RSS Submission

Final-gate audit for an RSS submission. Every rule below was verified for the 2026 cycle (CFP + review-process page, checked 2026-07-08); reopen the live CFP before enforcing any of it on a later cycle.

The three-deadline cascade

RSS 2026 staged submission across three AoE cutoffs one week apart:

Treat the abstract gate as real: no record by January 23 meant no paper submission. The supplement gate is later than the paper — a sequencing gift ML-venue habits squander by trying to finish everything for the paper date.

Format audit

• Page ceiling, not target: at most 8 pages excluding references, and the CFP states reviewers look favorably on papers that are not unnecessarily long. Padding to the limit is a negative signal at this venue, unlike most.
• Self-containment rule: the main PDF must contain everything needed for an expert to verify the central claims. Appendices, proofs, and data listings go to the supplement — but if a reviewer must open the supplement to believe the claim, the paper fails the rule.
• No external links: project pages, hosted videos, and repos are welcome only in the camera-ready. In the submission they are at best ignored and at worst an anonymity breach.

Double-blind sweep

RSS 2026 review was double-blind. Sweep in this order, because robotics leaks differently from ML:

1. Footage and figures — lab walls, robot decals, institutional colors, staff in frame. Robotics papers de-anonymize through pixels more often than through prose.
2. Platform genealogy — "our previously published gripper [12]" narrows authorship to one lab; rewrite as third-person hardware description.
3. Metadata — PDF author fields, video container tags, archive paths containing usernames.
4. Prose — acknowledgements, grant numbers, dataset names unique to one group.

Eligibility and overlap

• Declare and resolve dual-submission exposure: the same work under review at ICRA, IROS, CoRL, or a journal is a standard robotics failure pattern given the packed calendar (rss-workflow).
• arXiv preprints: handle per the current cycle's wording — do not assume either permission or prohibition (待核实 each year).
• There is no separate demo-paper category in 2026; do not submit a demo write-up expecting a lighter track.

OpenReview record audit

The 2026 venue group lived at roboticsfoundation.org/RSS/2026/Conference. Field errors are self-inflicted desk risks; check each against the PDF:

• Title and abstract match the PDF character-for-character (search indexing and AC assignment both read the form, not the file).
• Author list complete in the form even though the PDF is anonymous — conflicts and reviewer assignment run off form metadata.
• Every author's OpenReview profile has current affiliation and conflict domains; a stale profile silently corrupts conflict detection.
• Subject areas chosen for the AC you want: they route the paper to its jury.
• The supplement upload slot is a separate object with its own February deadline — verify it attached to the right paper ID.

Severity ladder

Last-72-hours sequence

1. Freeze the claims sentence and re-derive the title/abstract from it.
2. Rebuild the PDF from clean sources; check author metadata is empty.
3. Run the anonymity sweep above on paper and supplement together.
4. Verify every figure is legible at print scale — trial-count tables especially.
5. Enter OpenReview fields; confirm the abstract text matches the PDF verbatim.
6. Submit the paper with hours of margin; AoE is generous but robot demos are not the thing to be finishing at 11pm on deadline day.

Notes that save a cycle

• The abstract-gate week is a real strategic asset: title, abstract, and subject areas can be refined against the actual claim during the paper week, but the scope they promise should not swing wildly — ACs see both versions.
• Reference pages are excluded from the 8-page count in 2026; do not compress citations to save body space, and do not smuggle content into reference-page footnotes either.
• Templates and any style-file requirements were not verifiable through the gateway at check time (待核实): download whatever the live CFP links and diff your preamble against it rather than reusing last year's class file.
• If the robot produced new decisive results after January 30: they cannot enter the reviewed record. Bank them for the rebuttal only as clarification, or for the journal extension — not as smuggled supplement additions.

Output format

[Submission verdict] ready / needs fixes / not ready
[Gate status] abstract <done/pending> | paper <done/pending> | supplement <done/pending>
[Format risks] <pages / self-containment / links / template>
[Anonymity findings] <footage / genealogy / metadata / prose>
[Overlap exposure] <none / declared / unresolved>
[Fix order] <ranked list before the next gate>