Writing study: NitroGen: An Open Foundation Model for Generalist Gaming Agents

← CVPR 2026 oral writing corpus

Loïc Magne, Anas Awadalla, Guanzhi Wang, Yinzhen Xu, Joshua Belofsky, Fengyuan Hu, Joohwan Kim, Ludwig Schmidt, Georgia Gkioxari, Jan Kautz, Yisong Yue, Yejin Choi, Yuke Zhu, Jim Fan · Official CVPR 2026 oral record

Figure, table and section ledger

Editorial contribution type: Datasets, benchmarks & measurement (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: 11 physical pages
Before references / appendix: ≈ 8.05 page-equivalents
Main figures / tables: 7 / 0
Attached appendix figures / tables: 0 / 0
Attached appendix span: Separate file recorded below; not in this PDF total

Before references / appendix: ≈ 73.18%References / intervening material: ≈ 26.82%Estimated layout envelopes, not text/artwork area. Unresolved appendix boundaries remain within the undivided post-body material.

Extraction notes: No attached appendix boundary detected. Zero means no recognized caption, not independently verified absence.

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	≈ 0.57 pp.	5.18%
2. Multi-task multi-game evaluation suite. To assess	1	≈ 0.37 pp.	3.36%
3. Large-scale behavior-cloning pre-training. To	2	≈ 5.52 pp.	50.18%
4. Limitations and future work	7	≈ 0.43 pp.	3.91%
5. Related works	8	≈ 0.63 pp.	5.73%
6. Conclusion	8	≈ 0.19 pp.	1.73%
References	9	≈ 2.95 pp.	26.82%

All detected figure and table locations

Main text

Attached appendix

No recognized caption labels.

Separate source files—excluded from canonical-PDF totals

A separate author version may repeat main text and references. Whole-file lengths and counts are not extra appendix material and are never added to the canonical PDF totals. Read the study's source scope for which sections received substantive reading versus layout inspection.

Additional source 1

Separate PDF (or ZIP containing it): whole-file layout of 3 physical pages; 2 detected figures and 0 tables. These may include overlapping main-text material.

Located appendix only: starts on physical p.1; ≈ 2.94 page-equivalents, 2 figure labels and 0 table labels. This interval is distinct from the whole file.

Detected heading	PDF page	Estimated span
Supplementary Material	1	≈ 0.02 pp.
7. NitroGen model details	1	≈ 0.23 pp.
8. Synchronous inference	1	≈ 0.27 pp.
9. Action labeling	1	≈ 0.19 pp.
10. Use of Large Language Models (LLMs)	1	≈ 2.23 pp.

Argument and organization

Approach §§2.1–2.3
Explain the dataset acquisition bottleneck, evaluation interface and only then the foundation policy.
Experiments and limitations §§3–4
Audit action labels, gameplay and held-out-game finetuning, then delimit short-horizon capabilities and dataset bias.

Problem, claim and contribution

The opening identifies action labels as the scaling bottleneck, making on-screen controller overlays the enabling observation rather than presenting the model architecture as the sole novelty.

Abstract and §1, PDF p. 1; §2.1, PDF p. 2 · Main paper · Additional source 2

Argument and section order

Data, harness and model each have an approach subsection; the evaluation separately checks label quality, pretrained gameplay, held-out-game transfer and self-imitation.

§§2–3, PDF pp. 2–7 · Main paper · Additional source 2

Figures, captions and visual explanation

Figure 2 first shows heterogeneous real overlays, then follows video frame to template match, controller crop and button/joystick labels. Figure 5 groups task-completion bars by visual style and states rollouts per task.

Figure 2, PDF p. 3; Figure 5, PDF p. 6 · Main paper · Additional source 2

Tables, comparisons and evidence

Held-out-game finetuning is compared with the same architecture trained from scratch under the same data and compute budget, whereas the general gameplay figure reports a pretrained model without task-specific finetuning.

§3, PDF p. 7; Figure 7, PDF p. 8 · Main paper · Additional source 2

Appendix and reproducibility

The supplement gives the flow-matching/inference details, tests whether pause/resume changes game dynamics, and breaks action-label performance down by controller family.

Supplement §§7–10, PDF pp. 1–3, Figures 8–9 · Main paper · Additional source 2

Clarity and writing conventions

The limitations explicitly say the model is not an end-to-end game player, long-horizon planner or robust unseen-game zero-shot agent, an important boundary on the foundation-model title.

§4, PDF pp. 7–8 · Main paper · Additional source 2

What to reuse—and what not to infer

Transferable lesson: For a data-driven agent, audit the label extraction pipeline and separate pretrained task performance, transfer after finetuning and long-horizon limitations.

Boundary: Success on diverse games is not evidence of robust zero-shot play on unseen titles; noisy overlay timing and action-game/controller bias remain explicit limitations.

Apply this lesson using the argument and evidence worksheet.

Inspection record and sources

Checked: 2026-09-26
Reading status: Writing-focused manuscript scan + selected visuals inspected
Main text: Manuscript-wide writing scan of PDF pp. 1–11: framing, argument sequence, section organization, figure/table captions, evaluation interpretation and limitations; reference sections identified. This is a writing-focused scan, not an exhaustive technical proof or reproduction audit. Main PDF pages 3 and 6 were rendered and visually inspected.
Appendix / supplement: Writing-focused scan of the separate supplementary PDF, pp. 1–3, including its organization, captions, protocol/evidence detail and limitation material where present; proofs and code listings were inspected for exposition, not independently verified or executed. Supplement §§7–10, PDF pp. 1–3, Figures 8–9. Linked video/audio demonstrations and non-PDF archive contents were not assessed.
Visually inspected PDF pages: 3, 6

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.