Pith. sign in

REVIEW 4 major objections 4 minor 70 references

SAP-Nav: Spatial Semantic Representation Meets Active Perception for Hierarchical Open-Vocabulary Object Navigation

T0 review · 4 major / 4 minor · reviewed 2026-08-16 · deepseek-v4-flash

Pith's one-line read SAP-Nav claims a zero-shot, fully online navigation system can outperform training-based baselines on hierarchical open-vocabulary object navigation by actively choosing viewpoints and maintaining a queryable spatial-semantic…

desk verdict Solid zero-shot active-perception system for hierarchical OVON, but the headline 12.2-point region gain is mostly VLM scale rather than the architecture; worth reviewing with revisions. read the letter →

arxiv 2608.12707 v1 pith:DX75LDAH submitted 2026-08-13 cs.RO

classification cs.RO
keywords open-vocabularyobjectnavigationhierarchicalgoalzero-shotactiveperceptionspatialsemanticrepresentationvision-languagemodelviewpointselectiononlineroomsegmentation
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

Open-vocabulary object navigation usually asks an agent to find any instance of a category, but people actually give goals that mix scene, room, region, and instance cues, such as "the black glass-top coffee table in the living room with the white fireplace." SAP-Nav claims that these hierarchical goals can be solved fully online with no task-specific training and no precomputed map, by actively deciding where to look rather than passively using whatever the path happens to see. It builds a queryable spatial-semantic representation from room snapshots acquired on demand, and it repositions the robot before trusting a vision-language model's verdict on a candidate object. On the LangMap benchmark the best SAP-Nav configuration reaches 43.8% success on region-level goals, 12.2 points above the strongest training-based baseline, and on HM3D-OVON it reports the top success rate among compared methods.

What carries the argument

The load-bearing machinery is a pair of active-perception loops. QSSR fuses online room segmentation (door projections accumulated over a temporal window) with actively stitched room snapshots and a pretrained scene encoder's room-type logits, forming a BEV map where each cell carries room-type semantics and each room instance keeps a visual record. AVV scores the current view's sufficiency on visibility and perspective, samples navigable viewpoints on concentric rings, and selects the one maximizing the mean ray-cast visibility of the candidate footprint through an online 2.5D height map: $v^* = \arg\max_{v\in\mathcal{V}} \frac{1}{|U(o)|}\sum_{f\in U(o)} \mathrm{vis}(v,f)$. Only then does it ask the VLM to verify category and parsed attribute constraints, blacklisting rejected candidates.

What would settle it

On a fixed set of LangMap region-level episodes, replace the Eq. (6) viewpoint selection with a randomly chosen navigable viewpoint from the same ring samples; if success rate does not drop meaningfully, geometric visibility is not carrying the reported improvement.

Watch

Extended reading notes

Core claim

The paper's central claim is that active perception, applied at two distinct moments, is sufficient to close the gap that hierarchical open-vocabulary navigation opens: grounding room and region constraints needs persistent scene-level evidence, while verifying instance attributes needs a clear viewpoint, and no single passive trajectory reliably supplies both. To test this, SAP-Nav decomposes the problem into QSSR, which constructs an online BEV map whose cells carry room-type semantics and whose room instances carry stitched holistic snapshots, and AVV, which scores current-view sufficiency, selects a navigable and unoccluded viewpoint by ray-casting through an online 2.5D height map, and only then asks the VLM to verify category and parsed attributes. The reported results, best overall success rate on LangMap across scene, room, region, and instance granularities and the best success rate on HM3D-OVON, support the claim that a zero-shot system need not trade away multi-granularity competence for online operation.

Load-bearing premise

The weakest load-bearing premise is that a viewpoint with higher geometric ray-cast visibility of the candidate's footprint in the online 2.5D height map will actually give the vision-language model enough visual detail to verify the object, because the whole AVV gain rests on that correlation.

Editorial extensions

If this is right

  • A fully online robot without a precomputed map can match or beat training-based navigators on goals that specify rooms, regions, and instances, not just object categories.
  • Room-level semantics can be extracted from a few actively captured holistic snapshots, so two rooms of the same type can be told apart during a single episode.
  • Repositioning before verification reduces false stops, and rejected candidates go to a blacklist so exploration resumes without re-querying the same object.
  • The same framework transfers to standard category-level object navigation, since scene-level success on HM3D-OVON improves without task-specific training.

Reading between the lines

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

  • The paper leaves implicit that the geometric visibility score in Eq. (6) could serve as a general "should I look again" trigger for other VLM-based perception tasks, such as instance re-identification or scene-graph verification, without retraining.
  • The per-episode QSSR limitation suggests a persistence extension: retaining room snapshots across tasks would let the same representation support lifelong navigation, with each visit accumulating evidence for later goal searches.
  • A clean stress test of the paper's core premise would randomize viewpoint selection while keeping everything else fixed; if success holds up, the gain would come from repositioning per se rather than from the geometric visibility model.
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 / 4 minor

Summary. The paper presents SAP-Nav, a fully online, zero-shot framework for hierarchical open-vocabulary object navigation (OVON). It combines a Queryable Spatial-Semantic Representation (QSSR), built incrementally from online room segmentation and actively acquired room snapshots, with an Active Viewpoint Verification (AVV) module that repositions the agent when the current observation is judged insufficient for verifying a candidate target. The method is evaluated on the LangMap hierarchical benchmark and the HM3D-OVON standard benchmark, with additional qualitative real-world robot experiments. The headline result is a 43.8% region-level success rate on LangMap with a Qwen3-VL-235B-A22B backbone, reported as a 12.2-point improvement over the training-based PlaNaVid baseline, and a 49.7% success rate on HM3D-OVON with GPT-4o.

Significance. If the results hold, SAP-Nav would demonstrate that a zero-shot, fully online system can handle scene-, room-, region-, and instance-level goals without task-specific training or precomputed maps, outperforming training-based methods on hierarchical OVON. The paper's strengths are its broad benchmark coverage, internal ablations of QSSR and AVV, and a real-world deployment study. The central quantitative claim, however, is currently tied to the choice of a very large VLM; under a controlled same-backbone comparison the advantage over the training-based baseline is small and negative on most granularity levels. The active-perception design is a useful contribution, but the evidence that the framework itself, rather than the VLM scale, drives the headline gains requires a controlled comparison.

major comments (4)
  1. [Abstract and Table 1] The abstract's headline claim of a 12.2% improvement in SR over training-based methods on region-level navigation is confounded by VLM scale. In Table 1, the winning SAP-Nav row uses Qwen3-VL-235B-A22B-thinking and reports 43.8% region SR, while PlaNaVid is reported with Qwen2.5-VL-7B at 31.6%. The same table provides the needed control row: with the same Qwen2.5-VL-7B backbone, SAP-Nav reaches 33.6% region SR, a 2.0-point margin, and it trails PlaNaVid on Single-Goal (29.4 vs 31.4), Scene (28.8 vs 34.4), Room (30.1 vs 35.6), and Instance (25.1 vs 26.2). The 12.2-point claim should therefore be presented as a property of the full configuration with a 235B VLM, or better, supported by re-running PlaNaVid with the same large VLM. Without this controlled comparison, the paper's central claim that SAP-Nav beats training-based methods is not established as a property of the framework.
  2. [Experimental Setup and Tables 1-4] No error bars, variances, or numbers of seeds are reported in Tables 1, 2, 3, or 4. Given the stochastic nature of VLM prompting, object detection, and exploration, differences of a few percentage points—such as the Instance-level 26.8 vs 26.2 in Table 1 or several ablation entries in Table 4—may be within run-to-run noise. The authors should report means and standard deviations over at least three seeds for the main configurations, and ideally for all compared methods.
  3. [Tables 1 and 2] The comparison across baselines is not fully controlled for the perception stack. Only VLFM is described as re-run with its official implementation; for the other baselines it is not stated whether their reported numbers come from the original papers, nor which object detector, segmenter, VLM, and frontier-exploration policy are shared. Since Table 2 includes methods also using GPT-4o (TANGO, MetaNav) that may differ in other components, the reader cannot separate the proposed framework's contribution from the chosen off-the-shelf components. The authors should specify the exact shared components for every baseline and mark each number as re-run or taken from the cited paper.
  4. [Method, Viewpoint Selection, Eq. (6)] The viewpoint-selection module uses a geometric 2.5D height-map visibility score as a proxy for whether a new viewpoint will improve VLM verification, but the paper does not validate this proxy directly. The Limitations section acknowledges that AVV 'selects viewpoints based on geometric visibility, without considering their semantic utility or motion cost.' The Table 4 ablation shows that AVV improves performance, but it does not isolate whether the gain comes from the specific visibility model in Eq. (6) or simply from moving closer to the candidate and enlarging it in the image. An additional ablation that replaces Eq. (6) with a random navigable viewpoint near the candidate would help separate these explanations and support the transferability of AVV beyond these benchmarks.
minor comments (4)
  1. [Real-world Deployment] The real-world deployment section is qualitative: it provides illustrative examples but no instruction-level success counts or quantitative task outcomes. Since the paper states that this 'validate[s] its practical feasibility,' the claim would be stronger with at least a small table of task-level outcomes across a fixed instruction set.
  2. [Eq. (3)] In Eq. (3), the case C_k = {c} is well defined, but the behavior when C_k is empty or when the argmax over a multi-type set is not unique is unspecified. A sentence clarifying these edge cases would prevent ambiguity.
  3. [Abstract and text] The claim of a '12.2% improvement' should specify that it is in absolute percentage points and that it is measured relative to a different VLM configuration, as discussed in the major comments.
  4. [Table 1 caption] The caption says 'Single-Goal reports the average performance over all single-goal episodes,' which is vague about whether the average is over the four granularity levels or over a combined set of episodes. Please clarify the aggregation rule.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the pipeline is evaluated against external benchmarks and ablations are empirical, not derived from the paper's own outputs.

full rationale

SAP-Nav is a modular, training-free system whose equations are constructive definitions, not derivations from the evaluation targets. Equations (1)-(6) define how room snapshots are aggregated, how BEV room-type evidence is accumulated, and how candidate viewpoints are scored; none of these quantities is fitted to LangMap or HM3D-OVON success labels. The claim that QSSR and AVV improve performance is supported by ablations on held-out benchmarks, which is the appropriate empirical evidence rather than a circular reduction. The only author-overlap reference is the LangMap benchmark (Miao et al. 2026), but it is used as a public evaluation suite with multiple external baselines and is not invoked as a premise in any argument, nor does it forbid alternative architectures or viewpoints. The abstract's 12.2% region-level advantage compares SAP-Nav with a 235B VLM against PlaNaVid with a 7B VLM; that is a comparability/fairness limitation, not a circularity, because no parameter is fitted to that comparison and no result is renamed as a prediction. The AVV geometric-visibility heuristic may not always improve VLM verification, but the paper itself acknowledges this in Limitations and Future Work; that is a robustness concern, not a circular step. Overall, the paper's central claims are externally falsifiable and do not reduce by construction to their inputs.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

The central claims rest on a set of behavioral assumptions about the scene encoder, the VLM, and the visibility model, plus several hand-chosen constants. None are provided with values or sensitivity analysis. No new physical or conceptual entities are introduced beyond software modules.

free parameters (7)
  • Viewpoint sufficiency threshold tau = not reported
    Controls when AVV triggers repositioning. Strongly affects the SR/SPL trade-off; no sensitivity analysis or chosen value is given.
  • Temporal window m for door projection accumulation = not reported
    Online room segmentation adapts Bao et al. 2025 by accumulating door projections over m frames; m is never specified.
  • Number of frames Nv per room snapshot = not reported
    When first entering a room, the agent rotates in place to capture Nv frames stitched into a snapshot; Nv is not given.
  • Viewpoint sampling radii = {0.8, 1.2, 1.6, 2.0, 2.4} m
    Concentric ring distances used in Eq. (4) for viewpoint candidates; chosen by hand.
  • Angular sample count Ntheta = 24
    Number of angles per ring in Eq. (4).
  • Indoor room-type vocabulary T = not enumerated
    Static prior restricting the Places365 scene encoder outputs to indoor room types; the membership of T is not specified.
  • Room area growth threshold for expansion update = not reported
    Triggers new snapshots for previously classified rooms; the 'substantial' criterion is never quantified.
assumptions (5)
  • domain assumption The adapted online room segmentation of Bao et al. (2025) produces correct room boundaries from accumulated door projections over a temporal window of m frames.
    QSSR partitions explored free space into room instances Rk; incorrect boundaries would mis-assign room types to candidates.
  • domain assumption The Places365 scene encoder, when restricted to an indoor room-type vocabulary T, yields reliable per-patch room-type logits in unseen HM3D scenes.
    Sub-room refinement (Eq. 1-2) accumulates these logits into the probability field used to assign candidate room types.
  • domain assumption VLM zero-shot judgments of visibility and perspective (1-5 scores) are consistent across viewpoints and correlate with actual verification accuracy.
    The sufficiency threshold tau triggers viewpoint selection; if the scoring is miscalibrated, AVV would reposition unnecessarily or fail to reposition.
  • domain assumption The 2.5D height-map ray-casting visibility model in Eq. (6) accurately predicts which viewpoints will actually provide better verification evidence.
    Viewpoint selection picks v* by maximizing visible footprint cells; if this proxy does not track perceptual sufficiency, the chosen views may not help verification.
  • domain assumption Success is correctly measured by reaching any dataset-annotated target viewpoint within 1 m at 500 steps.
    All SR/SPL claims depend on this benchmark protocol.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SAP-Nav: Spatial Semantic Representation Meets Active Perception for Hierarchical Open-Vocabulary Object Navigation." pith.science (2026). https://pith.science/paper/DX75LDAH

@misc{pith2026260812707,
  author       = {Pith},
  title        = {Pith review of: SAP-Nav: Spatial Semantic Representation Meets Active Perception for Hierarchical Open-Vocabulary Object Navigation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DX75LDAH}},
  note         = {Machine review of arXiv:2608.12707}
}
read the original abstract

Hierarchical open-vocabulary object navigation (OVON) requires agents to follow free-form instructions that may specify targets through scene-, room-, region-, and instance-level cues in unseen environments. Although recent work LangMap has formalized this setting, reliably solving it under partial observations remains challenging: spatial grounding requires persistent environment-level evidence, whereas target verification requires clear and discriminative candidate views. We present SAP-Nav, a fully online, zero-shot framework that addresses both requirements through active perception. SAP-Nav incrementally constructs a Queryable Spatial-Semantic Representation from actively acquired room views, enabling spatial semantic queries from any explored location. It further employs Active Viewpoint Verification to assess whether the current observation provides sufficient evidence and, when necessary, reposition the agent to a more informative viewpoint before verifying candidates against category and attribute constraints. Although designed for hierarchical OVON, SAP-Nav supports both hierarchical and standard category-level OVON without task-specific training or precomputed scene maps. Experiments on LangMap and HM3D-OVON show that SAP-Nav achieves the overall best performance, including a 12.2% improvement in SR over training-based methods on region-level navigation. Real-world robot experiments further demonstrate its practical feasibility. Code will be made publicly available upon acceptance.

Figures

Figures reproduced from arXiv: 2608.12707 by the authors.

Figure 1
Figure 1. SAP-Nav introduces active perception into hierarchical OVON, where goals span scene-, room-, region-, and instance [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. The overview of our SAP-Nav. Given egocentric RGB-D observations, odometry, and an instruction at any granularity, SAP-Nav operates through two active-perception modules. QSSR converts actively acquired online observations into a queryable spatial semantic representation, from room segmentation to sub-room semantics. AVV closes the verification loop by assessing viewpoint sufficiency, actively repositioning to infor… view at source ↗
Figure 3
Figure 3. Real-world deployment. SAP-Nav can execute hierarchical OVON instructions in indoor scenes. Component LangMap HM3D-OVON SE VS TV Single-Goal Scene Room Region Instance Overall SR ↑ SPL ↑ SR SPL SR SPL SR SPL SR SPL SR SPL 24.1 12.2 12.8 7.2 38.3 20.7 40.0 18.9 3.5 1.7 24.5 16.2 ✓ ✓ 29.8 15.7 17.6 10.9 39.8 20.7 40.5 20.4 17.7 9.6 31.1 17.5 ✓ 36.0 19.2 33.1 19.9 44.0 22.7 42.5 22.0 25.0 13.4 46.8 26.2 ✓ ✓ 39.2 17.6 4… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

70 extracted references · 49 canonical work pages

  1. [1]

    Habitat Challenge 2023 , year =

  2. [2]

    arXiv preprint arXiv:1807.06757 , year=

    On evaluation of embodied navigation agents , author=. arXiv preprint arXiv:1807.06757 , year=

  3. [3]

    arXiv preprint arXiv:2305.16925 , year=

    How to not train your dragon: Training-free embodied object goal navigation with semantic frontiers , author=. arXiv preprint arXiv:2305.16925 , year=

  4. [4]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Cows on pasture: Baselines and benchmarks for language-driven zero-shot object navigation , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  5. [5]

    Advances in Neural Information Processing Systems , volume=

    Multion: Benchmarking semantic map memory using multi-object navigation , author=. Advances in Neural Information Processing Systems , volume=

  6. [6]

    Zhang, Jiazhao and Dai, Liu and Meng, Fanpeng and Fan, Qingnan and Chen, Xuelin and Xu, Kai and Wang, He , booktitle=

  7. [7]

    8th Annual Conference on Robot Learning , year=

    InstructNav: Zero-shot System for Generic Instruction Navigation in Unexplored Environment , author=. 8th Annual Conference on Robot Learning , year=

  8. [8]

    Advances in Neural Information Processing Systems , volume=

    Zson: Zero-shot object-goal navigation using multimodal goal embeddings , author=. Advances in Neural Information Processing Systems , volume=

Show all 70 references
  1. [9]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Towards long-horizon vision-language navigation: Platform, benchmark and method , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  2. [10]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Unigoal: Towards universal zero-shot goal-oriented navigation , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  3. [11]

    Proceedings of the 40th International Conference on Machine Learning , pages=

    ESC: exploration with soft commonsense constraints for zero-shot object navigation , author=. Proceedings of the 40th International Conference on Machine Learning , pages=

  4. [12]

    Hong, Yicong and Wu, Qi and Qi, Yuankai and Rodriguez-Opazo, Cristian and Gould, Stephen , booktitle=

  5. [13]

    2019 International Conference on Robotics and Automation (ICRA) , pages=

    Visual representations for semantic target driven navigation , author=. 2019 International Conference on Robotics and Automation (ICRA) , pages=. 2019 , organization=

  6. [14]

    Ramrakhya, Ram and Batra, Dhruv and Wijmans, Erik and Das, Abhishek , booktitle=

  7. [15]

    Wijmans, Erik and Essa, Irfan and Batra, Dhruv , journal=

  8. [16]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Auxiliary tasks and exploration enable objectgoal navigation , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  9. [17]

    8th Annual Conference on Robot Learning , year=

    PoliFormer: Scaling On-Policy RL with Transformers Results in Masterful Navigators , author=. 8th Annual Conference on Robot Learning , year=

  10. [18]

    Advances in Neural Information Processing Systems , volume=

    Object goal navigation using goal-oriented semantic exploration , author=. Advances in Neural Information Processing Systems , volume=

  11. [19]

    2023 IEEE International Conference on Robotics and Automation (ICRA) , pages=

    Frontier Semantic Exploration for Visual Target Navigation , author=. 2023 IEEE International Conference on Robotics and Automation (ICRA) , pages=. 2023 , organization=

  12. [20]

    Conference on Robot Learning , pages=

    Learning hierarchical relationships for object-goal navigation , author=. Conference on Robot Learning , pages=. 2021 , organization=

  13. [21]

    Learning to Map for Active Semantic Goal Navigation , author=

  14. [22]

    Chang, Angel and Dai, Angela and Funkhouser, Thomas and Halber, Maciej and Niessner, Matthias and Savva, Manolis and Song, Shuran and Zeng, Andy and Zhang, Yinda , journal=

  15. [23]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Poni: Potential functions for objectgoal navigation with interaction-free learning , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  16. [24]

    2023 IEEE International Conference on Robotics and Automation (ICRA) , pages=

    Open-vocabulary queryable scene representations for real world planning , author=. 2023 IEEE International Conference on Robotics and Automation (ICRA) , pages=. 2023 , organization=

  17. [25]

    arXiv preprint arXiv:2408.00714 , url=

    SAM 2: Segment Anything in Images and Videos , author=. arXiv preprint arXiv:2408.00714 , url=

  18. [26]

    2024 , organization=

    Yokoyama, Naoki and Ha, Sehoon and Batra, Dhruv and Wang, Jiuguang and Bucher, Bernadette , booktitle=. 2024 , organization=

  19. [27]

    Habitat-matterport

    Ramakrishnan, Santhosh K and Gokaslan, Aaron and Wijmans, Erik and Maksymets, Oleksandr and Clegg, Alex and Turner, John and Undersander, Eric and Galuba, Wojciech and Westbury, Andrew and Chang, Angel X and others , journal=. Habitat-matterport

  20. [28]

    2025 IEEE International Conference on Robotics and Automation (ICRA) , pages=

    Topology-based visual active room segmentation , author=. 2025 IEEE International Conference on Robotics and Automation (ICRA) , pages=. 2025 , organization=

  21. [29]

    arXiv preprint arXiv:2603.06914 , year=

    SysNav: Multi-Level Systematic Cooperation Enables Real-World, Cross-Embodiment Object Navigation , author=. arXiv preprint arXiv:2603.06914 , year=

  22. [30]

    arXiv preprint arXiv:2201.13360 , year=

    Hydra: A real-time spatial perception system for 3D scene graph construction and optimization , author=. arXiv preprint arXiv:2201.13360 , year=

  23. [31]

    Habitat: A platform for embodied

    Savva, Manolis and Kadian, Abhishek and Maksymets, Oleksandr and Zhao, Yili and Wijmans, Erik and Jain, Bhavana and Straub, Julian and Liu, Jia and Koltun, Vladlen and Malik, Jitendra and others , booktitle=. Habitat: A platform for embodied

  24. [32]

    arXiv preprint arXiv:2602.02220 , year=

    LangMap: A Hierarchical Benchmark for Open-Vocabulary Goal Navigation , author=. arXiv preprint arXiv:2602.02220 , year=

  25. [33]

    2024 , organization=

    Yokoyama, Naoki and Ramrakhya, Ram and Das, Abhishek and Batra, Dhruv and Ha, Sehoon , booktitle=. 2024 , organization=

  26. [34]

    Khanna, Mukul and Ramrakhya, Ram and Chhablani, Gunjan and Yenamandra, Sriram and Gervet, Theophile and Chang, Matthew and Kira, Zsolt and Chaplot, Devendra Singh and Batra, Dhruv and Mottaghi, Roozbeh , booktitle=

  27. [35]

    Robotics: Science and Systems , year=

    Uni-NaVid: A Video-based Vision-Language-Action Model for Unifying Embodied Navigation Tasks , author=. Robotics: Science and Systems , year=

  28. [36]

    International Conference on Computer Vision (ICCV) , year =

    Move to Understand a 3D Scene: Bridging Visual Grounding and Exploration for Efficient and Versatile Embodied Navigation , author =. International Conference on Computer Vision (ICCV) , year =

  29. [37]

    Yang, Yuncong and Yang, Han and Zhou, Jiachen and Chen, Peihao and Zhang, Hongxin and Du, Yilun and Gan, Chuang , booktitle=

  30. [38]

    arXiv preprint arXiv:2604.02318 , year=

    Stop Wandering: Efficient Vision-Language Navigation via Metacognitive Reasoning , author=. arXiv preprint arXiv:2604.02318 , year=

  31. [39]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    OVSegDT: Segmenting Transformer for Open-Vocabulary Object Goal Navigation , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  32. [40]

    Wang, Zihan and Lee, Seungjun and Lee, Gim Hee , journal=

  33. [41]

    arXiv preprint arXiv:2509.12129 , year=

    Embodied navigation foundation model , author=. arXiv preprint arXiv:2509.12129 , year=

  34. [42]

    arXiv preprint arXiv:2508.10925 , year=

    gpt-oss-120b & gpt-oss-20b model card , author=. arXiv preprint arXiv:2508.10925 , year=

  35. [43]

    Ziliotto, Filippo and Campari, Tommaso and Serafini, Luciano and Ballan, Lamberto , booktitle=

  36. [44]

    Artificial intelligence , volume=

    The spatial semantic hierarchy , author=. Artificial intelligence , volume=

  37. [45]

    Proceedings of the IEEE , volume=

    Active perception , author=. Proceedings of the IEEE , volume=

  38. [46]

    arXiv preprint arXiv:2511.21631 , year=

    Qwen3-VL Technical Report , author=. arXiv preprint arXiv:2511.21631 , year=

  39. [47]

    Proceedings of the 2026 International Conference on Multimedia Retrieval , pages=

    TAVEN: Task-driven Adaptive Viewpoint Exploration for Training-Free 3D Spatial Reasoning and Understanding , author=. Proceedings of the 2026 International Conference on Multimedia Retrieval , pages=

  40. [48]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Context-Nav: Context-Driven Exploration and Viewpoint-Aware 3D Spatial Reasoning for Instance Navigation , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  41. [49]

    European Conference on Computer Vision , pages=

    Prioritized semantic learning for zero-shot instance navigation , author=. European Conference on Computer Vision , pages=. 2024 , organization=

  42. [50]

    arXiv preprint arXiv:2603.05377 , year=

    OpenFrontier: General Navigation with Visual-Language Grounded Frontiers , author=. arXiv preprint arXiv:2603.05377 , year=

  43. [51]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Touchdown: Natural language navigation and spatial reasoning in visual street environments , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  44. [52]

    International Conference on Learning Representations , year=

    DD-PPO: Learning Near-Perfect PointGoal Navigators from 2.5 Billion Frames , author=. International Conference on Learning Representations , year=

  45. [53]

    2005 , publisher=

    The Cambridge handbook of visuospatial thinking , author=. 2005 , publisher=

  46. [54]

    Hierarchical open-vocabulary

    Werby, Abdelrhman and Huang, Chenguang and B. Hierarchical open-vocabulary. First Workshop on Vision-Language Models for Navigation and Manipulation at ICRA 2024 , year=

  47. [55]

    Spatial-Aware and Viewpoint-Robust Vision-Language Navigation , year=

    Zhong, Zhide and Lu, Jia and Liu, Xiangchen and Yu, Runze and Zheng, Xinhu and Liu, Zhe and Wang, Hesheng and Li, Haoang , journal=. Spatial-Aware and Viewpoint-Robust Vision-Language Navigation , year=

  48. [56]

    Proceedings of the International Conference on Automated Planning and Scheduling , volume=

    Saynav: Grounding large language models for dynamic planning to navigation in new environments , author=. Proceedings of the International Conference on Automated Planning and Scheduling , volume=

  49. [57]

    Yin, Hang and Xu, Xiuwei and Wu, Zhenyu and Zhou, Jie and Lu, Jiwen , journal=

  50. [58]

    International Journal of Computer Vision , volume=

    Active perception for visual-language navigation , author=. International Journal of Computer Vision , volume=. 2023 , publisher=

  51. [59]

    arXiv preprint arXiv:2601.06806 , year=

    SpatialNav: Leveraging Spatial Scene Graphs for Zero-Shot Vision-and-Language Navigation , author=. arXiv preprint arXiv:2601.06806 , year=

  52. [60]

    Active Scene Recognition for Domestic Robots: Observing, Moving, and Recognizing , year=

    Liu, Shaopeng and Huang, Chao and Huang, Hailong and Wu, Jingda , journal=. Active Scene Recognition for Domestic Robots: Observing, Moving, and Recognizing , year=

  53. [61]

    2017 , publisher=

    Zhou, Bolei and Lapedriza, Agata and Khosla, Aditya and Oliva, Aude and Torralba, Antonio , journal=. 2017 , publisher=

  54. [62]

    Yadav, Karmesh and Ramrakhya, Ram and Ramakrishnan, Santhosh Kumar and Gervet, Theo and Turner, John and Gokaslan, Aaron and Maestre, Noah and Chang, Angel Xuan and Batra, Dhruv and Savva, Manolis and others , booktitle=

  55. [63]

    Learning To Explore Using Active Neural SLAM , author=

  56. [64]

    arXiv preprint arXiv:2601.08325 , year=

    ActiveVLA: Injecting Active Perception into Vision-Language-Action Models for Precise 3D Robotic Manipulation , author=. arXiv preprint arXiv:2601.08325 , year=

  57. [65]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Active Perception Meets Rule-Guided RL: A Two-Phase Approach for Precise Object Navigation in Complex Environments , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  58. [66]

    Autonomous Robots , volume=

    Revisiting active perception , author=. Autonomous Robots , volume=. 2018 , publisher=

  59. [67]

    2024 , organization=

    Liu, Shilong and Zeng, Zhaoyang and Ren, Tianhe and Li, Feng and Zhang, Hao and Yang, Jie and Jiang, Qing and Li, Chunyuan and Yang, Jianwei and Su, Hang and others , booktitle=. 2024 , organization=

  60. [68]

    2026 , howpublished =

    Qwen3.5 , author =. 2026 , howpublished =

  61. [69]

    arXiv preprint arXiv:2410.21276 , year =

  62. [70]

    arXiv preprint arXiv:2502.13923 , year=

    Qwen2.5-VL Technical Report , author=. arXiv preprint arXiv:2502.13923 , year=

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.