Icra submission

ICRA Submission

Final-gate audit for a direct ICRA conference submission on PaperPlaza (ras.papercept.net), the RAS submission system used across ICRA, IROS, and RA-L. Everything here is anchored to the verified ICRA 2026 cycle; the numbers move, so open the current year's "call for papers" page before trusting any of them.

Hard limits (2026-cycle values — reverify each year)

Check	2026 rule	Consequence if violated
Page count at submission	6 pages of content + up to 2 pages containing only references	Returned without review
Format	IEEE two-column conference template (RAS PaperPlaza support page templates), PDF only	Compliance failure at upload
Anonymity	Double-anonymous: no authors/affiliations in the PDF; full author list entered in PaperPlaza metadata	Administrative rejection
Video attachment	≤ 180 seconds, ≤ 20 MB, one of mpeg/mp4/mpg	Rejected by the upload form
Video timing	Only inside the posted windows (2026: Aug 5–Sep 9 and Sep 17–22)	No video on record, no late path
Deadline	Sep 15, 2025, 23:59 Pacific (not AoE)	Closed portal

The Pacific-time detail is a live trap: authors calibrated to AoE deadlines at ML conferences lose up to five hours.

The 2026 anonymity turn

Through ICRA 2025, review was single-blind — author names printed on page one. The 2026 cycle switched to double-anonymous. Concretely:

• Strip the author block, affiliations, and identifying acknowledgements/grant numbers from the PDF; keep the full, final author list in PaperPlaza metadata (it cannot be extended casually later — treat the metadata list as binding).
• Rewrite platform tells: "our previously developed XYZ-Arm [12]" becomes "the XYZ-Arm [12]" with third-person citation.
• Blur or crop lab-identifying scenery in figures and in the video attachment — posters on the wall, safety vests with logos, institute door signs.
• Anonymize linked material: a GitHub URL with a lab organization name is an identity leak. Use an anonymized mirror or omit the link at submission.
• Because the policy is one cycle old, confirm it still holds for the target year before de-anonymizing or re-anonymizing anything.

PaperPlaza mechanics worth knowing in advance

• Every author needs a PaperPlaza PIN; collecting PINs from six co-authors on deadline night is a known time sink — do it the week the site opens.
• The system runs its own PDF compliance test (fonts embedded, page size, page count); a failing PDF can be re-uploaded until the deadline, so test early.
• Keywords/session topics are chosen from a fixed RAS taxonomy and steer which Associate Editor sees the paper — pick the two or three closest, not the broadest.
• The submission confirmation email is the only receipt; archive it.

PDF craft the compliance checker enforces

• Fonts must be embedded — the usual failures are figures exported from MATLAB/matplotlib with Type 3 or missing fonts; regenerate with embedded TrueType/Type 1 before deadline week.
• US Letter page size per the IEEE template; a European A4 build fails quietly.
• Vector figures where possible: reviewers zoom into plots of contact forces and trajectories, and rasterized 300-dpi exports turn to mush at 400%.
• Keep the final PDF a sane size — a 60 MB paper full of uncompressed photos strains the system and the reviewers' patience even where no hard cap bites.

Dual-submission and the RA-L boundary

The same manuscript may not be simultaneously under review at ICRA and anywhere else — including RA-L, arXiv exempted (a preprint is allowed; check the current policy text for wording). The clean patterns:

• ICRA reject in January → revise → RA-L in February. Legal and common.
• RA-L accept in autumn → transfer within the posted window for conference presentation. Legal; this is the designed pathway.
• Same PDF to ICRA (September) and RA-L (October) "to hedge": dual submission, discoverable trivially since both run on PaperPlaza infrastructure with overlapping editorial boards.

Metadata fields that outlive the deadline

Three PaperPlaza fields have consequences authors rarely anticipate:

• Author list: feeds the eventual IEEE Xplore record; a co-author omitted "to add later" creates an authorship dispute vector, not a placeholder.
• Keywords: route the paper to an Editor's area and its reviewer pool — effectively the only reviewer-selection influence authors get (see icra-review-process).
• Abstract: printed in the program and reviewer-bidding views; a stale abstract that diverges from the PDF makes bidding reviewers feel misled.

Return-without-review triage

severity ▲
  │ 7-page content body ............... certain return, no repair after deadline
  │ author names on page 1 (2026+) .... administrative reject
  │ template swapped for ML style ..... compliance flag, likely return
  │ references interleaved so page 7
  │   contains prose ................... counts against content limit — fix now
  │ video 200 s / 25 MB ............... attachment refused; trim or re-encode
  │ acknowledgements naming the grant
  │   and PI .......................... anonymity leak; delete for review
  ▼ metadata abstract ≠ PDF abstract ... sloppiness signal to the AE; sync them

Pre-upload sequence (final 48 hours)

1. Regenerate the PDF from a clean clone of the paper repo; confirm the IEEE template class file is untouched.
2. Run the page split check: content ends on or before page 6; pages 7-8 contain nothing but the bibliography.
3. Anonymity sweep: PDF metadata (pdfinfo), acknowledgements, self-citations, figure backgrounds, video frames, embedded links.
4. Play the video attachment start to finish on a machine without the lab codecs; verify duration, size, and that captions carry no names.
5. Enter/verify PaperPlaza metadata: authors, keywords, abstract synced to PDF.
6. Upload paper and video; pass the compliance check; archive the confirmation.

Output format

[ICRA upload verdict] go / fix-first / no-go
[Page audit] content <n>/6, reference pages <n>/2
[Anonymity audit] pass / leaks: <list>
[Video audit] <s>s / <MB>MB / format / window date ok?
[Metadata audit] authors final? keywords chosen? abstract synced?
[Dual-submission] clear / conflict: <venue>
[Deadline math] <local time> vs 23:59 Pacific = <hours remaining>