arxiv: 2603.26788 · v2 · submitted 2026-03-25 · 💻 cs.RO · cs.CV

Recognition: 2 theorem links

· Lean Theorem

ReMemNav: A Rethinking and Memory-Augmented Framework for Zero-Shot Object Navigation

Feng Wu , Wei Zuo , Wenliang Yang , Jun Xiao , Yang Liu , Xinhua Zeng

Authors on Pith no claims yet

Pith reviewed 2026-05-15 01:05 UTC · model grok-4.3

classification 💻 cs.RO cs.CV

keywords zero-shot object navigationvision-language modelsepisodic memoryrethinking mechanismhierarchical navigationsemantic buffer queueHM3D benchmarkMP3D benchmark

0 comments

The pith

ReMemNav combines episodic memory buffers and dual-modal rethinking with vision-language models to cut hallucinations and deadlocks in zero-shot object navigation.

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

The paper presents ReMemNav as a hierarchical framework that augments vision-language models with panoramic semantic priors and episodic memory for locating unseen objects in unfamiliar spaces. It tackles spatial hallucinations and exploration deadlocks through an adaptive rethinking step that checks target visibility against stored history. The system also converts depth masks into feasible actions so the model can choose concrete movements. On standard simulation benchmarks this produces higher success rates and shorter paths than prior training-free baselines.

Core claim

ReMemNav integrates panoramic semantic priors and episodic memory with VLMs by anchoring spatial reasoning via the Recognize Anything Model, applying an adaptive dual-modal rethinking mechanism on an episodic semantic buffer queue to verify target visibility and correct decisions, and deriving feasible actions from depth masks for low-level execution, which raises success rate and SPL on HM3D v0.1, v0.2, and MP3D relative to existing training-free zero-shot methods.

What carries the argument

The adaptive dual-modal rethinking mechanism built around an episodic semantic buffer queue, which stores historical observations and actively verifies whether the target is visible before committing to new actions.

If this is right

Navigation agents reach unseen targets more often across HM3D and MP3D scenes without task-specific training.
Exploration paths become shorter and more efficient as measured by SPL, because deadlocks are broken by memory-based corrections.
High-level semantic instructions from the VLM map more reliably onto low-level spatial moves via depth-derived action sequences.
The same buffer-and-rethink loop could reduce repeated visits to already-seen areas in other indoor navigation settings.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Adding similar memory queues to other VLM-driven robotic tasks might improve long-horizon instruction following in changing scenes.
The framework's reliance on episodic buffers suggests it could be tested on real robots to see how sensor noise affects the rethinking step.
Extending the buffer to store multi-step reasoning traces might allow handling of compound instructions like 'find the chair then the lamp'.

Load-bearing premise

The episodic semantic buffer queue and rethinking step can reliably detect and fix VLM mistakes such as false target detections without introducing fresh errors or needing per-environment adjustments.

What would settle it

An experiment in which targets are placed behind partial occlusions or near visually similar distractors, measuring whether success rate falls below the baseline when the buffer queue fails to override repeated VLM hallucinations.

Figures

Figures reproduced from arXiv: 2603.26788 by Feng Wu, Jun Xiao, Wei Zuo, Wenliang Yang, Xinhua Zeng, Yang Liu.

**Figure 2.** Figure 2: Overview of the ReMemNav framework. At each time step, the agent acquires six-directional RGB-D observations, [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Figure 3: Pipeline of the Episodic Memory Buffer Queue construction based on multi-modal perception. This process illustrates [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

**Figure 4.** Figure 4: VLM-guided process for safe action decision-making. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗

**Figure 5.** Figure 5: Ablation study on episodic memory capacity K. The experiment is conducted using Qwen3-VL-4B on HM3D v0.2. The dual y-axis plot illustrates the inverted U-shape trend for both SR and SPL, peaking at K = 10. framework, we conduct an ablation study across various VLMs on the HM3D v0.2 dataset, as presented in Table III. First, it is noteworthy that our ReMemNav framework achieves a solid baseline performance … view at source ↗

read the original abstract

Zero-shot object navigation requires agents to locate unseen target objects in unfamiliar environments without prior maps or task-specific training which remains a significant challenge. Although recent advancements in vision-language models(VLMs) provide promising commonsense reasoning capabilities for this task, these models still suffer from spatial hallucinations, local exploration deadlocks, and a disconnect between high-level semantic intent and low-level control. In this regard, we propose a novel hierarchical navigation framework named ReMemNav, which seamlessly integrates panoramic semantic priors and episodic memory with VLMs. We introduce the Recognize Anything Model to anchor the spatial reasoning process of the VLM. We also design an adaptive dual-modal rethinking mechanism based on an episodic semantic buffer queue. The proposed mechanism actively verifies target visibility and corrects decisions using historical memory to prevent deadlocks. For low-level action execution, ReMemNav extracts a sequence of feasible actions using depth masks, allowing the VLM to select the optimal action for mapping into actual spatial movement. Extensive evaluations on HM3D and MP3D demonstrate that ReMemNav outperforms existing training-free zero-shot baselines in both success rate and exploration efficiency. Specifically, we achieve significant absolute performance improvements, with SR and SPL increasing by 1.7% and 7.0% on HM3D v0.1, 18.2% and 11.1% on HM3D v0.2, and 8.7% and 7.9% on MP3D.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

ReMemNav combines an episodic buffer with VLM rethinking and depth-mask actions for zero-shot navigation, but the gains rest on unablated claims about the buffer's reliability.

read the letter

The core of this paper is a practical tweak to VLM-driven object navigation: it adds an episodic semantic buffer queue that feeds into a dual-modal rethinking step, plus depth masks for action selection. The authors anchor the VLM with the Recognize Anything Model and claim this setup cuts spatial hallucinations and deadlocks better than prior training-free baselines. On HM3D and MP3D they report SR lifts of 1.7–18.2% and SPL lifts of 7–11%, which is the main empirical hook. That specific stack—panoramic priors, buffer queue for visibility checks, rethinking loop, and mask-based control—is not laid out in the cited work, so the integration itself is the new piece. It does a clean job of turning high-level VLM output into executable steps without extra training. The writing stays focused on the engineering flow rather than overclaiming theory. The soft spot is exactly what the stress-test note flags: no ablation shows how often the rethinking corrects versus adds error, no threshold sensitivity data, and no protocol details on baselines or statistics. The abstract numbers are presented as direct evidence of the mechanism, but without those breakdowns the gains could trace to prompt choices or scene quirks instead. The framework is empirical, not circular, and the citations hit the right prior navigation and VLM papers. Readers working on embodied agents who need training-free methods will find the architecture description useful for replication attempts. It is worth a serious referee because the idea is concrete enough to test and the reported margins are large enough to matter if they hold up. I would send it to review and ask specifically for the rethinking ablation and full experimental setup.

Referee Report

3 major / 2 minor

Summary. The paper proposes ReMemNav, a hierarchical framework for zero-shot object navigation that integrates panoramic semantic priors and episodic memory with VLMs. It anchors VLM spatial reasoning via the Recognize Anything Model, introduces an adaptive dual-modal rethinking mechanism based on an episodic semantic buffer queue to verify target visibility and break deadlocks, and uses depth masks to extract feasible low-level actions. Evaluations on HM3D v0.1/v0.2 and MP3D report absolute SR/SPL gains of 1.7%/7.0%, 18.2%/11.1%, and 8.7%/7.9% over training-free baselines.

Significance. If the gains prove robust and directly attributable to the rethinking mechanism, the work would meaningfully advance training-free ZSON by addressing VLM hallucinations and deadlocks through memory augmentation without task-specific training. The combination of semantic priors, buffer-based correction, and VLM-guided action selection offers a practical engineering path for more reliable navigation in unseen environments.

major comments (3)

[Experiments] Experiments section: The abstract and results claim concrete SR/SPL gains (e.g., +18.2% SR on HM3D v0.2) but supply no experimental protocol details, baseline implementations, statistical tests, run variance, or ablation results. This leaves the attribution of improvements to the episodic buffer queue unverified and load-bearing for the central claim.
[Method] Method section (adaptive dual-modal rethinking): The headline performance is ascribed to the episodic semantic buffer queue verifying target visibility and correcting VLM decisions. No quantitative breakdown is given on correction frequency versus false-positive rate, nor on whether per-environment threshold tuning is required; without this or an ablation removing the queue, the net gains could be artifacts of prompt/scene distribution rather than a robust advance.
[§4.3] §4.3 (low-level control): The depth-mask action extraction and VLM selection step is described at high level but lacks concrete mapping rules from mask-derived candidates to spatial actions under partial observability, making reproducibility of the reported efficiency gains (SPL) difficult to assess.

minor comments (2)

[Method] Notation for the episodic semantic buffer queue (size, update rule, similarity metric) is introduced without a formal definition or pseudocode, complicating exact re-implementation.
[Figures] Figure captions for navigation trajectories do not indicate whether the shown paths include rethinking corrections or baseline runs, reducing clarity of the qualitative results.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and commit to revisions that strengthen experimental rigor, methodological transparency, and reproducibility without altering the core claims.

read point-by-point responses

Referee: [Experiments] Experiments section: The abstract and results claim concrete SR/SPL gains (e.g., +18.2% SR on HM3D v0.2) but supply no experimental protocol details, baseline implementations, statistical tests, run variance, or ablation results. This leaves the attribution of improvements to the episodic buffer queue unverified and load-bearing for the central claim.

Authors: We agree that the current Experiments section requires expansion for verifiability. In the revised manuscript we will add: full protocol details (episode counts, environment configurations, and evaluation metrics); explicit descriptions of baseline reproductions; statistical significance tests (paired t-tests with p-values); variance across multiple runs (standard deviations from at least three random seeds); and dedicated ablation studies that isolate the episodic semantic buffer queue. A new table will also quantify rethinking correction frequency and false-positive rates to directly support attribution of the reported gains. revision: yes
Referee: [Method] Method section (adaptive dual-modal rethinking): The headline performance is ascribed to the episodic semantic buffer queue verifying target visibility and correcting VLM decisions. No quantitative breakdown is given on correction frequency versus false-positive rate, nor on whether per-environment threshold tuning is required; without this or an ablation removing the queue, the net gains could be artifacts of prompt/scene distribution rather than a robust advance.

Authors: We will augment the Method section with quantitative metrics on the rethinking mechanism, including correction frequency, false-positive rate, and confirmation that the adaptive thresholds are fixed and environment-agnostic rather than tuned per scene. A new ablation study that disables the episodic buffer queue will be included to isolate its contribution and rule out prompt or distribution artifacts. revision: yes
Referee: [§4.3] §4.3 (low-level control): The depth-mask action extraction and VLM selection step is described at high level but lacks concrete mapping rules from mask-derived candidates to spatial actions under partial observability, making reproducibility of the reported efficiency gains (SPL) difficult to assess.

Authors: We will expand §4.3 with explicit mapping rules and pseudocode that detail how mask-derived candidate actions are converted to spatial movements. The revision will address partial observability explicitly, distinguishing cases where the target is absent from the current view (triggering exploration) from cases where it is visible (prioritizing directed approach), thereby enabling full reproduction of the SPL improvements. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical engineering framework with independent evaluation claims

full rationale

The paper describes ReMemNav as a hierarchical framework combining VLMs, panoramic priors, episodic memory, and an adaptive rethinking mechanism. All reported gains (SR/SPL improvements on HM3D v0.1/v0.2 and MP3D) are presented as outcomes of empirical evaluation rather than any derivation, equation, or fitted parameter. No self-definitional steps, fitted-input predictions, load-bearing self-citations, uniqueness theorems, or ansatz smuggling appear in the abstract or described structure. The central claims rest on external benchmark results and are not reduced to the inputs by construction. This is a standard empirical contribution whose performance numbers are falsifiable outside the paper's own definitions.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Central claim depends on the assumption that VLM spatial reasoning becomes reliable once anchored by Recognize Anything Model and episodic memory; no free parameters or new entities are introduced in the abstract.

axioms (1)

domain assumption VLMs provide reliable commonsense reasoning for navigation when anchored by Recognize Anything Model
Framework design explicitly relies on this to ground spatial reasoning and prevent hallucinations.

pith-pipeline@v0.9.0 · 5578 in / 1250 out tokens · 43341 ms · 2026-05-15T01:05:55.670463+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

adaptive dual-modal rethinking mechanism based on an episodic semantic buffer queue... verifies target visibility and corrects decisions
IndisputableMonolith/Foundation/ArithmeticFromLogic.lean LogicNat recovery unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

episodic memory buffer queue... sliding-window... capacity K=10

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

46 extracted references · 20 canonical work pages · 2 internal anchors

[1]

Objectnav revisited: On evaluation of embodied agents navigating to objects.arXiv preprint arXiv:2006.13171, 2020

D. Batra, A. Gokaslan, A. Kembhavi, O. Maksymets, R. Mottaghi, M. Savva, A. Toshev, and E. Wijmans, “Objectnav revisited: On evaluation of embodied agents navigating to objects,”arXiv preprint arXiv:2006.13171, 2020

work page arXiv 2006
[2]

Zson: Zero-shot object-goal navigation using multimodal goal embed- dings,

A. Majumdar, G. Aggarwal, B. Devnani, J. Hoffman, and D. Batra, “Zson: Zero-shot object-goal navigation using multimodal goal embed- dings,”Advances in Neural Information Processing Systems, vol. 35, pp. 32 340–32 352, 2022

2022
[3]

Offline visual representation learning for embodied navigation,

K. Yadav, R. Ramrakhya, A. Majumdar, V .-P. Berges, S. Kuhar, D. Ba- tra, A. Baevski, and O. Maksymets, “Offline visual representation learning for embodied navigation,” inWorkshop on Reincarnating Reinforcement Learning at ICLR 2023, 2023

2023
[4]

Ovrl-v2: A simple state-of-art baseline for imagenav and objectnav,

K. Yadav, A. Majumdar, R. Ramrakhya, N. Yokoyama, A. Baevski, Z. Kira, O. Maksymets, and D. Batra, “Ovrl-v2: A simple state-of-art baseline for imagenav and objectnav,”arXiv preprint arXiv:2303.07798, 2023

work page arXiv 2023
[5]

Bridging zero-shot object navigation and foundation models through pixel-guided navigation skill,

W. Cai, S. Huang, G. Cheng, Y . Long, P. Gao, C. Sun, and H. Dong, “Bridging zero-shot object navigation and foundation models through pixel-guided navigation skill,” in2024 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2024, pp. 5228–5234

2024
[6]

Esc: Exploration with soft commonsense constraints for zero-shot object navigation,

K. Zhou, K. Zheng, C. Pryor, Y . Shen, H. Jin, L. Getoor, and X. E. Wang, “Esc: Exploration with soft commonsense constraints for zero-shot object navigation,” inInternational Conference on Machine Learning. PMLR, 2023, pp. 42 829–42 842

2023
[7]

How to not train your dragon: Training-free embodied object goal navigation with semantic frontiers,

J. Chen, G. Li, S. Kumar, B. Ghanem, and F. Yu, “How to not train your dragon: Training-free embodied object goal navigation with semantic frontiers,”arXiv preprint arXiv:2305.16925, 2023

work page arXiv 2023
[8]

Topv-nav: Unlocking the top-view spatial reasoning potential of mllm for zero-shot object navigation,

L. Zhong, C. Gao, Z. Ding, Y . Liao, H. Ma, S. Zhang, X. Zhou, and S. Liu, “Topv-nav: Unlocking the top-view spatial reasoning potential of mllm for zero-shot object navigation,”arXiv preprint arXiv:2411.16425, 2024

work page arXiv 2024
[9]

V oronav: V oronoi-based zero-shot object navigation with large language model,

P. Wu, Y . Mu, B. Wu, Y . Hou, J. Ma, S. Zhang, and C. Liu, “V oronav: V oronoi-based zero-shot object navigation with large language model,” arXiv preprint arXiv:2401.02695, 2024

work page arXiv 2024
[10]

Thda: Treasure hunt data augmentation for semantic navigation,

O. Maksymets, V . Cartillier, A. Gokaslan, E. Wijmans, W. Galuba, S. Lee, and D. Batra, “Thda: Treasure hunt data augmentation for semantic navigation,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 15 374–15 383

2021
[11]

Simple but effective: Clip embeddings for embodied ai,

A. Khandelwal, L. Weihs, R. Mottaghi, and A. Kembhavi, “Simple but effective: Clip embeddings for embodied ai,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 14 829–14 838

2022
[12]

Habitat- web: Learning embodied object-search strategies from human demon- strations at scale,

R. Ramrakhya, E. Undersander, D. Batra, and A. Das, “Habitat- web: Learning embodied object-search strategies from human demon- strations at scale,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 5173–5183

2022
[13]

Learning active camera for multi-object navigation,

P. Chen, D. Ji, K. Lin, W. Hu, W. Huang, T. Li, M. Tan, and C. Gan, “Learning active camera for multi-object navigation,”Advances in Neural Information Processing Systems, vol. 35, pp. 28 670–28 682, 2022

2022
[14]

Film: Following instructions in language with modular methods.arXiv preprint arXiv:2110.07342, 2021

S. Y . Min, D. S. Chaplot, P. Ravikumar, Y . Bisk, and R. Salakhutdinov, “Film: Following instructions in language with modular methods,” arXiv preprint arXiv:2110.07342, 2021

work page arXiv 2021
[15]

Jarvis: A neuro-symbolic commonsense reason- ing framework for conversational embodied agents,

K. Zheng, K. Zhou, J. Gu, Y . Fan, J. Wang, Z. Di, X. He, and X. E. Wang, “Jarvis: A neuro-symbolic commonsense reason- ing framework for conversational embodied agents,”arXiv preprint arXiv:2208.13266, 2022

work page arXiv 2022
[16]

Openfmnav: Towards open-set zero-shot object navigation via vision-language foundation models,

Y . Kuang, H. Lin, and M. Jiang, “Openfmnav: Towards open-set zero-shot object navigation via vision-language foundation models,” inFindings of the Association for Computational Linguistics: NAACL 2024, 2024, pp. 338–351

2024
[17]

L3mvn: Leveraging large language models for visual target navigation,

B. Yu, H. Kasaei, and M. Cao, “L3mvn: Leveraging large language models for visual target navigation,” in2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2023, pp. 3554–3560

2023
[18]

Navigation with large language models: Semantic guesswork as a heuristic for planning,

D. Shah, M. R. Equi, B. Osi ´nski, F. Xia, B. Ichter, and S. Levine, “Navigation with large language models: Semantic guesswork as a heuristic for planning,” inConference on Robot Learning. PMLR, 2023, pp. 2683–2699

2023
[19]

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,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 23 171–23 181

2023
[20]

Prioritized semantic learning for zero-shot instance navigation,

X. Sun, L. Liu, H. Zhi, R. Qiu, and J. Liang, “Prioritized semantic learning for zero-shot instance navigation,” inEuropean Conference on Computer Vision. Springer, 2024, pp. 161–178

2024
[21]

Imagine before go: Self-supervised generative map for object goal navigation,

S. Zhang, X. Yu, X. Song, X. Wang, and S. Jiang, “Imagine before go: Self-supervised generative map for object goal navigation,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2024, pp. 16 414–16 425

2024
[22]

End-to-end navigation with vision language models: Transforming spatial reasoning into question-answering,

D. Goetting, H. G. Singh, and A. Loquercio, “End-to-end navigation with vision language models: Transforming spatial reasoning into question-answering,”arXiv preprint arXiv:2411.05755, 2024

work page arXiv 2024
[23]

Grounded language-image pre-training,

L. H. Li, P. Zhang, H. Zhang, J. Yang, C. Li, Y . Zhong, L. Wang, L. Yuan, L. Zhang, J.-N. Hwanget al., “Grounded language-image pre-training,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 10 965–10 975

2022
[24]

Surds: Benchmarking spatial un- derstanding and reasoning in driving scenarios with vision language models,

X. Guo, R. Zhang, Y . Duan, Y . He, D. Nie, W. Huang, C. Zhang, S. Liu, H. Zhao, and L. Chen, “Surds: Benchmarking spatial un- derstanding and reasoning in driving scenarios with vision language models,”arXiv preprint arXiv:2411.13112, 2024

work page arXiv 2024
[25]

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,” in2024 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2024, pp. 42–48

2024
[26]

Recognize anything: A strong image tagging model,

Y . Zhang, X. Huang, J. Ma, Z. Li, Z. Luo, Y . Xie, Y . Qin, T. Luo, Y . Li, S. Liuet al., “Recognize anything: A strong image tagging model,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 1724–1732

2024
[27]

Zero-shot object searching using large-scale object relationship prior,

H. Chen, R. Xu, S. Cheng, P. A. Vela, and D. Xu, “Zero-shot object searching using large-scale object relationship prior,”arXiv preprint arXiv:2303.06228, 2023

work page arXiv 2023
[28]

Zero experience required: Plug & play modular transfer learning for semantic visual navigation,

Z. Al-Halah, S. K. Ramakrishnan, and K. Grauman, “Zero experience required: Plug & play modular transfer learning for semantic visual navigation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 17 031–17 041

2022
[29]

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

E. Wijmans, A. Kadian, A. Morcos, S. Lee, I. Essa, D. Parikh, M. Savva, and D. Batra, “Dd-ppo: Learning near-perfect pointgoal navigators from 2.5 billion frames,”arXiv preprint arXiv:1911.00357, 2019

work page arXiv 1911
[30]

Visual language maps for robot navigation,

C. Huang, O. Mees, A. Zeng, and W. Burgard, “Visual language maps for robot navigation,” in2023 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2023, pp. 10 608–10 615

2023
[31]

Wmnav: Integrating vision-language models into world models for object goal navigation,

D. Nie, X. Guo, Y . Duan, R. Zhang, and L. Chen, “Wmnav: Integrating vision-language models into world models for object goal navigation,” in2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2025, pp. 2392–2399

2025
[32]

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

S. Chen, P.-L. Guhur, M. Tapaswi, C. Schmid, and I. Laptev, “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
[33]

Clash: Collaborative large-small hierarchical frame- work for continuous vision-and-language navigation,

L. Wang, Z. He, J. Li, R. Xia, M. Hu, C. Yao, C. Liu, Y . Tang, and Q. Chen, “Clash: Collaborative large-small hierarchical frame- work for continuous vision-and-language navigation,”arXiv preprint arXiv:2512.10360, 2025

work page arXiv 2025
[34]

Doraemon: Decentralized ontology-aware reliable agent with enhanced memory oriented navigation,

T. Gu, L. Li, X. Wang, C. Gong, J. Gong, Z. Zhang, Y . Xie, L. Ma, and X. Tan, “Doraemon: Decentralized ontology-aware reliable agent with enhanced memory oriented navigation,”arXiv preprint arXiv:2505.21969, 2025

work page arXiv 2025
[35]

Swiftsage: A generative agent with fast and slow thinking for complex interactive tasks,

B. Y . Lin, Y . Fu, K. Yang, F. Brahman, S. Huang, C. Bhagavatula, P. Ammanabrolu, Y . Choi, and X. Ren, “Swiftsage: A generative agent with fast and slow thinking for complex interactive tasks,”Advances in Neural Information Processing Systems, vol. 36, pp. 23 813–23 825, 2023

2023
[36]

Ground slow, move fast: A dual-system foundation model for generalizable vision-and-language navigation,

M. Wei, C. Wan, J. Peng, X. Yu, Y . Yang, D. Feng, W. Cai, C. Zhu, T. Wang, J. Panget al., “Ground slow, move fast: A dual-system foundation model for generalizable vision-and-language navigation,” arXiv preprint arXiv:2512.08186, 2025

work page arXiv 2025
[37]

Learning to explore using active neural slam,

D. S. Chaplot, D. Gandhi, S. Gupta, A. Gupta, and R. Salakhutdi- nov, “Learning to explore using active neural slam,”arXiv preprint arXiv:2004.05155, 2020

work page arXiv 2004
[38]

Heal: An empirical study on hallucinations in embodied agents driven by large language models,

T. Chakraborty, U. Ghosh, X. Zhang, F. F. Niloy, Y . Dong, J. Li, A. K. Roy-Chowdhury, and C. Song, “Heal: An empirical study on hallucinations in embodied agents driven by large language models,” arXiv preprint arXiv:2506.15065, 2025

work page arXiv 2025
[39]

Ecbench: Can multi-modal foundation models understand the egocentric world? a holistic embodied cog- nition benchmark,

R. Dang, Y . Yuan, W. Zhang, Y . Xin, B. Zhang, L. Li, L. Wang, Q. Zeng, X. Li, and L. Bing, “Ecbench: Can multi-modal foundation models understand the egocentric world? a holistic embodied cog- nition benchmark,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025, pp. 24 593–24 602

2025
[40]

Habitat-Matterport 3D Dataset (HM3D): 1000 Large-scale 3D Environments for Embodied AI

S. K. Ramakrishnan, A. Gokaslan, E. Wijmans, O. Maksymets, A. Clegg, J. Turner, E. Undersander, W. Galuba, A. Westbury, A. X. Changet al., “Habitat-matterport 3d dataset (hm3d): 1000 large-scale 3d environments for embodied ai,”arXiv preprint arXiv:2109.08238, 2021

work page internal anchor Pith review Pith/arXiv arXiv 2021
[41]

Matterport3D: Learning from RGB-D Data in Indoor Environments

A. Chang, A. Dai, T. Funkhouser, M. Halber, M. Niessner, M. Savva, S. Song, A. Zeng, and Y . Zhang, “Matterport3d: Learning from rgb-d data in indoor environments,”arXiv preprint arXiv:1709.06158, 2017

work page internal anchor Pith review Pith/arXiv arXiv 2017
[42]

Sg-nav: Online 3d scene graph prompting for llm-based zero-shot object navigation,

H. Yin, X. Xu, Z. Wu, J. Zhou, and J. Lu, “Sg-nav: Online 3d scene graph prompting for llm-based zero-shot object navigation,”Advances in neural information processing systems, vol. 37, pp. 5285–5307, 2024

2024
[43]

Unigoal: Towards universal zero-shot goal-oriented navigation,

H. Yin, X. Xu, L. Zhao, Z. Wang, J. Zhou, and J. Lu, “Unigoal: Towards universal zero-shot goal-oriented navigation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recog- nition, 2025, pp. 19 057–19 066

2025
[44]

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

Y . Long, W. Cai, H. Wang, G. Zhan, and H. Dong, “Instructnav: Zero-shot system for generic instruction navigation in unexplored environment,”arXiv preprint arXiv:2406.04882, 2024

work page arXiv 2024
[45]

Multi-floor zero-shot object navigation policy,

L. Zhang, H. Wang, E. Xiao, X. Zhang, Q. Zhang, Z. Jiang, and R. Xu, “Multi-floor zero-shot object navigation policy,” in2025 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2025, pp. 6416–6422

2025
[46]

Panonav: Mapless zero-shot object navigation with panoramic scene parsing and dynamic memory,

Q. Jin, Y . Wu, and C. Chen, “Panonav: Mapless zero-shot object navigation with panoramic scene parsing and dynamic memory,”arXiv preprint arXiv:2511.06840, 2025

work page arXiv 2025