Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

OpenBench: A New Benchmark and Baseline for Semantic Navigation in Smart Logistics

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · 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 robot can deliver to a named unit using only OpenStreetMap and vision-language models, with no pre-mapping and no GPS.

desk verdict A real new benchmark and a plausible OSM+LLM+VLM baseline, but the GPS-free claim needs direct localization validation and the metric equations have a bug. read the letter →

arxiv 2502.09238 v1 pith:DEJ6C5NW submitted 2025-02-13 cs.RO

Junhui Wang , Dongjie Huo , Zehui Xu , Yongliang Shi , Yimin Yan , Yuanxin Wang , Chao Gao , Yan Qiao
show 1 more author
Guyue Zhou
This is my paper · ORCID
classification cs.RO
keywords semanticnavigationlast-miledeliveryOpenStreetMapLLMtaskplanningVLMgloballocalizationGPS-freebenchmarkmobilerobots
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 argues that outdoor semantic navigation for last-mile delivery does not need pre-built high-precision maps or GPS. It proposes the OPEN system, which turns a natural-language delivery instruction into a task sequence with an LLM, routes the robot over OpenStreetMap road data, and uses a vision-language model to localize globally and recognize house numbers. To test this, the authors introduce OpenBench, a benchmark with three simulated residential worlds and corresponding OSM maps, plus long-term metrics that weight earlier tasks in a multi-drop sequence. The central claim is that this combination delivers reliable, efficient navigation: 100% success rate in the small and medium simulated worlds, and successful real-world completion of a two-building delivery with path efficiency close to human teleoperation.

What carries the argument

The central mechanism is a closed loop that turns OpenStreetMap into both a route graph and a localization anchor. An LLM parses a delivery instruction into hierarchical addresses and queries OSM; a routing module converts OSM into a hierarchical graph and runs Multi-Level Dijkstra for navigation mode, or samples around a building's inflated concave hull for exploration mode. At the same time, a VLM-based pipeline — MobileSAM segmentation, CLIP semantic labeling, BEV projection, and 2D registration against OSM geometry — produces global poses that correct FastLIO2 odometry drift through factor-graph prior factors. Map updates write newly detected house-number plates back into OSM, which is what later deliveries exploit.

What would settle it

Run the OPEN system in the medium simulation world while recording the VLM registration's estimated global pose against the simulator's ground-truth pose; if the median registration error exceeds the 10 m success radius or the registration fails on more than a small fraction of frames, the GPS-free navigation claim is contradicted.

Watch

Extended reading notes

Core claim

The central claim is that a robot can deliver parcels to a named unit of a named building without any pre-mapping and without GPS, using only publicly available OpenStreetMap data and off-the-shelf foundation models. The OPEN system realizes this by having an LLM parse and verify addresses, group nearby stops, and query OSM; by generating waypoints either from OSM routing (navigation mode) or from uniform sampling around a building's concave hull (exploration mode); and by performing global localization with MobileSAM segmentation, CLIP semantic labeling, BEV projection, and 2D registration against OSM geometry. This VLM-based registration corrects FastLIO2 odometry drift in a factor graph, keeping localization bounded. In simulation the system reaches 100% SR with 35.93% SPL in the small world and 100% SR with 75.58% LSR in the medium world; in a real campus it completed a sequential two-building delivery with SPL near the 96.1% achieved by a human teleoperator.

Load-bearing premise

The whole system works only if the VLM-based global localization (MobileSAM plus CLIP plus BEV registration) is accurate enough to keep FastLIO2 odometry drift inside the 10 m success radius, and the paper reports no error, failure rate, or sensitivity analysis for that module.

Editorial extensions

If this is right

  • Deploying a delivery robot in a new residential area reduces to downloading OSM and running the system, since no LiDAR pre-mapping or GPS infrastructure is required.
  • The OpenBench metrics SRTP, SR, SPL, LSR, and LSPL give a shared protocol for comparing outdoor semantic navigation systems, including long-horizon multi-drop behavior.
  • Because OSM maps are roughly 1 percent the storage size of point cloud maps and 0.01 percent the size of the tested topological maps, onboard map storage ceases to be a bottleneck for small robots.
  • Map updates that add house-number locations to OSM improve path efficiency by 6 to 48 percent on the tested destinations, so a robot fleet operating in an area would keep getting more efficient over time.

Reading between the lines

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

  • A natural extension would be to measure the VLM global localization's pose error against ground truth in each benchmark world; the paper reports none, and every downstream metric depends on that error staying below the 10 m success radius.
  • The exponential-decay weighting in LSR and LSPL makes metrics sensitive to early-task success; a testable variant would use delivery deadlines or customer-priority weights instead of a fixed decay rate.
  • In areas where OSM lacks building footprints or house numbers, exploration mode becomes the bottleneck; evaluating the system in sparse-OSM rural environments would reveal how much the benchmark's residential worlds overstate readiness.
  • Since CLIP labels are zero-shot, the same pipeline could be pointed at other semantic targets such as gate numbers or loading docks without retraining; the paper demonstrates only house-number recognition.
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.

Referee Report

4 major / 5 minor

Summary. The paper introduces OpenBench, a simulated and real-world benchmark for last-mile delivery in residential areas, and OPEN, a baseline system that combines OpenStreetMap (OSM) with LLM-based task planning, VLM-based global localization and map updating, and classical planning/localization. The benchmark defines new long-term metrics (LSR and LSPL) in addition to SR and SPL. In simulation, OPEN is reported to outperform the ViNT and NoMaD baselines on SR/SPL, and ablations show that map updates and OSM's compact representation yield efficiency and storage benefits. A real-world two-building delivery task is reported as successful, with performance said to be close to a human teleoperation baseline. The authors state that the code and benchmark are publicly available.

Significance. If the claims hold, the paper makes a useful contribution: a publicly available benchmark targeting an underexplored outdoor last-mile delivery setting, a lightweight OSM-based map representation that avoids pre-mapping, and a concrete integration of LLMs, VLMs, and classical navigation algorithms. The explicit release of code and benchmark data is a strength, and the use of GPS-free global localization as a central design goal is timely. However, the current evidence is not yet sufficient to certify the central GPS-free semantic navigation claim: the VLM global-localization module is never evaluated in isolation, navigation results are single runs without error bars, and the metric equations contain a normalization error that contradicts the reported table values.

major comments (4)
  1. [§III-C, Eqs. (2)-(3)] The definitions of LSR and LSPL include an extra factor 1/N in front of a weighted average whose weights already sum to one. As written, the maximum possible value of LSR is 1/N, not 1, which is inconsistent with Table I where LSR is reported as 100% in the small environment and 75.58% in the medium environment. Please correct the normalization and verify all LSR/LSPL entries, since the current formulas cannot produce the reported numbers.
  2. [§V-B, Table I] All navigation results are reported from what appears to be a single run per environment, with no seeds, repetitions, or error bars. The SR/SPL/LSR/LSPL differences between OPEN and the baselines therefore rest on point estimates without any measure of variance. Please add multiple trials with different random seeds and report mean and standard deviation (or per-trial values), and state the exact number of runs used for each entry in Table I.
  3. [§V-A.3 and §V-C] The success threshold is defined as reaching within 10 m of the destination, which is quite loose for a last-mile delivery task whose goal is the customer's front door, and the paper does not report the distribution of final distances to the destination. In the real-world experiment, Section V-A.2 states that GPS recorded the robot's actual trajectory, yet Section V-C does not compare the OPEN trajectory against this ground truth and does not report OPEN's SPL, even though it states that OPEN's performance is 'closely aligned' with the human-operated control's SPL of 96.1%. Please report the OPEN SPL, the final positioning error, and the distribution of success distances so that the headline success rates can be interpreted.
  4. [§IV-D and §IV-E] The GPS-free claim rests on the VLM global-localization pipeline (MobileSAM segmentation, CLIP labeling, BEV projection, and 2D registration against OSM geometry), but no experiment isolates this module to report its positional error, failure rate, or sensitivity to label noise, repeated building footprints, or occlusion. Section IV-E's assertion that the factor graph 'ensures bounded localization error over time, even without GPS' is stated without proof or empirical support. Because every downstream metric in Table I depends on this module, the central contribution is currently unsupported at its load-bearing point; an instrumented evaluation of the VLM localization module is needed.
minor comments (5)
  1. [§V-B.2] The text says 'For LSP and LSPL' but LSP is never defined; this appears to be a typo for LSR, and should be corrected.
  2. [§III-C] After Eq. (4), the paper should state explicitly that the weights ci sum to one and define n consistently with the number of sequential tasks, since the notation is otherwise ambiguous.
  3. [§V-A.1] The heading 'Sucess Rate of Task Planning' contains a typo and should read 'Success Rate of Task Planning'.
  4. [§V-B.2 and §II] The comparison against ViNT and NoMaD is presented as the only available option because 'no open-source methods align with our objectives'; the limitations of this comparison should be stated explicitly, since these baselines are not designed for multi-task semantic delivery and their failure modes are not analyzed in detail.
  5. [§IV-C and §IV-D] The implementation details of the VLM pipeline are underspecified: the specific CLIP model, the prompt templates, the number of label candidates, and the 2D registration algorithm are not described with enough precision for reproduction; please add these details.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the OPEN system is an integration of external components evaluated against external baselines; no predicted quantity reduces by construction to a fitted input or self-citation.

full rationale

The paper's derivation chain is not circular. OPEN combines off-the-shelf OSM data, LLM/VLM services, FastLIO2 odometry, factor-graph smoothing, and classical A*/TEB planning. The central success metrics (SR, SPL, and the proposed LSR/LSPL) are defined independently of OPEN's internal parameters; the decay rate r=0.9 is a hand-chosen evaluation setting, not fitted to make OPEN's results come out favorable. Table I compares OPEN against external baselines NoMaD and ViNT, and the real-world trial is benchmarked against a human teleoperation trajectory, giving independent reference points. The GPS-free claim rests on the VLM global-localization module (Section IV-D), and the absence of direct localization-accuracy or failure-rate experiments is a missing-evidence concern, not a circular reduction: the module's output is not defined in terms of the navigation metrics, nor are the metrics derived from the module. The only potentially self-referential citation is the use of MobileSAM [24], which includes a co-author, but this is a standard citation of a public off-the-shelf segmentation model and is not load-bearing in a way that defines OPEN's results. No equation in the paper reduces to its own input, and no fitted parameter is renamed as a prediction.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central claim rests on domain assumptions about OSM completeness and VLM reliability rather than on a derivation. The only free numeric choices in the benchmark are the decay rate r=0.9 and the 10 m success threshold, both set by hand. No invented physical entities are introduced.

free parameters (2)
  • decay rate r in LSR and LSPL = 0.9
    Set by hand in Section V-A3. It controls the exponential weighting of earlier tasks in Eqs. 2 to 4 and changes the reported long-term scores with no sensitivity analysis.
  • success distance threshold = 10 m
    Section V-A3 defines successful navigation as reaching within 10 m of the destination. This loose threshold determines all SR, SPL, LSR, and LSPL values.
assumptions (4)
  • domain assumption OpenStreetMap provides a sufficiently complete and geometrically aligned representation of the operating area for routing and global localization.
    Used in Sections IV-C and IV-D. If residential OSM lacks building or unit data, the system must fall back to exploration, and localization registration assumes OSM geometry matches the vision data.
  • domain assumption CLIP and MobileSAM zero-shot labels can recognize buildings, roads, and house-number plates in unseen environments.
    Section IV-D relies on VLM segmentation and CLIP embedding matching to assign semantic labels and perform map updates. No per-label accuracy is reported.
  • standard math The geometric normalization in Eq. 4 is the intended weighting for LSR and LSPL.
    Eq. 4 is a standard normalized geometric series, but it makes the extra 1/N in Eqs. 2 and 3 algebraically inconsistent with the values in Table I.
  • domain assumption A single continuous run in each environment and one real-world route is enough to support the stated conclusions.
    Section V presents no repeated trials or variance estimates, yet the conclusions generalize to various residential settings.

how reviews work

0 comments
Cite this review

Pith. "Pith review of OpenBench: A New Benchmark and Baseline for Semantic Navigation in Smart Logistics." pith.science (2026). https://pith.science/paper/DEJ6C5NW

@misc{pith2026250209238,
  author       = {Pith},
  title        = {Pith review of: OpenBench: A New Benchmark and Baseline for Semantic Navigation in Smart Logistics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DEJ6C5NW}},
  note         = {Machine review of arXiv:2502.09238}
}
read the original abstract

The increasing demand for efficient last-mile delivery in smart logistics underscores the role of autonomous robots in enhancing operational efficiency and reducing costs. Traditional navigation methods, which depend on high-precision maps, are resource-intensive, while learning-based approaches often struggle with generalization in real-world scenarios. To address these challenges, this work proposes the Openstreetmap-enhanced oPen-air sEmantic Navigation (OPEN) system that combines foundation models with classic algorithms for scalable outdoor navigation. The system uses off-the-shelf OpenStreetMap (OSM) for flexible map representation, thereby eliminating the need for extensive pre-mapping efforts. It also employs Large Language Models (LLMs) to comprehend delivery instructions and Vision-Language Models (VLMs) for global localization, map updates, and house number recognition. To compensate the limitations of existing benchmarks that are inadequate for assessing last-mile delivery, this work introduces a new benchmark specifically designed for outdoor navigation in residential areas, reflecting the real-world challenges faced by autonomous delivery systems. Extensive experiments in simulated and real-world environments demonstrate the proposed system's efficacy in enhancing navigation efficiency and reliability. To facilitate further research, our code and benchmark are publicly available.

Figures

Figures reproduced from arXiv: 2502.09238 by the authors.

Figure 1
Figure 1. Overview of the proposed benchmark framework. The diagram [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Simulation environment for last-mile delivery. [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Overview of the OPEN system for autonomous last-mile delivery. The system initiates with a natural language delivery request, processed by [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: The robot used in real-world navigation experiments. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Illustration of the real-world experiment. The top-left part presents [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. LogisticsVLN: Vision-Language Navigation For Low-Altitude Terminal Delivery Based on Agentic UAVs

    cs.RO 2025-05 conditional novelty 5.0 of 10

    An off-the-shelf MLLM-based UAV system achieves 54.7% success on a new 300-task simulated window-level delivery benchmark.

Reference graph

Works this paper leans on

28 extracted references · 25 canonical work pages · cited by 1 Pith paper

  1. [1]

    Artificial intelligence in smart logistics cyber-physical systems: State-of-the-arts and potential applications,

    Y . Liu, X. Tao, X. Li, A. W. Colombo, and S. Hu, “Artificial intelligence in smart logistics cyber-physical systems: State-of-the-arts and potential applications,” IEEE Transactions on Industrial Cyber- Physical Systems, vol. 1, p. 1–20, 2023

    work page 2023

  2. [2]

    In-car positioning and navigation tech- nologies—a survey,

    I. Skog and P. Handel, “In-car positioning and navigation tech- nologies—a survey,” IEEE Transactions on Intelligent Transportation Systems, vol. 10, no. 1, p. 4–21, Mar. 2009

    work page 2009

  3. [3]

    GNM: A General Navigation Model to Drive Any Robot,

    D. Shah, A. Sridhar, A. Bhorkar, N. Hirose, and S. Levine, “GNM: A General Navigation Model to Drive Any Robot,” in International Conference on Robotics and Automation (ICRA) , 2023

    work page 2023

  4. [4]

    ViNT: A foundation model for visual navigation,

    D. Shah, A. Sridhar, N. Dashora, K. Stachowicz, K. Black, N. Hirose, and S. Levine, “ViNT: A foundation model for visual navigation,” in 7th Annual Conference on Robot Learning , 2023

    work page 2023

  5. [5]

    NoMaD: Goal Masked Diffusion Policies for Navigation and Exploration,

    A. Sridhar, D. Shah, C. Glossop, and S. Levine, “NoMaD: Goal Masked Diffusion Policies for Navigation and Exploration,” arXiv pre- print, 2023

    work page 2023

  6. [6]

    Openstreetmap: Challenges and opportunities in machine learning and remote sensing,

    J. E. Vargas-Munoz, S. Srivastava, D. Tuia, and A. X. Falcao, “Openstreetmap: Challenges and opportunities in machine learning and remote sensing,” IEEE Geoscience and Remote Sensing Magazine, vol. 9, no. 1, p. 184–199, Mar. 2021

    work page 2021

  7. [7]

    SOON: Scenario oriented object navigation with graph-based exploration,

    F. Zhu, X. Liang, Y . Zhu, Q. Yu, X. Chang, and X. Liang, “SOON: Scenario oriented object navigation with graph-based exploration,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2021, pp. 12 689–12 699

    work page 2021

  8. [8]

    Cows on pasture: Baselines and benchmarks for language-driven zero-shot object navigation,

    S. Y . Gadre, M. Wortsman, G. Ilharco, L. Schmidt, and S. Song, “Cows on pasture: Baselines and benchmarks for language-driven zero-shot object navigation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2023, pp. 23 171–23 181

    2023

Show all 28 references
  1. [9]

    Interactive gibson benchmark: A benchmark for interactive navigation in cluttered environments,

    F. Xia, W. B. Shen, C. Li, P. Kasimbeg, M. E. Tchapmi, A. Toshev, R. Martin-Martin, and S. Savarese, “Interactive gibson benchmark: A benchmark for interactive navigation in cluttered environments,” IEEE Robotics and Automation Letters, vol. 5, no. 2, p. 713–720, Apr. 2020

    2020

  2. [10]

    Past, present, and future of simultaneous localization and mapping: Toward the robust-perception age,

    C. Cadena, L. Carlone, H. Carrillo, Y . Latif, D. Scaramuzza, J. Neira, I. Reid, and J. J. Leonard, “Past, present, and future of simultaneous localization and mapping: Toward the robust-perception age,” IEEE Transactions on Robotics , vol. 32, no. 6, p. 1309–1332, Dec. 2016

    2016

  3. [11]

    Path planning for autonomous mobile robots: A review,

    J. R. S ´anchez-Ib´a˜nez, C. J. P ´erez-del Pulgar, and A. Garc ´ıa-Cerezo, “Path planning for autonomous mobile robots: A review,” Sensors, vol. 21, no. 23, p. 7898, Nov. 2021

    2021

  4. [12]

    Mobile robot control and navigation: A global overview,

    S. G. Tzafestas, “Mobile robot control and navigation: A global overview,” Journal of Intelligent & Robotic Systems , vol. 91, no. 1, p. 35–58, Mar. 2018

    2018

  5. [13]

    Exploration in deep reinforcement learning: From single- agent to multiagent domain,

    J. Hao, T. Yang, H. Tang, C. Bai, J. Liu, Z. Meng, P. Liu, and Z. Wang, “Exploration in deep reinforcement learning: From single- agent to multiagent domain,” IEEE Transactions on Neural Networks and Learning Systems , vol. 35, no. 7, p. 8762–8782, Jul. 2024

    2024

  6. [14]

    Crowd- steer: Realtime smooth and collision-free robot navigation in densely crowded scenarios trained using high-fidelity simulation,

    J. Liang, U. Patel, A. J. Sathyamoorthy, and D. Manocha, “Crowd- steer: Realtime smooth and collision-free robot navigation in densely crowded scenarios trained using high-fidelity simulation,” in Proceed- ings of the Twenty-Ninth International Joint Conference on Artificial I...

    2020

  7. [15]

    MTG: Mapless trajectory generator with traversabil- ity coverage for outdoor navigation,

    J. Liang, P. Gao, X. Xiao, A. J. Sathyamoorthy, M. Elnoor, M. C. Lin, and D. Manocha, “MTG: Mapless trajectory generator with traversabil- ity coverage for outdoor navigation,” in 2024 IEEE International Conference on Robotics and Automation (ICRA) , vol. 28. IEEE, May 2024, p...

    2024

  8. [16]

    TGS: Trajectory generation and selection using vision language models in mapless outdoor environments,

    D. Song, J. Liang, X. Xiao, and D. Manocha, “TGS: Trajectory generation and selection using vision language models in mapless outdoor environments,” 2024

    2024

  9. [17]

    A survey on vision- language-action models for embodied ai,

    Y . Ma, Z. Song, Y . Zhuang, J. Hao, and I. King, “A survey on vision- language-action models for embodied ai,” 2024

    2024

  10. [18]

    A survey on large language model based autonomous agents,

    L. Wang, C. Ma, X. Feng, Z. Zhang, H. Yang, J. Zhang, Z. Chen, J. Tang, X. Chen, Y . Lin, W. X. Zhao, Z. Wei, and J. Wen, “A survey on large language model based autonomous agents,” Frontiers of Computer Science , vol. 18, no. 6, Mar. 2024

    2024

  11. [19]

    Learning transferable visual models from natural language supervi- sion,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark, G. Krueger, and I. Sutskever, “Learning transferable visual models from natural language supervi- sion,” in Proceedings of the 38th International Conference on Machine...

    2021

  12. [20]

    Visual language maps for robot navigation,

    C. Huang, O. Mees, A. Zeng, and W. Burgard, “Visual language maps for robot navigation,” in 2023 IEEE International Conference on Robotics and Automation (ICRA) . IEEE, May 2023

    2023

  13. [21]

    VLFM: Vision-language frontier maps for zero-shot semantic navigation,

    N. Yokoyama, S. Ha, D. Batra, J. Wang, and B. Bucher, “VLFM: Vision-language frontier maps for zero-shot semantic navigation,” in 2024 IEEE International Conference on Robotics and Automation (ICRA), vol. 35. IEEE, May 2024, p. 42–48

    2024

  14. [22]

    Lm-nav: Robotic navigation with large pre-trained models of language, vision, and action,

    D. Shah, B. Osi ´nski, S. Levine et al. , “Lm-nav: Robotic navigation with large pre-trained models of language, vision, and action,” in Conference on robot learning . PMLR, 2023, pp. 492–504

    2023

  15. [23]

    On evaluation of embodied navigation agents,

    P. Anderson, A. Chang, D. S. Chaplot, A. Dosovitskiy, S. Gupta, V . Koltun, J. Kosecka, J. Malik, R. Mottaghi, M. Savva, and A. R. Zamir, “On evaluation of embodied navigation agents,” 2018

    2018

  16. [24]

    Faster segment anything: Towards lightweight sam for mobile applications,

    C. Zhang, D. Han, Y . Qiao, J. U. Kim, S.-H. Bae, S. Lee, and C. S. Hong, “Faster segment anything: Towards lightweight sam for mobile applications,” 2023

    2023

  17. [25]

    Real-time routing with openstreetmap data,

    D. Luxen and C. Vetter, “Real-time routing with openstreetmap data,” in Proceedings of the 19th ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems , ser. GIS ’11. New York, NY , USA: ACM, 2011, pp. 513–516

    2011

  18. [26]

    Fast-lio: A fast, robust lidar-inertial odometry package by tightly-coupled iterated kalman filter,

    W. Xu and F. Zhang, “Fast-lio: A fast, robust lidar-inertial odometry package by tightly-coupled iterated kalman filter,” IEEE Robotics and Automation Letters, vol. 6, no. 2, p. 3317–3324, Apr. 2021

    2021

  19. [27]

    iSAM2: Incremental smoothing and mapping with fluid relinearization and incremental variable reordering,

    M. Kaess, H. Johannsson, R. Roberts, V . Ila, J. Leonard, and F. Dellaert, “iSAM2: Incremental smoothing and mapping with fluid relinearization and incremental variable reordering,” in 2011 IEEE International Conference on Robotics and Automation . IEEE, May 2011

    2011

  20. [28]

    Integrated online trajec- tory planning and optimization in distinctive topologies,

    C. R ¨osmann, F. Hoffmann, and T. Bertram, “Integrated online trajec- tory planning and optimization in distinctive topologies,” Robotics and Autonomous Systems, vol. 88, p. 142–153, Feb. 2017

    2017

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.