brycewang-stanford ICRA Author Response
At most conferences "author response" means a rebuttal window. Classic ICRA has
none (verify the current cycle — see icra-review-process): reviews arrive
together with the January decision. Author response at ICRA therefore happens in
three real channels, each with its own craft: the camera-ready revision after
acceptance, the RA-L response letter if the work travels the journal
pathway, and the resubmission memo after a reject. Doing these well is a
skill; pretending a rebuttal exists is a planning error.
Channel 1: the camera-ready revision (accepted papers)
Reviewer comments on an accepted ICRA paper are advisory — no one re-checks compliance — but ignoring them is shortsighted: reviewers are your first readers, their confusions predict everyone else's, and the same people may review the journal extension.
Triage each comment:
| Comment type | Action in camera-ready | Budget guard |
|---|---|---|
| Factual/math error caught | fix unconditionally | always fits |
| Ambiguity that misled a reviewer | rewrite the passage | high priority |
| Missing reference | add it | cheap |
| "Add experiment X" | usually defer to journal version; add only if data exists | 8-page total final limit |
| Stylistic taste | your call | lowest |
| Requests contradicting each other | pick one, note the tension in limitations | — |
Remember the geometry: submissions ran 6 content + 2 reference pages, while the 2026 final limit was 8 pages total including references — additions compete with your bibliography. Never silently change claims: if a fix weakens a number, the number changes and the text stays honest.
Channel 2: the RA-L response letter (journal pathway)
Work routed through RA-L (see icra-topic-selection) can receive a revise
decision, and here a real response letter exists with journal norms:
Response structure per point:
R1.3 (reviewer text quoted verbatim, trimmed fairly)
> Response: <direct answer first — yes/no/agree/disagree>
> Change: Sec. III-B, para 2 rewritten; Table II adds the n=15
> wrist-mounted condition; Fig. 4 caption now states 1× speed.
> (If disagreeing) Evidence: <measurement, citation, or derivation —
> never volume of prose>
- Answer every point, including the painful ones; a skipped point reads as concession or evasion.
- Lead with agreement where genuine — robotics reviewers respond well to "the reviewer is right; trials with the deformable set were added" because added hardware evidence is costly and shows commitment.
- Keep the letter's tone flat and technical; the AE reads it as a signal of whether a second round will converge.
- Respect the clock: RA-L revisions run on a fixed schedule within its ~6-month publish-or-reject envelope; start hardware reruns the day the decision lands, not after the letter is drafted.
Channel 3: the resubmission memo (rejected papers)
A January ICRA reject reaches March IROS or same-week RA-L only through disciplined conversion. Write an internal memo, not a grievance:
- Extract every substantive objection; deduplicate across reviewers.
- Classify: evidence gap / clarity failure / novelty challenge / scope mismatch (the categories predict different fixes).
- For each, decide: fix with new data, fix with rewriting, or consciously accept the risk — with an owner and a date.
- Diff the venue: what IROS or RA-L reviewers will additionally expect (RA-L reviewers, for instance, read with journal-permanence standards; a "promising demo" framing must mature into a validated-claim framing).
- Reviewer-pool overlap is high across ICRA/IROS/RA-L — assume at least one repeat reader who will check whether their objection was addressed.
Tone calibration (robotics reviewers, specifically)
- Reviewers here are practitioners; "in a real deployment this would not matter" arguments fail unless backed by deployment data they can see.
- Concede hardware limits plainly — "the F/T sensor saturates at 50 N, so higher-load trials were not possible" — rather than reframing constraints as choices; the community respects physical honesty.
- Avoid ML-rebuttal reflexes (citing three new preprints, promising a camera- ready appendix): ICRA has no appendix, and the channels above run on data and page-anchored edits, not citation volume.
Discipline common to all three channels
- Quote before answering. Restating a comment in your own words invites the accusation of answering a strawman.
- Evidence outranks assertion. One added table row beats three paragraphs of explanation; in robotics, one new uncut video clip beats both.
- Track everything. Keep a comment → action → location ledger; it becomes the camera-ready cover note, the RA-L response letter skeleton, or the resubmission changelog respectively.
- Never fabricate agility. "We have added extensive new experiments" backed by nothing is checkable and fatal with overlapping reviewer pools.
Ledger template
| # | Reviewer point (quoted) | Class | Action | Where | Status |
|---|-------------------------|-------|--------|-------|--------|
| R2.1 | "success criterion never defined" | clarity | verbatim criterion added | Sec V-A | done |
| R1.4 | "compare against MPC baseline" | evidence | n=20 MPC trials run | Tab III | data 6/12 |
| R3.2 | "extend to deformables" | scope | deferred; limitation stated | Sec VII | done |
Output format
[Channel] camera-ready / RA-L letter / resubmission memo
[Comment ledger] <n> points: fix-now <n> / defer-stated <n> / risk-accepted <n>
[Evidence adds] <new trials/tables/clips committed, with dates>
[Page impact] fits final limit? <y/n + what moves>
[Repeat-reviewer check] objections addressed verbatim-quotably: y/n
[Deadline] <channel deadline and working backward date>