Writing study: Proactive Instance Navigation with Comparative Judgment for Ambiguous User Queries

NeurIPS overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← NeurIPS 2026 oral writing corpus

Junhyuk Kwon, Seungjoon Lee, Hyejin Park, Kyle Min, Jungseul Ok · Official NeurIPS 2026 oral record

Observed writing choices, not acceptance rules. This study examines the inspected manuscript version. It does not establish why the paper was accepted or that its presentation choices should be copied unchanged.

Figure, table and section ledger

Editorial contribution type: Domain-grounded methods & physical systems (primary).

Layout extraction, not a required paper template. Counts are distinct caption labels; a multi-panel figure counts once. Section spans are estimated from heading positions, including floats and whitespace. Method and limitations.

Canonical PDF

21 physical pages

Before references / appendix

≈ 9.77 page-equivalents

Main figures / tables

5 / 7

Attached appendix figures / tables

5 / 8

Attached appendix span

≈ 8.91 page-equivalents

Before references / appendix: ≈ 46.52%References / intervening material: ≈ 11.05%Attached appendix: ≈ 42.43%Estimated layout envelopes, not text/artwork area. Unresolved appendix boundaries remain within the undivided post-body material.

Source-bound correction: Top-level heading and reference/appendix boundaries checked against source PDF text. Bibliography titles/author initials, numbered contribution/property bullets, diagram steps, equations and TOC entries are not substantive headings. Section spans are page-coordinate estimates, not ink area or prose-only length. This is an author manuscript mirror, not a verified accepted-version template. Source layout uses physical single-column y/page-height, including equation-heavy pages.

Section-by-section lengths and source pages

The title/abstract/front matter precede the first numbered heading. A span ends at the next detected top-level heading, not at the last sentence; end-page whitespace is included. References and statements are listed separately. These are not measurements of text-only or artwork area.

Top-level heading	Starts on PDF page	Estimated span	Share of entire PDF
1 Introduction	1	≈ 1.43 pp.	6.81%
2 Related Works	3	≈ 0.53 pp.	2.52%
3 Problem Formulation	3	≈ 0.72 pp.	3.43%
4 Proposed Method	4	≈ 1.98 pp.	9.43%
5 Experiments	6	≈ 2.77 pp.	13.19%
6 Analysis	9	≈ 1.37 pp.	6.52%
7 Conclusion	10	≈ 0.31 pp.	1.48%
References	10	≈ 2.32 pp.	11.05%
A Baselines	13	≈ 0.27 pp.	1.29%
B Pool Construction Stage Details	13	≈ 0.2 pp.	0.95%
C NLI-based Entailment Scoring	13	≈ 0.19 pp.	0.9%
D Adaptation of ProCompNav to TextNav	13	≈ 0.85 pp.	4.05%
E Adapting Object Navigation for Instance-Level Exploration	14	≈ 0.86 pp.	4.1%
F Adaptation of AIUTA to TextNav	15	≈ 0.16 pp.	0.76%
G Increasing Query Specificity	15	≈ 0.23 pp.	1.1%
H Effect of The Refinement Threshold	15	≈ 0.8 pp.	3.81%
I Effect of the Pool Size Threshold	16	≈ 0.83 pp.	3.95%
J Stability Across Random Seeds	17	≈ 0.08 pp.	0.38%
K Results with Different MLLMs	17	≈ 0.11 pp.	0.52%
L Robustness Across User Simulators	17	≈ 0.03 pp.	0.14%
M Computational Cost	17	≈ 0.69 pp.	3.29%
N Instance Merge Analysis	18	≈ 0.29 pp.	1.38%
O Human Evaluation Details	18	≈ 1.7 pp.	8.1%
P Failure Attribution Procedure and Limitations	20	≈ 0.22 pp.	1.05%
Q Prompts	20	≈ 1.39 pp.	6.62%

All detected figure and table locations

Main text

• Table 1 · p. 2
• Figure 1 · p. 2
• Figure 2 · p. 4
• Table 2 · p. 7
• Table 3 · p. 7
• Table 4 · p. 8
• Table 5 · p. 8
• Table 6 · p. 8
• Figure 3 · p. 9
• Table 7 · p. 9
• Figure 4 · p. 9
• Figure 5 · p. 10

Attached appendix

• Figure 6 · p. 14
• Table 8 · p. 15
• Figure 7 · p. 16
• Table 9 · p. 16
• Figure 8 · p. 17
• Table 10 · p. 17
• Table 11 · p. 18
• Table 12 · p. 18
• Table 13 · p. 18
• Table 14 · p. 19
• Figure 9 · p. 19
• Table 15 · p. 20
• Figure 10 · p. 21

Argument and organization

1. Problem and contribution

   Defines the user burden of identifying one ambiguous object, contrasting independent matching with candidate-pool comparison rather than framing navigation only as higher success rate.

2. Argument sequence

   Formal task inputs precede pool construction and recursive comparative judgment; experiments measure task success, response length, number of questions, noisy feedback and human response format.

3. Evidence and verification

   The main comparison identifies reproduced baselines sharing language/vision backbones. Failure analysis shows that many failures occur before comparative reasoning is possible, limiting attribution of total success to the proposed component.

Problem, claim and contribution

Defines the user burden of identifying one ambiguous object, contrasting independent matching with candidate-pool comparison rather than framing navigation only as higher success rate.

Introduction, Figure 1 and Table 1, pp. 1–2 · Main paper

Argument and section order

Formal task inputs precede pool construction and recursive comparative judgment; experiments measure task success, response length, number of questions, noisy feedback and human response format.

Sections 3–5, pp. 3–8 · Main paper

Figures, captions and visual explanation

Figure 4 decomposes all episodes into success/failure, target presence, pruning and navigation failures using aligned stacked strips. Table 7 further attributes the pruning subset with explicit counts and percentages. Figure 2 works through a concrete ambiguous query, a candidate-image pool, two comparison groups, attribute-score gaps and a yes/no branch returning to the updated pool. A visible stop condition and repeated state notation make the recursive decision procedure inspectable.

Figure 4 and Table 7, p. 9 (render inspected); Figure 2, p. 4 (additional rendered inspection) · Main paper

Tables, comparisons and evidence

The main comparison identifies reproduced baselines sharing language/vision backbones. Failure analysis shows that many failures occur before comparative reasoning is possible, limiting attribution of total success to the proposed component.

Table 2, p. 7; Section 6.2, p. 9 · Main paper

Appendix and reproducibility

Threshold sweeps retain navigation-cost trade-offs; multi-seed results, alternate user simulators and a counterbalanced human-study interface separate robustness from a single favorable configuration.

Appendix results, Tables 9–15 and Figure 10, pp. 16–21 · Main paper

Clarity and writing conventions

Metric directions and operational definitions keep shorter answers distinct from fewer questions; the conclusion does not erase limitations inherited from exploration and candidate generation.

Sections 5–7, pp. 6–10 · Main paper

What to reuse—and what not to infer

Transferable lesson: For an interactive system, pair task outcomes with human burden and a staged failure decomposition that identifies where the proposed component can actually help.

Boundary: The failure table conditions on 160 target-pruning failures; its percentages are not percentages of all episodes. Preserve the denominator and avoid attributing upstream retrieval failures to the comparison module.

Apply this lesson using the argument and evidence worksheet.

Inspection record and sources

Checked

2026-10-02

Reading status

Writing-focused manuscript scan + selected visuals inspected

Main text

Writing-oriented scan across manuscript framing, section sequence, figure/table captions, selected argument/evaluation prose and conclusion; not exhaustive technical or proof verification. Sections 1–7, pp. 1–10; references pp. 10–12; framing, interactive protocol, method, evaluation and failure analysis scanned. Source is the exact public author version listed below, not a verified accepted/camera-ready revision.

Appendix / supplement

Writing-oriented appendix/documentation scan: Appendices on pp. 13–21: baseline adaptation, pool construction, NLI, navigation mechanics, threshold/feedback/model sensitivity, cost and human-study details. Source code, data, model outputs and external video assets were not executed or independently validated.

Visually inspected PDF pages

4, 9

This is a writing-focused scan of the manuscript and the stated supplementary scope: framing, organization, visual explanation, evidence presentation and wording. It is not a line-by-line proof audit, reproduction of results or exhaustive scientific review.

• Public author arXiv manuscript 2605.06223v4; not verified identical to the accepted or camera-ready OpenReview revision · 21 pages