam

    2022

Pith tools

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