brycewang-stanford RSS Related Work
Position the paper so its claim's novelty is checkable. At a selective single-track venue, related work is not a courtesy section — it is where the Area Chair verifies that the field genuinely lacks what the paper asserts.
The lanes an RSS reviewer expects covered
| Lane | Where it lives | The question your coverage must answer |
|---|---|---|
| RSS's own line | roboticsproceedings.org (free) / dblp | What did the last 2-3 editions say on this problem? |
| The IEEE siblings | ICRA / IROS proceedings | Is this "known at the mega-conferences" already? |
| Robot learning | CoRL, ML venues | If learning-based: what does the CoRL line already show? |
| Journal depth | T-RO, IJRR, AURO | Does a journal version of this idea predate you? |
| Enabling fields | vision/ML/control theory venues | Is the core mechanism imported? Say so and cite the source |
The venue's own lane deserves priority: RSS proceedings are open-access, so "we couldn't access it" is never a defense, and reviewers often are the recent-edition authors under the anonymity veil.
Contrast, don't catalogue
Rewrite tour-style paragraphs into claim-anchored contrasts:
- Weak: "Much work has studied cloth manipulation [3,7,12,19]."
- Strong: "Prior bimanual systems couple contact timing with trajectory optimization [3,7]; we show decoupling them isolates the dominant failure source, which [3,7] cannot attribute."
One contrast sentence per nearest neighbor; the neighbor set is usually 3-5 papers, not 30. Everything else earns at most a grouped citation.
arXiv concurrency in a fast field
Robot learning moves preprint-first, so the January deadline lands amid unrefereed neighbors:
- Cite visible preprints neutrally. A same-quarter arXiv paper with your idea is a fact reviewers will know; acknowledging it with a technical distinction beats hoping nobody noticed.
- Do not claim priority races. "Concurrent work [x] differs in A and B" is verifiable; "we were first" is not.
- Cutoff hygiene: state comparisons against the literature as of submission; the rebuttal (if invited) can acknowledge anything that landed after.
Double-blind self-citation
- Cite your own prior papers in third person, exactly like a stranger's work.
- The robotics-specific trap is the platform chain: citing the paper that built your exact rig, in a sentence that only makes sense if it is yours, de-anonymizes as surely as a name. Describe hardware provenance generically at submission and restore attribution at camera-ready.
- Unpublished own work needed for the argument: include as anonymized supplement, not as a citation to a findable preprint.
Coverage protocol (an hour, well spent)
A repeatable sweep that catches the misses reviewers catch:
1. roboticsproceedings.org: open the last three volumes' tables of contents;
keyword-scan titles (all free, no login).
2. dblp.org/db/conf/rss: confirm venue attribution for anything you plan to
call "prior RSS work" — folklore misfiles ICRA papers as RSS constantly.
3. dblp author pages of your 3-5 nearest neighbors: find their newest
follow-ups, including the IROS/CoRL versions of ideas you know from RSS.
4. arXiv: search your claim's key nouns filtered to the last 12 months;
list hits in a concurrency ledger with a one-line distinction each.
5. Reverse pass: for each baseline you compare against, read who cites it
*this year* — that citation neighborhood is your reviewer pool's memory.
Common positioning failures at this venue
| Failure | How it reads to the AC |
|---|---|
| Only ML venues cited for a manipulation claim | Authors don't know the robotics line |
| The venue's own recent papers absent | Didn't check the free proceedings |
| Nearest neighbor cited but never contrasted | The delta may not exist |
| "First to..." with a findable counterexample | Credibility damage spreads to the claims |
| 60-reference tour, 3 contrasts | Coverage substituting for positioning |
Overlap declarations
- Prior workshop or abstract versions of this work: verify the current CFP's stance and declare rather than gamble — workshop archival status varies.
- Same-cycle submissions to sibling venues with shared content are the routing
hazard of the packed robotics calendar (
rss-workflow); resolve before, not after, the abstract deadline.
Vignette: positioning a contact-rich manipulation claim
A paper claims contact-timing decoupling explains bimanual failure rates. Its neighbor set after the protocol sweep: a recent RSS paper on bimanual planning (contrast: couples timing and motion, cannot attribute failures), an ICRA system with higher raw success on one garment class (contrast: no mechanism isolation, so the number and the claim answer different questions), a CoRL policy-learning line (contrast: learns timing implicitly; this paper measures it explicitly), and a T-RO stability analysis (contrast: quasi-static assumption excludes the regime studied here). Four contrasts, four sentences, and the novelty claim is now checkable by any reviewer who knows one of the four lanes.
Output format
[Lane coverage] rss-line / icra-iros / corl / journals / enabling -> gaps
[Nearest neighbors] <3-5 papers, each with a contrast sentence>
[Concurrent preprints] <item -> neutral handling>
[Self-citation sweep] third-person ok / platform-chain risk found
[Overlap declarations] <none / listed>
[Novelty sentence] <the claim's delta, one line>
Source: brycewang-stanford/Awesome-Journal-Skills → RSS-Skills/skills/rss-related-work/SKILL.md