Pith. sign in

REVIEW 2 major objections 5 minor 282 references

VLN methods that look strong in simulation often fail on real robots; hierarchical stacks hold up better than pure RGB end-to-end ones under the tested setups.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-14 15:39 UTC pith:GBH3EJZG

load-bearing objection Solid survey plus a carefully caveated 10-scene robot study that quantifies a large sim-to-real gap; the architectural comparison is system-level, not pure, and the authors say so. the 2 major comments →

arxiv 2607.09792 v1 pith:GBH3EJZG submitted 2026-07-09 cs.RO cs.AI

A Comprehensive Survey and Systematic Real-World Evaluation of Embodied Vision-and-Language Navigation

Liuyi Wang , Kai Sheng , Zongtao He , Jinlong Li , Yongrui Qin , Haojie Dai , Xiangyi Wang , Jingwei Yang
show 3 more authors
Qingqing Yan Chengju Liu Qijun Chen
This is my paper
classification cs.RO cs.AI
keywords vision-and-language navigationembodied AIsim-to-real gaphierarchical navigationmonolithic policiesreal-world robot evaluationcollision avoidancesemantic stopping
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Vision-and-language navigation asks a robot to follow free-form spoken or written instructions using only what it sees, without a pre-built map. Most published numbers come from clean simulators, where success rates can exceed 85 percent. This paper organizes the whole field into hierarchical versus monolithic action designs and discriminative versus generative model designs, then puts two representative systems on a physical wheeled robot in ten real scenes. A monolithic RGB-only method that scores 61 percent success in simulation falls to 22 percent in the real world, while a hierarchical system that plans waypoints and uses LiDAR/SLAM reaches 51 percent success and far fewer collisions. The gap is not just a number: real cameras introduce blur and lighting change, robots slip and collide, and agents often cannot decide when an instruction has been satisfied. The authors treat these results as evidence that current VLN research must close perception, stopping, and safety gaps before the technology is ready for open environments.

Core claim

Under the tested configurations, current VLN systems suffer a large simulation-to-real gap: a representative monolithic RGB-only method drops from 61 percent success in simulation to 22 percent on a physical robot across ten diverse scenes, while a hierarchical waypoint-based system reaches 51 percent real-world success with a markedly lower collision rate, indicating greater robustness for the hierarchical stack in this evaluation setting.

What carries the argument

A two-by-two taxonomy of VLN methods (hierarchical vs. monolithic action paradigms crossed with discriminative vs. generative model paradigms) used both to structure the literature review and to select the two representative system configurations that are then run head-to-head on a real wheeled robot.

Load-bearing premise

That the two chosen systems, the fixed three-meter success rule, the single random seed, and the ten chosen scenes are enough to speak for architectural differences rather than just the specific sensors, robot, and rooms used.

What would settle it

Rerun the same two system configurations on a different physical robot and a new set of real scenes that still follow the paper’s instruction mix; if the hierarchical system no longer shows substantially higher success and lower collisions than the monolithic RGB-only system, the claimed robustness difference does not generalize.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 5 minor

Summary. This manuscript surveys embodied Vision-and-Language Navigation (VLN) and organizes methods along two orthogonal axes: action paradigms (hierarchical waypoint-based vs. monolithic action-based) and model paradigms (discriminative vs. generative). It reviews problem formulations, datasets, simulators, and metrics, then analyzes strengths and limitations of each paradigm. The distinctive contribution is a systematic real-world evaluation on a wheeled robot across ten diverse scenes (200 episodes total): under the tested configurations, a representative monolithic RGB-only method (JanusVLN) drops from 61% SR in simulation to 22% in the real world, while a hierarchical system (CLASH with panoramic sensing and LiDAR/SLAM) achieves 51% real-world SR and substantially lower collision rate (7% vs. 51%). The authors introduce stricter metrics (SSR, CR), analyze intention/backtracking instructions, and discuss sim-to-real failure modes and future directions.

Significance. If the reported measurements hold, the paper supplies a timely dual contribution: a structured taxonomy of a rapidly diversifying VLN literature, and rare physical-robot evidence of a large sim-to-real gap under controlled, multi-scene conditions. The careful scoping of claims to system configurations (rather than pure architecture isolation), the addition of SSR and collision rate, and the qualitative collision analysis are concrete strengths that the community can use for deployment-oriented research. The work is valuable as a reference and as an empirical baseline for safer, more transferable VLN systems.

major comments (2)
  1. [Abstract; Sec. I; Sec. V-B] Sec. V-A/V-B and abstract: The central comparison confounds action paradigm with sensing and control stack (panoramic + LiDAR/SLAM hierarchical vs. monocular RGB-only monolithic). The manuscript already caveats this in Sec. V-B, but the abstract and Sec. I still lead with hierarchical vs. monolithic labels. Please rephrase the abstract and contribution bullets so that system configuration (sensors, mapping, controller) is primary and paradigm labels secondary, so readers cannot misread the 51% vs. 22% result as a pure architecture effect.
  2. [Sec. V-A; Sec. V-B; Figs. 15–17] Sec. V-A.3–V-B: The real-world protocol uses one trial per episode with a fixed seed (42) and reports point estimates only (e.g., SR 22%/51%, CR 51%/7%). For a claim of a substantial sim-to-real gap across ten scenes, please add uncertainty quantification (bootstrap or binomial CIs per metric, and/or per-scene variance) and state whether any episodes were re-run after collisions or communication failures. This is load-bearing for the quantitative headline numbers.
minor comments (5)
  1. [Fig. 3; Index Terms; Sec. I] Fig. 3 and keywords contain typos: “Anlysis,” “Strenghs,” “evluation” (Sec. I), “Moultimodal.” Please proofread section titles and the keyword list.
  2. [Table III; Sec. IV-C] Table III and related text: GPU-hour and training-data summaries for monolithic foundation models are useful; a short parallel table or paragraph for hierarchical generative systems (e.g., CLASH, InternVLA-N1) would improve balance.
  3. [Sec. II-C; Sec. V-A.4] Sec. II-C metrics: nDTW/sDTW and RGS/RGSPL are defined carefully; briefly note which of these (if any) were computable in the real-world study and why SR/SSR/CR were preferred.
  4. [Fig. 12] Fig. 12 scene labels mix Chinese and English in the source layout description; ensure the published figure uses consistent English labels matching the caption (S1–S10).
  5. [Sec. VI-B; Fig. 20] Sec. VI-B future directions are broad; one sentence each tying lifelong memory, counterfactual world models, and safety metrics back to the specific failure modes in Fig. 20 would tighten the link from evaluation to roadmap.

Circularity Check

0 steps flagged

No significant circularity: taxonomy is descriptive and the load-bearing sim-to-real numbers are new physical measurements, not rearrangements of fitted inputs or self-defined quantities.

full rationale

This is a survey-plus-empirical-evaluation paper, not a first-principles derivation. The methodological taxonomy (hierarchical vs. monolithic; discriminative vs. generative) is an organizational classification of prior work and does not claim to derive performance from axioms. The central quantitative claim—monolithic RGB-only SR falling from 61% in simulation to 22% in real-world deployment, versus hierarchical real-world SR of 51% and much lower collision rate—is obtained by running two representative systems on a physical wheeled platform across ten scenes and 200 episodes, then computing standard navigation metrics (SR, SSR, OSR, CR). Those rates are external measurements against ground-truth goals and human semantic annotations, not quantities defined in terms of the paper’s own parameters. Self-citations (e.g., CLASH as the hierarchical baseline, and other author-related methods in the survey) are normal for a survey and for choosing a SoTA representative; they do not force the real-world numbers by construction, and the authors explicitly caveat sensing/mapping/control confounds rather than smuggling uniqueness or ansatz results. No fitted parameter is renamed as a prediction, no uniqueness theorem is imported to forbid alternatives, and no equation reduces to its own input. Score 0 is therefore the correct, proportionate finding.

Axiom & Free-Parameter Ledger

3 free parameters · 3 axioms · 0 invented entities

As a survey-plus-evaluation paper the central claims rest on standard POMDP formalization of VLN, conventional success thresholds, and the representativeness of two chosen systems and ten scenes. Few free parameters are fitted; the main modeling choices are experimental design decisions rather than invented physical entities.

free parameters (3)
  • success distance threshold d_th = 3 m
    Fixed at the conventional 3 m used by R2R/VLN-CE; directly determines SR/SSR counts.
  • instruction mix ratios = 70/20/10
    70% step-by-step / 20% intention / 10% backtracking chosen by the authors for the real-world suite.
  • camera height and robot footprint = 1.5 m
    Default 1.5 m tripod height and differential-drive chassis geometry affect both perception and collision statistics.
axioms (3)
  • domain assumption VLN can be formalized as a POMDP with observation function over egocentric RGB(-D) and language instruction I.
    Stated in Sec. II-B; standard in the field since Anderson et al. 2018.
  • ad hoc to paper The two selected systems (CLASH hierarchical, JanusVLN monolithic) are sufficiently representative of their respective paradigms for comparative claims under the tested stacks.
    Sec. V-A; authors note the entanglement of sensing and architecture.
  • domain assumption A 3 m Euclidean threshold plus human semantic judgment defines success in real scenes.
    Sec. V-A metrics; inherits the community convention while adding SSR.

pith-pipeline@v1.1.0-grok45 · 52892 in / 2603 out tokens · 26749 ms · 2026-07-14T15:39:01.500689+00:00 · methodology

0 comments
read the original abstract

Navigation is a fundamental capability of autonomous systems, yet most existing approaches rely on highly structured models and strong prior assumptions, limiting their robustness in open and uncertain real-world environments. Vision-and-Language Navigation (VLN) offers a promising direction by enabling robots to integrate natural language understanding with visual perception in a data-driven manner. Although VLN has attracted increasing research attention, systematic methodological taxonomy and real-world validation remain limited. This survey presents a comprehensive review of VLN research. Specifically, state-of-the-art methods are organized along two orthogonal dimensions: action paradigms, including hierarchical and monolithic frameworks, and model paradigms, including discriminative and generative approaches. A critical analysis of their respective strengths and limitations is provided. Additionally, we conduct a systematic real-world evaluation of representative VLN system configurations on a physical robotic platform. Experiments across ten diverse real-world scenes show a substantial performance gap between simulation and real-world deployment under the tested configurations: a representative monolithic RGB-only method achieves 61% success in simulation but drops to 22% in real-world deployment, while a hierarchical framework achieves a higher real-world success rate of 51%, suggesting stronger robustness in our evaluation setting. Finally, we highlight key challenges in perception, decision-making, and control that must be addressed in future research.

Figures

Figures reproduced from arXiv: 2607.09792 by Chengju Liu, Haojie Dai, Jingwei Yang, Jinlong Li, Kai Sheng, Liuyi Wang, Qijun Chen, Qingqing Yan, Xiangyi Wang, Yongrui Qin, Zongtao He.

Figure 1
Figure 1. Figure 1: Overview of the vision-and-language navigation (VLN) research. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Trend of vision-and-language navigation (VLN) publications from [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overview of this survey. First, the origins of VLN are traced, followed by the problem definition and paradigm classification. Hierarchical and monolithic [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Timeline of representative VLN methods categorized by action space and model paradigm. The four quadrants correspond to different combinations [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Performance trends of representative VLN methods on the validation unseen splits of R2R [5] and R2R-CE [8], measured by SR and SPL. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Comparison of different VLN action paradigms. [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: The structure of the hierarchical framework section. [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Representative VLN methods under the waypoint-based hierarchical framework. (1) Topological graph-based policies incrementally construct an online [PITH_FULL_IMAGE:figures/full_fig_p009_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Additionally, the representative VLN methods under [PITH_FULL_IMAGE:figures/full_fig_p013_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Representative VLN methods under the monolithic action-based framework. (1) Recurrent policies employ recurrent neural networks to encode [PITH_FULL_IMAGE:figures/full_fig_p014_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: The used wheeled robot for the real-world evaluation. [PITH_FULL_IMAGE:figures/full_fig_p018_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Real-world scenes for evaluation of vision-and-language navigation, including 10 different environments. [PITH_FULL_IMAGE:figures/full_fig_p019_12.png] view at source ↗
Figure 15
Figure 15. Figure 15 [PITH_FULL_IMAGE:figures/full_fig_p019_15.png] view at source ↗
Figure 14
Figure 14. Figure 14: Comparison of the first-person visual observations between simula [PITH_FULL_IMAGE:figures/full_fig_p019_14.png] view at source ↗
Figure 17
Figure 17. Figure 17: Comparison of the performance of the hierarchical and monolithic [PITH_FULL_IMAGE:figures/full_fig_p020_17.png] view at source ↗
Figure 16
Figure 16. Figure 16: Comparison of the performance of the hierarchical and monolithic [PITH_FULL_IMAGE:figures/full_fig_p020_16.png] view at source ↗
Figure 18
Figure 18. Figure 18: Comparison of the model inference time and the average step length [PITH_FULL_IMAGE:figures/full_fig_p021_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Examples of real-world navigation trajectories produced by hierarchical and monolithic VLN frameworks [PITH_FULL_IMAGE:figures/full_fig_p022_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: Examples of collision failure cases in the real-world evaluation. [PITH_FULL_IMAGE:figures/full_fig_p022_20.png] view at source ↗
Figure 22
Figure 22. Figure 22: The long-term evolution trends of VLN research. [PITH_FULL_IMAGE:figures/full_fig_p024_22.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

282 extracted references · 28 linked inside Pith

  1. [1]

    On the representation and estimation of spatial uncertainty,

    R. C. Smith and P. Cheeseman, “On the representation and estimation of spatial uncertainty,”The International Journal of Robotics Research, vol. 5, no. 4, pp. 56–68, 1986. [Online]. Available: https://doi.org/10.1177/027836498600500404

    doi:10.1177/027836498600500404 1986

  2. [2]

    Smith, M

    R. Smith, M. Self, and P. Cheeseman,Estimating Uncertain Spatial Relationships in Robotics. New York, NY: Springer New York, 1990, pp. 167–193. [Online]. Available: https://doi.org/10.1007/ 978-1-4613-8997-2 14

    1990

  3. [3]

    Visual odometry,

    D. Nister, O. Naroditsky, and J. Bergen, “Visual odometry,” inPro- ceedings of the 2004 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2004. CVPR 2004., vol. 1, 2004, pp. I–I

    2004

  4. [4]

    Orb-slam2: An open-source slam system for monocular, stereo, and rgb-d cameras,

    R. Mur-Artal and J. D. Tard ´os, “Orb-slam2: An open-source slam system for monocular, stereo, and rgb-d cameras,”IEEE Transactions on Robotics, vol. 33, no. 5, pp. 1255–1262, 2017

    2017

  5. [5]

    Vision-and-language navigation: Interpret- ing visually-grounded navigation instructions in real environments,

    P. Anderson, Q. Wu,et al., “Vision-and-language navigation: Interpret- ing visually-grounded navigation instructions in real environments,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018, pp. 3674–3683

    2018

  6. [6]

    Stay on the path: Instruction fidelity in vision-and-language navigation,

    V . Jain, G. Magalhaes,et al., “Stay on the path: Instruction fidelity in vision-and-language navigation,” inProceedings of the 57th Annual Meeting of the Association for Computational Linguistics, 2019, pp. 1862–1872

    2019

  7. [7]

    Room-across-room: Multilingual vision- and-language navigation with dense spatiotemporal grounding,

    A. Ku, P. Anderson,et al., “Room-across-room: Multilingual vision- and-language navigation with dense spatiotemporal grounding,” in Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), 2020, pp. 4392–4412

    2020

  8. [8]

    Beyond the nav-graph: Vision-and- language navigation in continuous environments,

    J. Krantz, E. Wijmans,et al., “Beyond the nav-graph: Vision-and- language navigation in continuous environments,” inComputer Vision – ECCV 2020, A. Vedaldi, H. Bischof,et al., Eds. Springer International Publishing, pp. 104–120. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 25

    2020

  9. [9]

    Towards long-horizon vision-language navigation: Platform, benchmark and method,

    X. Song, W. Chen,et al., “Towards long-horizon vision-language navigation: Platform, benchmark and method,” inProceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 12 078–12 088

    2025

  10. [10]

    Habitat 2.0: Training home assistants to rearrange their habitat,

    A. Szot, A. Clegg,et al., “Habitat 2.0: Training home assistants to rearrange their habitat,”Advances in neural information processing systems, vol. 34, pp. 251–266, 2021

    2021

  11. [11]

    Ai2-thor: An interactive 3d environment for visual ai,

    E. Kolve, R. Mottaghi,et al., “Ai2-thor: An interactive 3d environment for visual ai,” inarXiv preprint arXiv:1712.05474, 2017

    Pith/arXiv arXiv 2017

  12. [12]

    Behavior-1k: A human-centered, embodied ai benchmark with 1, 000 everyday activities and realistic simulation,

    C. Li, R. Zhang,et al., “Behavior-1k: A human-centered, embodied ai benchmark with 1, 000 everyday activities and realistic simulation,” CoRR, 2024

    2024

  13. [13]

    Rethinking the embodied gap in vision- and-language navigation: A holistic study of physical and visual disparities,

    L. Wang, X. Xia,et al., “Rethinking the embodied gap in vision- and-language navigation: A holistic study of physical and visual disparities,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), October 2025, pp. 9455–9465

    2025

  14. [14]

    Speaker-follower models for vision-and- language navigation,

    D. Fried, R. Hu,et al., “Speaker-follower models for vision-and- language navigation,”Advances in Neural Information Processing Systems, vol. 31, 2018

    2018

  15. [15]

    Learning to navigate unseen environ- ments: Back translation with environmental dropout,

    H. Tan, L. Yu, and M. Bansal, “Learning to navigate unseen environ- ments: Back translation with environmental dropout,” inProceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), 2019, pp. 2610–2621

    2019

  16. [16]

    Vln bert: A recurrent vision-and-language bert for navigation,

    Y . Hong, Q. Wu,et al., “Vln bert: A recurrent vision-and-language bert for navigation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 1643–1653

    2021

  17. [17]

    Think global, act local: Dual-scale graph transformer for vision-and-language navigation,

    S. Chen, P.-L. Guhur,et al., “Think global, act local: Dual-scale graph transformer for vision-and-language navigation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 16 537–16 547

    2022

  18. [18]

    g3d-lf: Generalizable 3d-language feature fields for embodied tasks,

    Z. Wang and G. H. Lee, “g3d-lf: Generalizable 3d-language feature fields for embodied tasks,” inProceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 14 191–14 202

    2025

  19. [19]

    Vln-r1: Vision-language navigation via reinforcement fine-tuning,

    Z. Qi, Z. Zhang,et al., “Vln-r1: Vision-language navigation via reinforcement fine-tuning,”arXiv preprint arXiv:2506.17221, 2025

    Pith/arXiv arXiv 2025

  20. [20]

    Source-free elastic model adaptation for vision-and-language navigation,

    M. Tan, P. Chen,et al., “Source-free elastic model adaptation for vision-and-language navigation,”IEEE Transactions on Multimedia, vol. 27, pp. 3953–3965, 2025

    2025

  21. [21]

    Navq: Learning a q-model for foresighted vision-and-language navigation,

    P. Xu, X. Gong, and Y . Mu, “Navq: Learning a q-model for foresighted vision-and-language navigation,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), October 2025, pp. 6327–6341

    2025

  22. [22]

    Navid: Video-based vlm plans the next step for vision-and-language navigation,

    J. Zhang, K. Wang,et al., “Navid: Video-based vlm plans the next step for vision-and-language navigation,”Robotics: Science and Systems, 2024

    2024

  23. [23]

    Uni-navid: A video-based vision-language- action model for unifying embodied navigation tasks,

    J. Zhang, K. Wang,et al., “Uni-navid: A video-based vision-language- action model for unifying embodied navigation tasks,”Robotics: Sci- ence and Systems, 2025

    2025

  24. [24]

    Navila: Legged robot vision-language-action model for navigation,

    A.-C. Cheng, Y . Ji,et al., “Navila: Legged robot vision-language-action model for navigation,” inRSS, 2025

    2025

  25. [25]

    Streamvln: Streaming vision-and- language navigation via slowfast context modeling,

    M. Wei, C. Wan,et al., “Streamvln: Streaming vision-and- language navigation via slowfast context modeling,”arXiv preprint arXiv:2507.05240, 2025

    Pith/arXiv arXiv 2025

  26. [26]

    Clash: Collaborative large-small hierarchi- cal framework for continuous vision-and-language navigation,

    L. Wang, Z. He,et al., “Clash: Collaborative large-small hierarchi- cal framework for continuous vision-and-language navigation,”arXiv preprint arXiv:2512.10360, 2025

    arXiv 2025

  27. [27]

    Janusvln: Decoupling semantics and spatiality with dual implicit memory for vision-language navigation,

    S. Zeng, D. Qi,et al., “Janusvln: Decoupling semantics and spatiality with dual implicit memory for vision-language navigation,” inPro- ceedings of the International Conference on Learning Representations (ICLR), 2026

    2026

  28. [28]

    Bootstrapping language-guided navigation learn- ing with self-refining data flywheel,

    Z. Wang, J. Li,et al., “Bootstrapping language-guided navigation learn- ing with self-refining data flywheel,” inThe Thirteenth International Conference on Learning Representations, 2024

    2024

  29. [29]

    Waypoint models for instruction- guided navigation in continuous environments,

    J. Krantz, A. Gokaslan,et al., “Waypoint models for instruction- guided navigation in continuous environments,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 15 162–15 171

    2021

  30. [30]

    Etpnav: Evolving topological planning for vision-language navigation in continuous environments,

    D. An, H. Wang,et al., “Etpnav: Evolving topological planning for vision-language navigation in continuous environments,”IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024

    2024

  31. [31]

    Internvla-n1: An open dual-system vision-language navigation foundation model with learned latent plans,

    I.-N. Team, “Internvla-n1: An open dual-system vision-language navigation foundation model with learned latent plans,” Technical report, Shanghai AI Laboratory, 2025. [Online]. Available: https://internrobotics.github.io/internvla-n1.github.io/static/ pdfs/InternVLA N1.pdf

    2025

  32. [32]

    Multimodal evolutionary encoder for con- tinuous vision-language navigation,

    Z. He, L. Wang,et al., “Multimodal evolutionary encoder for con- tinuous vision-language navigation,” in2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Oct. 2024, pp. 1443–1450

    2024

  33. [33]

    Language-aligned waypoint (law) supervision for vision-and-language navigation in continuous environ- ments,

    S. Raychaudhuri, S. Wani,et al., “Language-aligned waypoint (law) supervision for vision-and-language navigation in continuous environ- ments,” inProceedings of the 2021 conference on empirical methods in natural language processing, 2021, pp. 4018–4028

    2021

  34. [34]

    Navgpt: Explicit reasoning in vision- and-language navigation with large language models,

    G. Zhou, Y . Hong, and Q. Wu, “Navgpt: Explicit reasoning in vision- and-language navigation with large language models,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 7, 2024, pp. 7641–7649

    2024

  35. [35]

    Navcot: Boosting llm-based vision-and-language navigation via learning disentangled reasoning,

    B. Lin, Y . Nie,et al., “Navcot: Boosting llm-based vision-and-language navigation via learning disentangled reasoning,”IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025

    2025

  36. [36]

    Dynam3d: Dynamic layered 3d tokens empower vlm for vision-and-language navigation,

    Z. Wang, S. Lee, and G. H. Lee, “Dynam3d: Dynamic layered 3d tokens empower vlm for vision-and-language navigation,” inAdvances in Neural Information Processing Systems, 2025

    2025

  37. [37]

    History aware multimodal transformer for vision-and-language navigation,

    S. Chen, P.-L. Guhur,et al., “History aware multimodal transformer for vision-and-language navigation,”Advances in Neural Information Processing Systems, vol. 34, 2021

    2021

  38. [38]

    Vision-and-language navigation via causal learning,

    L. Wang, J. Tang,et al., “Vision-and-language navigation via causal learning,”Conference on Computer Vision and Pattern Recognition (CVPR), 2024

    2024

  39. [39]

    Lookahead exploration with neural radiance representation for continuous vision-language navigation,

    Z. Wang, X. Li,et al., “Lookahead exploration with neural radiance representation for continuous vision-language navigation,” inProceed- ings of the IEEE/CVF conference on computer vision and pattern recognition, 2024, pp. 13 753–13 762

    2024

  40. [40]

    Learning depth representation from rgb-d videos by time-aware contrastive pre-training,

    Z. He, L. Wang,et al., “Learning depth representation from rgb-d videos by time-aware contrastive pre-training,”IEEE Transactions on Circuits and Systems for Video Technology, vol. 34, no. 6, pp. 4143– 4158, 2023

    2023

  41. [41]

    Chatgpt: Optimizing language models for dialogue,

    OpenAI, “Chatgpt: Optimizing language models for dialogue,” https: //openai.com/blog/chatgpt, 2022

    2022

  42. [42]

    Qwen2.5-vl technical report,

    S. Bai, K. Chen,et al., “Qwen2.5-vl technical report,”arXiv preprint arXiv:2502.13923, 2025

    Pith/arXiv arXiv 2025

  43. [43]

    Internvl: Scaling up vision foundation models and aligning for generic visual-linguistic tasks,

    Z. Chen, J. Wu,et al., “Internvl: Scaling up vision foundation models and aligning for generic visual-linguistic tasks,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 24 185–24 198

    2024

  44. [44]

    Embodied navigation foundation model,

    J. Zhang, A. Li,et al., “Embodied navigation foundation model,” in Proceedings of the International Conference on Learning Representa- tions (ICLR), 2026

    2026

  45. [45]

    Vision-and-language navigation today and tomorrow: A survey in the era of foundation models,

    Y . Zhang, Z. Ma,et al., “Vision-and-language navigation today and tomorrow: A survey in the era of foundation models,”Transactions on Machine Learning Research, 2024

    2024

  46. [46]

    Vision-language navigation with embodied intelligence: A survey,

    P. Gao, P. Wang,et al., “Vision-language navigation with embodied intelligence: A survey,”arXiv preprint arXiv:2402.14304, 2024

    Pith/arXiv arXiv 2024

  47. [47]

    Advances in embodied navigation using large language models: A survey,

    J. Lin, H. Gao,et al., “Advances in embodied navigation using large language models: A survey,”arXiv preprint arXiv:2311.00530, 2023

    Pith/arXiv arXiv 2023

  48. [48]

    Safety of embodied navigation: a survey,

    Z. Wang, J. Hu, and R. Mu, “Safety of embodied navigation: a survey,” inProceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence, 2025, pp. 10 714–10 722

    2025

  49. [49]

    Aeroverse-review: Comprehensive survey on aerial embodied vision-and-language navigation,

    F. Yao, Y . Liu,et al., “Aeroverse-review: Comprehensive survey on aerial embodied vision-and-language navigation,”The Innovation Informatics, vol. 1, no. 1, pp. 100 015–1, 2025

    2025

  50. [50]

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

    Y . Ma, Z. Song,et al., “A survey on vision–language–action models for embodied ai,”IEEE Transactions on Neural Networks and Learning Systems, 2026

    2026

  51. [51]

    Vision-language-action models for robotics: A review towards real-world applications,

    K. Kawaharazuka, J. Oh,et al., “Vision-language-action models for robotics: A review towards real-world applications,”IEEE Access, vol. 13, pp. 162 467–162 504, 2025

    2025

  52. [52]

    Agentvln: Towards agentic vision-and-language navigation,

    Z. Xin, W. Li,et al., “Agentvln: Towards agentic vision-and-language navigation,”arXiv preprint arXiv:2603.17670, 2026

    arXiv 2026

  53. [53]

    Affordances-oriented planning using foundation models for continuous vision-language navigation,

    J. Chen, B. Lin,et al., “Affordances-oriented planning using foundation models for continuous vision-language navigation,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 22, 2025, pp. 23 568–23 576

    2025

  54. [54]

    Instructnav: Zero-shot system for generic instruction navigation in unexplored environment,

    Y . Long, W. Cai,et al., “Instructnav: Zero-shot system for generic instruction navigation in unexplored environment,” in8th Annual Conference on Robot Learning

  55. [55]

    Mapgpt: Map-guided prompting with adaptive path planning for vision-and-language navigation,

    J. Chen, B. Lin,et al., “Mapgpt: Map-guided prompting with adaptive path planning for vision-and-language navigation,” inProceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 2024, pp. 9796–9810. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 26

    2024

  56. [56]

    Visual language maps for robot navigation,

    C. Huang, O. Mees,et al., “Visual language maps for robot navigation,” inProceedings of the IEEE International Conference on Robotics and Automation (ICRA), London, UK, 2023

    2023

  57. [57]

    Correctnav: Self-correction flywheel empowers vision-language-action navigation model,

    Z. Yu, Y . Long,et al., “Correctnav: Self-correction flywheel empowers vision-language-action navigation model,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 40, no. 22, 2026, pp. 18 737– 18 745

    2026

  58. [58]

    Omninav: A unified framework for prospective exploration and visual-language navigation,

    X. Xue, J. Hu,et al., “Omninav: A unified framework for prospective exploration and visual-language navigation,” inProceedings of the International Conference on Learning Representations (ICLR), 2026

    2026

  59. [59]

    Astranav-world: World model for foresight control and consistency,

    J. Hu, J. Chen,et al., “Astranav-world: World model for foresight control and consistency,”arXiv preprint arXiv:2512.21714, 2025

    Pith/arXiv arXiv 2025

  60. [60]

    Decovln: Decoupling observation, reasoning, and correction for vision-and-language navigation,

    Z. Xin, W. Li,et al., “Decovln: Decoupling observation, reasoning, and correction for vision-and-language navigation,” 2026. [Online]. Available: https://arxiv.org/abs/2603.13133

    arXiv 2026

  61. [61]

    Abot-n0: Technical report on the vla foundation model for versatile embodied navigation,

    Z. Chu, S. Xie,et al., “Abot-n0: Technical report on the vla foundation model for versatile embodied navigation,”arXiv preprint arXiv:2602.11598, 2026

    arXiv 2026

  62. [62]

    Scaling data generation in vision-and-language navigation,

    Z. Wang, J. Li,et al., “Scaling data generation in vision-and-language navigation,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 12 009–12 020

    2023

  63. [63]

    Envedit: Environment editing for vision- and-language navigation,

    J. Li, H. Tan, and M. Bansal, “Envedit: Environment editing for vision- and-language navigation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 15 407–15 417

    2022

  64. [64]

    Bevbert: Multimodal map pre-training for language-guided navigation,

    D. An, Y . Qi,et al., “Bevbert: Multimodal map pre-training for language-guided navigation,”Proceedings of the IEEE/CVF Interna- tional Conference on Computer Vision, 2023

    2023

  65. [65]

    Towards learning a generic agent for vision-and- language navigation via pre-training,

    W. Hao, C. Li,et al., “Towards learning a generic agent for vision-and- language navigation via pre-training,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020, pp. 13 137–13 146

    2020

  66. [66]

    Semantically-aware spatio-temporal reasoning agent for vision-and-language navigation in continuous en- vironments,

    M. Z. Irshad, N. C. Mithun,et al., “Semantically-aware spatio-temporal reasoning agent for vision-and-language navigation in continuous en- vironments,” in2022 26th International conference on pattern recog- nition (ICPR). IEEE, 2022, pp. 4065–4071

    2022

  67. [67]

    Cross-modal map learning for vision and language navigation,

    G. Georgakis, K. Schmeckpeper,et al., “Cross-modal map learning for vision and language navigation,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 15 460–15 470

    2022

  68. [68]

    A multilevel attention network with sub- instructions for continuous vision-and-language navigation,

    Z. He, L. Wang,et al., “A multilevel attention network with sub- instructions for continuous vision-and-language navigation,”Applied Intelligence, vol. 55, no. 7, p. 657, Apr. 2025

    2025

  69. [69]

    Singer: An onboard generalist vision- language navigation policy for drones,

    M. Adang, J. Low,et al., “Singer: An onboard generalist vision- language navigation policy for drones,” inIROS 2025 Workshop: Open World Navigation in Human-centric Environments

    2025

  70. [70]

    Qwen-robotnav technical report: A scalable navigation model designed for an agentic navigation system,

    J. Zhang, G. Zhou,et al., “Qwen-robotnav technical report: A scalable navigation model designed for an agentic navigation system,”arXiv preprint arXiv:2606.18112, 2026

    Pith/arXiv arXiv 2026

  71. [71]

    Vision and navigation for the carnegie- mellon navlab,

    C. Thorpe, M. Hebert,et al., “Vision and navigation for the carnegie- mellon navlab,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 10, no. 3, pp. 362–373, 1988

    1988

  72. [72]

    A visual navigation system for autonomous land vehicles,

    A. Waxman, J. LeMoigne,et al., “A visual navigation system for autonomous land vehicles,”IEEE Journal on Robotics and Automation, vol. 3, no. 2, pp. 124–141, 1987

    1987

  73. [73]

    An experimental system for incremental environment modelling by an autonomous mobile robot,

    P. Moutarlier and R. Chatila, “An experimental system for incremental environment modelling by an autonomous mobile robot,” inThe First International Symposium on Experimental Robotics I, 1989, pp. 327– 346

    1989

  74. [74]

    Mobile robot localization using a single image,

    E. Krotkov, “Mobile robot localization using a single image,” inPro- ceedings, 1989 International Conference on Robotics and Automation, 1989, pp. 978–983 vol.2

    1989

  75. [75]

    Monoslam: Real-time single camera slam,

    A. J. Davison, I. D. Reid,et al., “Monoslam: Real-time single camera slam,”IEEE Transactions on Pattern Analysis and Machine Intelli- gence, vol. 29, no. 6, pp. 1052–1067, 2007

    2007

  76. [76]

    Parallel tracking and mapping for small ar workspaces,

    G. Klein and D. Murray, “Parallel tracking and mapping for small ar workspaces,” in2007 6th IEEE and ACM International Symposium on Mixed and Augmented Reality, 2007, pp. 225–234

    2007

  77. [77]

    Lsd-slam: Large-scale direct monocular slam,

    J. Engel, T. Sch ¨ops, and D. Cremers, “Lsd-slam: Large-scale direct monocular slam,” inComputer Vision – ECCV 2014, D. Fleet, T. Pajdla, et al., Eds. Cham: Springer International Publishing, 2014, pp. 834– 849

    2014

  78. [78]

    Deepvo: Towards end-to-end visual odom- etry with deep recurrent convolutional neural networks,

    S. Wang, R. Clark,et al., “Deepvo: Towards end-to-end visual odom- etry with deep recurrent convolutional neural networks,” in2017 IEEE International Conference on Robotics and Automation (ICRA), 2017, pp. 2043–2050

    2017

  79. [79]

    Droid-slam: Deep visual slam for monocular, stereo, and rgb-d cameras,

    Z. Teed and J. Deng, “Droid-slam: Deep visual slam for monocular, stereo, and rgb-d cameras,” inAdvances in Neural Information Pro- cessing Systems, M. Ranzato, A. Beygelzimer,et al., Eds., vol. 34. Curran Associates, Inc., 2021, pp. 16 558–16 569

    2021

  80. [80]

    Dd-ppo: Learning near-perfect point- goal navigators from 2.5 billion frames,

    E. Wijmans, A. Kadian,et al., “Dd-ppo: Learning near-perfect point- goal navigators from 2.5 billion frames,” inInternational Conference on Learning Representations

Showing first 80 references.