Pith. sign in

REVIEW 4 major objections 6 minor 1 cited by

The paper claims that nine state-of-the-art VLLMs all exhibit notable deficiencies in robust path planning, collision avoidance, and active exploration in dynamic industrial navigation, based on a new Unity-based benchmark with safety-orien

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 · deepseek-v4-flash

2026-08-03 20:55 UTC pith:ZUVSB27K

load-bearing objection Promising industrial navigation benchmark, but the action-space inconsistency in the prompt makes the headline VLLM deficiency results unreliable until fixed. the 4 major comments →

arxiv 2511.17384 v2 pith:ZUVSB27K submitted 2025-11-21 cs.RO cs.CV

IndustryNav: Exploring Spatial Reasoning of Embodied Agents in Dynamic Industrial Navigation

Yifan Li , Lichi Li , Anh Dao , Xinyu Zhou , Wenjun Huang , Tianyi Ma , Yicheng Qiao , Zheda Mai
show 10 more authors
Daeun Lee Zichen Chen Pan Wang Lehan Yang Tianlong Wang Zhen Tan Sheng Li Mohit Bansal Yang Ni Yu Kong
This is my paper
classification cs.RO cs.CV
keywords spatial reasoningembodied agentsnavigation benchmarkvision-language modelswarehouse simulationcollision avoidancesafety metricsdynamic environments
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.

The paper sets out to show that spatial reasoning in embodied agents is not adequately tested by existing static, household benchmarks and that it remains unsolved for dynamic industrial navigation. To that end it builds IndustryNav, a benchmark of 12 manually designed Unity warehouse scenes with moving vehicles and workers, and proposes a zero-shot PointGoal pipeline that supplies the agent an egocentric image, global coordinates, and action history each step. Nine vision-language models were run through four start–target pairs per scene, and all of them posted low success rates, double-digit collision rates, and frequent proximity warnings. The authors argue this demonstrates a real gap in local–global planning, distance estimation, and active exploration, and they introduce collision rate and warning rate as metrics for future work on safe embodied navigation.

Core claim

The central claim is a measured capability gap: under a zero-shot PointGoal protocol in dynamic warehouse scenes, none of the nine tested VLLMs reaches the target in more than 65% of runs, closed-source models consistently beat open-source ones, and safety behaviors—collisions and near-miss warnings—are poor across all models. The authors interpret this as evidence that current VLLMs lack robust spatial reasoning for active, dynamic environments, particularly in combining local obstacle avoidance with global path planning.

What carries the argument

The load-bearing mechanism is the IndustryNav evaluation pipeline. At each step the agent receives an egocentric 1024×1024 image, a textual global state (position, heading, target, distance), and the last ten action-state pairs; it must output a JSON action from a discrete set (forward, turn left/right, stop) plus a reasoning trace. The benchmark's distinctive instruments are the two safety metrics: Collision Ratio, computed by checking whether a forward action fails to change position, and Warning Ratio, computed by feeding the egocentric image to a monocular depth estimator and flagging frames where any pixel in a forward region of interest is below a one-meter threshold. A fixed red-cone

Load-bearing premise

The paper's conclusion that VLLMs lack active spatial reasoning rests on the evaluation protocol genuinely measuring that competence: one 70-step run per start–target pair, exact coordinates and distances typed into the prompt, and an action space that differs between the method text and the actual prompt.

What would settle it

Re-run the nine models with a single consistent action space (no back/strafe actions), give each model several independent runs per start–target pair, and add a simple greedy planner as a baseline. If the greedy planner reaches near-perfect success while the VLLMs stay below 70%, or if any VLLM jumps above 90% once the prompt is fixed, the paper's deficiency claim would be strong evidence of prompt-sensitivity rather than a fundamental spatial-reasoning gap.

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

If this is right

  • Embodied AI evaluation should include dynamic, domain-specific scenarios with explicit safety metrics, not only static household question-answering.
  • No current VLLM meets the safety bar for real warehouse deployment: collision rates and warning rates remain far too high.
  • Closed-source models' consistent edge over open-source models points to a concrete target for open-model development; Nemotron is the closest open baseline.
  • Action-state histories materially improve success and safety, while adding a top-down map does not—a practical design hint for future navigation pipelines.
  • The benchmark's 12 scenes and five metrics offer a reusable test suite for the next generation of embodied agents.

Where Pith is reading between the lines

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

  • The reported model ranking may be underpowered: with one run per start–target pair (48 episodes per model), differences like 54% vs 55% success are within the noise; a reader should treat exact ordering cautiously.
  • The paper's action-space description is internally inconsistent: Section 3.2 lists forward/turn/stop, while the prompt in Figure 11 (and Figure 12) includes back and strafe actions. If any model attempted those actions, the simulator may have mis-handled them, potentially deflating success scores.
  • A natural next experiment is to compare against a simple classical planner (e.g., greedy heading toward the target with obstacle avoidance). If such a planner clears the success bar easily, the benchmark is mainly measuring VLLM limitations; if it also struggles, the scenes themselves are unusually hard.
  • The Warning Ratio could be sharpened into a time-to-collision or motion-prediction metric, which would reward agents for anticipating moving forklifts rather than merely reacting to instantaneous depth.

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

4 major / 6 minor

Summary. IndustryNav introduces a Unity-based benchmark of 12 manually constructed warehouse scenes with dynamic objects and workers, together with a zero-shot PointGoal navigation pipeline that feeds egocentric images, global odometry text, and action-state history to VLLMs. Five metrics are proposed: Success Ratio, Distance Ratio, Average Steps, Collision Ratio, and Warning Ratio. Nine VLLMs are evaluated (five closed-source, four open-source) on four start–target pairs per scene with 70-step episodes. The paper reports that no model reaches high success, closed-source models outperform open-source ones, Nemotron-nano-12B is the best open-source model, and safety remains poor. Ablations show action-state history helps and a top-down minimap does not. The supplementary material includes prompts, additional cases, further ablations, and a candid limitations section.

Significance. If the benchmark and evaluation are sound, IndustryNav would fill a real gap: existing spatial-reasoning benchmarks are mostly passive and household-oriented, while industrial navigation with moving obstacles and safety-oriented metrics is practically important. The proposed Collision Ratio and Warning Ratio are useful additions beyond standard success/efficiency metrics, and the paper contains concrete case analyses and ablations. The strengths are the manually built dynamic warehouse assets, the effort to combine local egocentric perception with global odometry, and the explicit attention to safety. However, the paper's central empirical conclusions are currently under-supported by the evaluation protocol and clouded by an internal action-space inconsistency.

major comments (4)
  1. [Sec. 3.2 vs. Fig. 11/Fig. 12] The defined action space is internally inconsistent. Sec. 3.2 states that the agent selects from {forward, turn left, turn right, stop} and provides an Action–State Mapping only for those four actions. The actual navigation prompt in Fig. 11 (and the minimap prompt in Fig. 12) instructs the model to output exactly one action from {forward, back, strafe right, strafe left, stop}, with no dynamics or state-transition rules for back or strafe. The paper never reports how these additional actions were parsed, mapped, or penalized. This is not a cosmetic mismatch: a model that follows the literal prompt may emit back or strafe. If those outputs were ignored, the agent freezes, inflating Average Steps and damaging Distance Ratio; if they were mapped to arbitrary motions, collisions and warnings would be artificially generated. Since the load-bearing claims in Sec. 4.1 — universal deficiency an
  2. [Sec. 4.1, Table 2] The evaluation is statistically underpowered, with no error bars, confidence intervals, or significance tests. Each model is run once on each of 4 start–target pairs per scene, i.e., N=48 episodes per model. For binary success outcomes, the standard error at p≈0.65 and N=48 is about 6.9 percentage points. Differences such as Claude-Haiku-4.5 at 61.81% vs. Gemini-2.5-flash at 65.28% are therefore within one standard error. The conclusion that closed-source models 'consistently outperform' open-source models and the claim that Nemotron 'approaches' closed-source performance are not established without such statistical support. Additionally, no non-VLLM baseline is reported; a simple heading-to-target policy or a random policy is needed to calibrate whether the low success ratios reflect a deficiency of VLLM spatial reasoning or the inherent difficulty of the episodes under the 70-step cap.
  3. [Sec. 3.1, Sec. 4.1] No code, scene files, prompts, raw logs, or evaluation harness are released. For a benchmark paper whose contribution is the 12 Unity warehouse scenarios and the navigation pipeline, the absence of an artifact link is a major reproducibility gap. The community cannot audit the collision detection, warning ROI, action parsing, or trajectory computation without the simulator and scenes. The manuscript should include a release URL or explain clearly why the benchmark cannot be released; without this, the benchmark is not usable as proposed.
  4. [Sec. 3.2, Sec. 4.2] The construct validity of 'active spatial reasoning' is weaker than claimed because the pipeline supplies exact coordinates, target coordinates, and distance-to-target as text at every step. Under this protocol, much of the global reasoning can be performed by language-level arithmetic (current heading vs. desired heading and distance minimization) rather than visual-spatial reasoning from the egocentric image. The qualitative 'distance estimation deficiency' discussion in Sec. 4.2 is then hard to reconcile with the fact that distance is explicitly provided numerically. Either the pipeline should ablate the textual odometry/distance information, or the paper should frame the task as testing instruction-following plus local visual obstacle avoidance rather than holistic spatial reasoning.
minor comments (6)
  1. [Abstract] The abstract says 'fourteen state-of-the-art VLLMs (including models such as GPT-5.2, Claude-4.6, and Gemini-3)', while the main text and Table 2 evaluate nine models, named GPT-5-mini, Claude-4.5, and Gemini-2.5-flash. These numbers and model names must be aligned.
  2. [Sec. 4.1] The results analysis says 'as shown in Tab. 1', but the performance table is Table 2; Table 1 is the comparison with prior benchmarks. Please fix the cross-reference.
  3. [Appendix B.1, Fig. 13] The text refers to 'seven evaluated embodied agents' and mentions 'GPT-3.5-mini', while the main evaluation has nine models including GPT-5-mini. Also, some displayed headings are 'θ=360.0°' and 'Heading 360.0°(West)', although the coordinate system in Sec. 3.2 defines West as θ=0°; normalize angles to [0,360).
  4. [Sec. 3.3, Eq. (4)] Collision Ratio only counts collisions when a forward action is executed and position does not change. Collisions caused by dynamic vehicles or workers moving into the agent while it is stationary or turning are not counted. This should be stated explicitly, and the metric definition should be justified.
  5. [Sec. 3.3, Eq. (5) and Sec. 4.1] The warning threshold is given as '1 meter', but the depth estimates come from Depth Pro and the ROI geometry is not defined. Please specify how the metric depth is calibrated and the exact ROI used, including its size and placement relative to the agent.
  6. [Appendix C.3] The limitation that 'the environment remains static between steps' is important and should appear in the main paper. The word 'dynamic' in the title and abstract should be qualified accordingly; the benchmark is step-based with discrete dynamics, not a continuous asynchronous environment.

Circularity Check

0 steps flagged

No significant circularity: the benchmark's claims are empirical measurements with metric definitions, not derivations that reduce to their own inputs.

full rationale

IndustryNav is a benchmark paper; its load-bearing claims are direct empirical measurements of nine VLLMs. The metrics SR, DR, AS, CR, and WR are defined by explicit equations (Eqs. 1-5) rather than derived from fitted parameters, and the reported performance table is a measurement of model outputs, not a prediction generated from the same data. No parameter is fitted to a subset and then 'predicted' for a closely related quantity, and no uniqueness theorem or prior self-authored result is invoked to force the conclusions. The only self-citation to the same authors' prior work (IndustryEQA, ref. [27]) appears in related work as a description of an existing industrial embodied-QA benchmark and is not load-bearing for the present claims. The internal inconsistency between the Section 3.2 action space (forward/turn left/turn right/stop) and the prompt in Fig. 11 (forward|back|strafe right|strafe left|stop, with no dynamics mapping for back/strafe) is a real protocol-validity concern, but it is not circularity: it does not make any success, efficiency, or safety result true by construction. The ablations are empirical comparisons, and the stated limitations in the appendix concern engineering constraints rather than circular derivation. Overall, the paper's derivations are self-contained and its conclusions are not equivalent to its inputs by definition.

Axiom & Free-Parameter Ledger

6 free parameters · 4 axioms · 0 invented entities

The central empirical claims depend on hand-chosen thresholds and protocol choices (delta, warning threshold, step size, history length, episode cap, number of pairs) rather than on fitted physical parameters. The benchmark itself is the resource, so the evaluation is sensitive to these choices but not derived from them.

free parameters (6)
  • success threshold δ = 20 px
    Used in the SR definition (Eq. 1); hand-chosen; changing it changes every success rate.
  • warning distance threshold = 1 m
    Used to mark warning states in Depth Pro depth maps (Sec. 4.1); hand-chosen with no calibration reported.
  • action step size Δ = 34 px
    Defines forward displacement (Sec. 3.2); affects all trajectory metrics.
  • action-state history length = 10 steps
    Fixed across experiments; ablation (Sec. 4.3) shows it changes success, so the choice matters.
  • episode step limit = 70 steps
    Caps each run (Sec. 4.1); interacts with SR and AS, and a different cap could change model rankings.
  • start-target pairs per scene = 4
    Hand-selected 'varying difficulty' pairs; determines N=48 episodes per model and statistical power.
axioms (4)
  • domain assumption Unity colliders plus 'position changed after forward' is a valid collision detector
    The CR definition (Eq. 4) assumes blocked forward movement equals collision; dynamic object interactions could create false positives/negatives.
  • domain assumption Depth Pro monocular depth is accurate enough to set safety warnings in these scenes
    The WR metric depends on depth estimates (Sec. 3.3, Fig. 4); no validation of Depth Pro accuracy in Unity warehouse scenes is reported.
  • domain assumption The 12 manually designed scenes and four hand-picked start-target pairs per scene are representative of industrial navigation
    Sec. 3.1/4.1; no automated scene generation or quantitative diversity metrics, yet conclusions generalize to 'dynamic industrial environments'.
  • domain assumption OpenRouter API responses at evaluation time represent stable, comparable model capability
    Sec. 4.1; model versions are named but no temperature/decoding settings, repetition counts, or API dates are pinned, so API non-determinism and version drift are uncontrolled.

pith-pipeline@v1.3.0-alltime-deepseek · 18811 in / 14774 out tokens · 138765 ms · 2026-08-03T20:55:47.565260+00:00 · methodology

0 comments
read the original abstract

While Visual Large Language Models (VLLMs) show great promise as embodied agents, they continue to face substantial challenges in spatial reasoning. Existing embodied benchmarks largely focus on passive, static household environments and evaluate isolated capabilities, failing to capture holistic performance in interactive and dynamic complexity of specific domains. To fill this gap, we present IndustryNav, the first dynamic industrial navigation benchmark for active spatial reasoning. IndustryNav leverages 12 manually created, high-fidelity Unity warehouse scenarios featuring dynamic objects and human movement. We proposes a zero-shot PointGoal navigation pipeline that effectively combines egocentric vision with global odometry to assess holistic local-global planning. Furthermore, we introduce the "collision rate" and "warning rate" metrics to measure safety-oriented behaviors. A comprehensive study of fourteen state-of-the-art VLLMs (including models such as GPT-5.2, Claude-4.6, and Gemini-3) reveals that closed-source models maintain a consistent advantage; however, all agents exhibit notable deficiencies in robust path planning, collision avoidance and active exploration. This highlights a critical need for embodied research to move beyond passive perception and toward tasks that demand stable planning, active exploration, and safe behavior in vivid, dynamic environments.

Figures

Figures reproduced from arXiv: 2511.17384 by Anh Dao, Daeun Lee, Lehan Yang, Lichi Li, Mohit Bansal, Pan Wang, Sheng Li, Tianlong Wang, Tianyi Ma, Wenjun Huang, Xinyu Zhou, Yang Ni, Yicheng Qiao, Yifan Li, Yu Kong, Zheda Mai, Zhen Tan, Zichen Chen.

Figure 1
Figure 1. Figure 1: Illustration of IndustryNav benchmark. Industry [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the IndustryNav benchmark. IndustryNav is built on Unity and consists of 12 dynamic warehouse environments. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Visualization of the camera setup for the IndustryNav [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: An illustration of the warning detection. The warning is [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Illustration of both correct (first row) and incorrect (second and third rows) action behaviors of GPT-5-mini under the IndustryNav [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: The effectiveness of action–state histories on (a) success [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 8
Figure 8. Figure 8: Top-down illustrations of 12 manually created warehouses. These layouts are designed in Unity by five experts using diverse [PITH_FULL_IMAGE:figures/full_fig_p013_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Warning detection illustration. corrections, and detours, leading to longer completion times compared to the straightforward static path. These results demonstrate that the navigation task is substantially more challenging in the dynamic industrial scenario. The presence of moving obstacles introduces higher uncertainty and variability, which critically strains the agent’s robustness and its ability to mai… view at source ↗
Figure 10
Figure 10. Figure 10: Examples of key dynamic and static components used to build our industrial warehouse environment. (a) Multiple forklift types, [PITH_FULL_IMAGE:figures/full_fig_p015_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Navigation prompt for embodied agents with egocentric image, history and global odometry information. Critical simulation [PITH_FULL_IMAGE:figures/full_fig_p016_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Navigation prompt for embodied agents with egocentric image and top-down view minimap information. Critical simulation [PITH_FULL_IMAGE:figures/full_fig_p017_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Illustration of incorrect action behaviors of seven embodied agents (from top to bottom): Gemini-2.5-flash, Claude-Sonnet-4.5, [PITH_FULL_IMAGE:figures/full_fig_p019_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Ablation study of the action-state history. The blue bar represents our full (default) navigation pipeline, while the green bar with [PITH_FULL_IMAGE:figures/full_fig_p019_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Ablation study of the top-down map. The blue bar represents our default navigation pipeline, while the textured orange bar [PITH_FULL_IMAGE:figures/full_fig_p020_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Ablation of dynamic vs. static scenarios. The blue bar represents our default navigation pipeline operating in the dynamic [PITH_FULL_IMAGE:figures/full_fig_p020_16.png] view at source ↗

discussion (0)

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

Forward citations

Cited by 1 Pith paper

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

  1. IndustryAssetEQA: A Neurosymbolic Operational Intelligence System for Embodied Question Answering in Industrial Asset Maintenance

    cs.AI 2026-04 unverdicted novelty 5.0

    IndustryAssetEQA integrates episodic telemetry representations with an FMEA knowledge graph to support embodied question answering over industrial assets, showing large gains in validity and reduced overclaims versus ...

Reference graph

Works this paper leans on

62 extracted references · 7 linked inside Pith · cited by 1 Pith paper

  1. [1]

    Vision-and-language navigation: Interpreting visually-grounded navigation instructions in real environments

    Peter Anderson, Qi Wu, Damien Teney, Jake Bruce, Mark Johnson, Niko S ¨underhauf, Ian Reid, Stephen Gould, and Anton Van Den Hengel. Vision-and-language navigation: Interpreting visually-grounded navigation instructions in real environments. InCVPR, pages 3674–3683, 2018. 3

    2018

  2. [2]

    Claude sonnet 4.5, 2025

    Anthropic. Claude sonnet 4.5, 2025. 7

    2025

  3. [3]

    Scanqa: 3d question answering for spatial scene understanding

    Daichi Azuma, Taiki Miyanishi, Shuhei Kurita, and Motoaki Kawanabe. Scanqa: 3d question answering for spatial scene understanding. InCVPR, pages 19129–19139, 2022. 2

    2022

  4. [4]

    The r2r framework: Publishing and discovering mappings on the web.COLD, 665:97–108, 2010

    Christian Bizer and Andreas Schultz. The r2r framework: Publishing and discovering mappings on the web.COLD, 665:97–108, 2010. 3

    2010

  5. [5]

    Depth pro: Sharp monocular metric depth in less than a second

    Alexey Bochkovskiy, Ama ¨el Delaunoy, Hugo Germain, Marcel Santos, Yichao Zhou, Stephan Richter, and Vladlen Koltun. Depth pro: Sharp monocular metric depth in less than a second. InICLR, 2025. 6

    2025

  6. [6]

    Spatialbot: Pre- cise spatial understanding with vision language models

    Wenxiao Cai, Iaroslav Ponomarenko, Jianhao Yuan, Xiaoqi Li, Wankou Yang, Hao Dong, and Bo Zhao. Spatialbot: Pre- cise spatial understanding with vision language models. In ICRA, pages 9490–9498. IEEE, 2025. 2

    2025

  7. [7]

    Partnr: A benchmark for planning and rea- soning in embodied multi-agent tasks

    Matthew Chang, Gunjan Chhablani, Alexander Clegg, Mikael Dallaire Cote, Ruta Desai, Michal Hlavac, Vladimir Karashchuk, Jacob Krantz, Roozbeh Mottaghi, Priyam Parashar, et al. Partnr: A benchmark for planning and rea- soning in embodied multi-agent tasks. InICLR, 2025. 2, 3

    2025

  8. [8]

    Neural topological slam for vi- sual navigation

    Devendra Singh Chaplot, Ruslan Salakhutdinov, Abhinav Gupta, and Saurabh Gupta. Neural topological slam for vi- sual navigation. InCVPR, pages 12875–12884, 2020. 4

    2020

  9. [9]

    Spatialvlm: Endow- ing vision-language models with spatial reasoning capabili- ties

    Boyuan Chen, Zhuo Xu, Sean Kirmani, Brain Ichter, Dorsa Sadigh, Leonidas Guibas, and Fei Xia. Spatialvlm: Endow- ing vision-language models with spatial reasoning capabili- ties. InCVPR, pages 14455–14465, 2024. 2

    2024

  10. [10]

    Spatial- rgpt: Grounded spatial reasoning in vision-language models

    An-Chieh Cheng, Hongxu Yin, Yang Fu, Qiushan Guo, Rui- han Yang, Jan Kautz, Xiaolong Wang, and Sifei Liu. Spatial- rgpt: Grounded spatial reasoning in vision-language models. InNeurIPS, pages 135062–135093, 2024. 1, 2

    2024

  11. [11]

    Efficienteqa: An efficient approach for open vocabulary embodied question answering.arXiv preprint arXiv:2410.20263, 2024

    Kai Cheng, Zhengyuan Li, Xingpeng Sun, Byung-Cheol Min, Amrit Singh Bedi, and Aniket Bera. Efficienteqa: An efficient approach for open vocabulary embodied question answering.arXiv preprint arXiv:2410.20263, 2024. 3

    Pith/arXiv arXiv 2024

  12. [12]

    Lota-bench: Benchmarking language- oriented task planners for embodied agents

    Jaewoo Choi, Youngwoo Yoon, Hyobin Ong, Jaehong Kim, and Minsu Jang. Lota-bench: Benchmarking language- oriented task planners for embodied agents. InICLR, 2024. 2, 3

    2024

  13. [13]

    Embodied question answer- ing

    Abhishek Das, Samyak Datta, Georgia Gkioxari, Stefan Lee, Devi Parikh, and Dhruv Batra. Embodied question answer- ing. InCVPR, pages 1–10, 2018. 3

    2018

  14. [14]

    Mm-spatial: Exploring 3d spatial understanding in multimodal llms

    Erik Daxberger, Nina Wenzel, David Griffiths, Haiming Gang, Justin Lazarow, Gefen Kohavi, Kai Kang, Marcin Eichner, Yinfei Yang, Afshin Dehghan, et al. Mm-spatial: Exploring 3d spatial understanding in multimodal llms. In ICCV, pages 7395–7408, 2025. 2, 3

    2025

  15. [15]

    EmbSpatial-bench: Benchmarking spatial un- derstanding for embodied tasks with large vision-language models

    Mengfei Du, Binhao Wu, Zejun Li, Xuanjing Huang, and Zhongyu Wei. EmbSpatial-bench: Benchmarking spatial un- derstanding for embodied tasks with large vision-language models. InACL, pages 346–355, 2024. 2

    2024

  16. [16]

    Blink: Multimodal large language models can see but not perceive

    Xingyu Fu, Yushi Hu, Bangzheng Li, Yu Feng, Haoyu Wang, Xudong Lin, Dan Roth, Noah A Smith, Wei-Chiu Ma, and Ranjay Krishna. Blink: Multimodal large language models can see but not perceive. InECCV, pages 148–166, 2024. 1

    2024

  17. [17]

    Gemini 2.5 Flash Preview Model Card

    Google DeepMind. Gemini 2.5 Flash Preview Model Card. Technical report, Google, Mountain View, CA, 2025. 7

    2025

  18. [18]

    Autotag & tagmap: Llm-powered moodle plugins for peda- gogical alignment checks.SN Computer Science, 6(7):1–11,

    Christian Gr ´evisse, Claude Braun, and Jos´e Batista da Costa. Autotag & tagmap: Llm-powered moodle plugins for peda- gogical alignment checks.SN Computer Science, 6(7):1–11,

  19. [19]

    3d-llm: In- jecting the 3d world into large language models.Advances in Neural Information Processing Systems, 36:20482–20494,

    Yining Hong, Haoyu Zhen, Peihao Chen, Shuhong Zheng, Yilun Du, Zhenfang Chen, and Chuang Gan. 3d-llm: In- jecting the 3d world into large language models.Advances in Neural Information Processing Systems, 36:20482–20494,

  20. [20]

    Enhancing visualization and interaction of complex spatial data through augmented reality.The International Journal of Advanced Manufacturing Technology, 134(11):5891–5906, 2024

    Zhenghang Hou, Weiping He, and Shuxia Wang. Enhancing visualization and interaction of complex spatial data through augmented reality.The International Journal of Advanced Manufacturing Technology, 134(11):5891–5906, 2024. 3

    2024

  21. [21]

    Visual language maps for robot navigation

    Chenguang Huang, Oier Mees, Andy Zeng, and Wolfram Burgard. Visual language maps for robot navigation. In ICRA, pages 10608–10615. IEEE, 2023. 3

    2023

  22. [22]

    Gqa: A new dataset for real-world visual reasoning and compositional question answering

    Drew A Hudson and Christopher D Manning. Gqa: A new dataset for real-world visual reasoning and compositional question answering. InCVPR, pages 6700–6709, 2019. 2

    2019

  23. [23]

    Clevr: A diagnostic dataset for compositional language and elementary visual reasoning

    Justin Johnson, Bharath Hariharan, Laurens Van Der Maaten, Li Fei-Fei, C Lawrence Zitnick, and Ross Girshick. Clevr: A diagnostic dataset for compositional language and elementary visual reasoning. InCVPR, pages 2901–2910, 2017. 2

    2017

  24. [24]

    Openvla: An open-source vision-language-action model.arXiv preprint arXiv:2406.09246, 2024

    Moo Jin Kim, Karl Pertsch, Siddharth Karamcheti, Ted Xiao, Ashwin Balakrishna, Suraj Nair, Rafael Rafailov, Ethan Foster, Grace Lam, Pannag Sanketi, et al. Openvla: An open-source vision-language-action model.arXiv preprint arXiv:2406.09246, 2024. 1

    Pith/arXiv arXiv 2024

  25. [25]

    Llava-next: Stronger llms supercharge multimodal capa- bilities in the wild, 2024

    Bo Li, Kaichen Zhang, Hao Zhang, Dong Guo, Renrui Zhang, Feng Li, Yuanhan Zhang, Ziwei Liu, and Chunyuan Li. Llava-next: Stronger llms supercharge multimodal capa- bilities in the wild, 2024. 2

    2024

  26. [26]

    Behavior-1k: A benchmark for embodied ai with 1,000 ev- eryday activities and realistic simulation

    Chengshu Li, Ruohan Zhang, Josiah Wong, Cem Gokmen, Sanjana Srivastava, Roberto Mart ´ın-Mart´ın, Chen Wang, Gabrael Levine, Michael Lingelbach, Jiankai Sun, et al. Behavior-1k: A benchmark for embodied ai with 1,000 ev- eryday activities and realistic simulation. InCoRL, pages 80–93. PMLR, 2023. 2, 3

    2023

  27. [27]

    Industryeqa: Push- ing the frontiers of embodied question answering in indus- trial scenarios.arXiv preprint arXiv:2505.20640, 2025

    Yifan Li, Yuhang Chen, Anh Dao, Lichi Li, Zhongyi Cai, Zhen Tan, Tianlong Chen, and Yu Kong. Industryeqa: Push- ing the frontiers of embodied question answering in indus- trial scenarios.arXiv preprint arXiv:2505.20640, 2025. 2

    Pith/arXiv arXiv 2025

  28. [28]

    Visual large language models for generalized and specialized applications.arXiv preprint arXiv:2501.02765, 2025

    Yifan Li, Zhixin Lai, Wentao Bao, Zhen Tan, Anh Dao, Kewei Sui, Jiayi Shen, Dong Liu, Huan Liu, and Yu Kong. Visual large language models for generalized and specialized applications.arXiv preprint arXiv:2501.02765, 2025. 1

    Pith/arXiv arXiv 2025

  29. [29]

    Toa: Task-oriented active vqa.Advances in Neural Information Processing Systems, 36:54061–54074, 2023

    Mingfu Liang, Ying Wu, et al. Toa: Task-oriented active vqa.Advances in Neural Information Processing Systems, 36:54061–54074, 2023. 3

    2023

  30. [30]

    Reasoning paths with reference objects elicit quanti- tative spatial reasoning in large vision-language models

    Yuan-Hong Liao, Rafid Mahmood, Sanja Fidler, and David Acuna. Reasoning paths with reference objects elicit quanti- tative spatial reasoning in large vision-language models. In EMNLP, pages 17028–17047, 2024. 2

    2024

  31. [31]

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

    Bingqian Lin, Yunshuang Nie, Ziming Wei, Jiaqi Chen, Shikui Ma, Jianhua Han, Hang Xu, Xiaojun Chang, and Xiaodan Liang. Navcot: Boosting llm-based vision-and- language navigation via learning disentangled reasoning. IEEE TPAMI, 2025. 3

    2025

  32. [32]

    3dsrbench: A comprehensive 3d spatial reasoning benchmark

    Wufei Ma, Haoyu Chen, Guofeng Zhang, Yu-Cheng Chou, Jieneng Chen, Celso de Melo, and Alan Yuille. 3dsrbench: A comprehensive 3d spatial reasoning benchmark. InICCV, pages 6924–6934, 2025. 2

    2025

  33. [33]

    Openeqa: Embodied question answering in the era of foun- dation models

    Arjun Majumdar, Anurag Ajay, Xiaohan Zhang, Pranav Putta, Sriram Yenamandra, Mikael Henaff, Sneha Silwal, Paul Mcvay, Oleksandr Maksymets, Sergio Arnaud, et al. Openeqa: Embodied question answering in the era of foun- dation models. InCVPR, pages 16488–16498, 2024. 1, 2, 3

    2024

  34. [34]

    Llama 4 model card, 2025

    Meta AI. Llama 4 model card, 2025. 7

    2025

  35. [35]

    Efficient hybrid mamba- transformer reasoning model.arXiv preprint arXiv:2508.14444, 2025

    NVIDIA Nemotron Nano. Efficient hybrid mamba- transformer reasoning model.arXiv preprint arXiv:2508.14444, 2025. 7

    Pith/arXiv arXiv 2025

  36. [36]

    Gpt-4o system card, 2024

    OpenAI. Gpt-4o system card, 2024. 7

    2024

  37. [37]

    Is map- ping necessary for realistic pointgoal navigation? InCVPR, pages 17232–17241, 2022

    Ruslan Partsey, Erik Wijmans, Naoki Yokoyama, Oles Do- bosevych, Dhruv Batra, and Oleksandr Maksymets. Is map- ping necessary for realistic pointgoal navigation? InCVPR, pages 17232–17241, 2022. 4

    2022

  38. [38]

    Reverie: Remote embodied visual referring expres- sion in real indoor environments

    Yuankai Qi, Qi Wu, Peter Anderson, Xin Wang, William Yang Wang, Chunhua Shen, and Anton van den Hengel. Reverie: Remote embodied visual referring expres- sion in real indoor environments. InCVPR, pages 9982– 9991, 2020. 3

    2020

  39. [39]

    Habitat: A plat- form for embodied ai research

    Manolis Savva, Abhishek Kadian, Oleksandr Maksymets, Yili Zhao, Erik Wijmans, Bhavana Jain, Julian Straub, Jia Liu, Vladlen Koltun, Jitendra Malik, et al. Habitat: A plat- form for embodied ai research. InCVPR, pages 9339–9347,

  40. [40]

    Learning to navigate using mid-level visual priors

    Alexander Sax, Jeffrey O Zhang, Bradley Emi, Amir Za- mir, Silvio Savarese, Leonidas Guibas, and Jitendra Malik. Learning to navigate using mid-level visual priors. InCoRL, pages 791–812, 2020. 4

    2020

  41. [41]

    Alfred: A benchmark for interpreting grounded instructions for everyday tasks

    Mohit Shridhar, Jesse Thomason, Daniel Gordon, Yonatan Bisk, Winson Han, Roozbeh Mottaghi, Luke Zettlemoyer, and Dieter Fox. Alfred: A benchmark for interpreting grounded instructions for everyday tasks. InCVPR, 2020. 3

    2020

  42. [42]

    Robospatial: Teaching spatial understanding to 2d and 3d vision-language models for robotics

    Chan Hee Song, Valts Blukis, Jonathan Tremblay, Stephen Tyree, Yu Su, and Stan Birchfield. Robospatial: Teaching spatial understanding to 2d and 3d vision-language models for robotics. InCVPR, pages 15768–15780, 2025. 2

    2025

  43. [43]

    Habitat 2.0: Training home assistants to rearrange their habitat

    Andrew Szot, Alexander Clegg, Eric Undersander, Erik Wi- jmans, Yili Zhao, John Turner, Noah Maestre, Mustafa Mukadam, Devendra Singh Chaplot, Oleksandr Maksymets, et al. Habitat 2.0: Training home assistants to rearrange their habitat. InNeurIPS, pages 251–266, 2021. 4

    2021

  44. [44]

    Cambrian-1: A fully open, vision-centric explo- ration of multimodal llms

    Peter Tong, Ellis Brown, Penghao Wu, Sanghyun Woo, Adithya Jairam Vedagiri IYER, Sai Charitha Akula, Shusheng Yang, Jihan Yang, Manoj Middepogu, Ziteng Wang, et al. Cambrian-1: A fully open, vision-centric explo- ration of multimodal llms. InNeurIPS, pages 87310–87356,

  45. [45]

    Hierarchical open- vocabulary 3d scene graphs for language-grounded robot navigation

    Abdelrhman Werby, Chenguang Huang, Martin B ¨uchner, Abhinav Valada, and Wolfram Burgard. Hierarchical open- vocabulary 3d scene graphs for language-grounded robot navigation. InICRA, 2024. 2

    2024

  46. [46]

    Vsp: Diagnosing the dual challenges of perception and reasoning in spatial planning tasks for mllms

    Qiucheng Wu, Handong Zhao, Michael Saxon, Trung Bui, William Yang Wang, Yang Zhang, and Shiyu Chang. Vsp: Diagnosing the dual challenges of perception and reasoning in spatial planning tasks for mllms. InICCV, pages 2270– 2280, 2025. 2

    2025

  47. [47]

    The rise and potential of large language model based agents: A survey.SCIS, 68(2):121101, 2025

    Zhiheng Xi, Wenxiang Chen, Xin Guo, Wei He, Yiwen Ding, Boyang Hong, Ming Zhang, Junzhe Wang, Senjie Jin, Enyu Zhou, et al. The rise and potential of large language model based agents: A survey.SCIS, 68(2):121101, 2025. 1

    2025

  48. [48]

    Gibson env: Real-world percep- tion for embodied agents

    Fei Xia, Amir R Zamir, Zhiyang He, Alexander Sax, Jitendra Malik, and Silvio Savarese. Gibson env: Real-world percep- tion for embodied agents. InCVPR, pages 9068–9079, 2018. 4

    2018

  49. [49]

    Expand vsr benchmark for vllm to expertize in spatial rules

    Peijin Xie, Lin Sun, Bingquan Liu, Dexin Wang, Xi- angzheng Zhang, Chengjie Sun, and Jiajia Zhang. Expand vsr benchmark for vllm to expertize in spatial rules. InAAAI, pages 8745–8752, 2025. 2

    2025

  50. [50]

    Point2graph: An end-to-end point cloud- based 3d open-vocabulary scene graph for robot navigation

    Yifan Xu, Ziming Luo, Qianwei Wang, Vineet Kamat, and Carol Menassa. Point2graph: An end-to-end point cloud- based 3d open-vocabulary scene graph for robot navigation. InICRA, pages 2853–2860. IEEE, 2025. 2

    2025

  51. [51]

    Qwen3 technical report.arXiv preprint arXiv:2505.09388, 2025

    An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, et al. Qwen3 technical report.arXiv preprint arXiv:2505.09388, 2025. 7

    Pith/arXiv arXiv 2025

  52. [52]

    Thinking in space: How mul- timodal large language models see, remember, and recall spaces

    Jihan Yang, Shusheng Yang, Anjali W Gupta, Rilyn Han, Li Fei-Fei, and Saining Xie. Thinking in space: How mul- timodal large language models see, remember, and recall spaces. InCVPR, pages 10632–10643, 2025. 1, 2

    2025

  53. [53]

    Rila: Re- flective and imaginative language agent for zero-shot seman- tic audio-visual navigation

    Zeyuan Yang, Jiageng Liu, Peihao Chen, Anoop Cherian, Tim K Marks, Jonathan Le Roux, and Chuang Gan. Rila: Re- flective and imaginative language agent for zero-shot seman- tic audio-visual navigation. InCVPR, pages 16251–16261,

  54. [54]

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

    Hang Yin, Xiuwei Xu, Zhenyu Wu, Jie Zhou, and Jiwen Lu. Sg-nav: Online 3d scene graph prompting for llm-based zero-shot object navigation. InNeurIPS, pages 5285–5307,

  55. [55]

    L3mvn: Leveraging large language models for visual target naviga- tion

    Bangguo Yu, Hamidreza Kasaei, and Ming Cao. L3mvn: Leveraging large language models for visual target naviga- tion. InIROS, pages 3554–3560. IEEE, 2023. 3

    2023

  56. [56]

    A continual learning approach for em- bodied question answering with generative adversarial imi- tation learning

    Xin Zeng, Haonan Luo, Zihang Wang, Sijia Li, Zhixuan Shen, and Tianrui Li. A continual learning approach for em- bodied question answering with generative adversarial imi- tation learning. InICASSP, pages 1–5. IEEE, 2025. 3

    2025

  57. [57]

    Open3d-vqa: A benchmark for embodied spatial concept reasoning with multimodal large language model in open space

    Weichen Zhang, Zile Zhou, Xin Zeng, Liu Xuchen, Jianjie Fang, Chen Gao, Jinqiang Cui, Yong Li, Xinlei Chen, and Xiao-Ping Zhang. Open3d-vqa: A benchmark for embodied spatial concept reasoning with multimodal large language model in open space. InACMMM, pages 12784–12791,

  58. [58]

    Dsi- bench: A benchmark for dynamic spatial intelligence.arXiv preprint arXiv:2510.18873, 2025

    Ziang Zhang, Zehan Wang, Guanghao Zhang, Weilong Dai, Yan Xia, Ziang Yan, Minjie Hong, and Zhou Zhao. Dsi- bench: A benchmark for dynamic spatial intelligence.arXiv preprint arXiv:2510.18873, 2025. 2

    arXiv 2025

  59. [59]

    Cityeqa: A hierarchical llm agent on embodied question answering benchmark in city space.arXiv preprint arXiv:2502.12532, 2025

    Yong Zhao, Kai Xu, Zhengqiu Zhu, Yue Hu, Zhiheng Zheng, Yingfeng Chen, Yatai Ji, Chen Gao, Yong Li, and Jincai Huang. Cityeqa: A hierarchical llm agent on embodied question answering benchmark in city space.arXiv preprint arXiv:2502.12532, 2025. 3

    Pith/arXiv arXiv 2025

  60. [60]

    3d- vla: a 3d vision-language-action generative world model

    Haoyu Zhen, Xiaowen Qiu, Peihao Chen, Jincheng Yang, Xin Yan, Yilun Du, Yining Hong, and Chuang Gan. 3d- vla: a 3d vision-language-action generative world model. In ICML, pages 61229–61245, 2024. 2

    2024

  61. [61]

    Towards learning a generalist model for embodied navigation

    Duo Zheng, Shijia Huang, Lin Zhao, Yiwu Zhong, and Liwei Wang. Towards learning a generalist model for embodied navigation. InCVPR, pages 13624–13634, 2024. 3

    2024

  62. [62]

    reasoning

    Filippo Ziliotto, Tommaso Campari, Luciano Serafini, and Lamberto Ballan. Tango: training-free embodied ai agents for open-world tasks. InCVPR, pages 24603–24613, 2025. 1 IndustryNav : Exploring Spatial Reasoning of Embodied Agents in Dynamic Industrial Navigation Supplementary Material A. More Details about IndustryNav A.1. Scene Layouts We provide the t...

    2025