corl-workflow

CoRL Workflow

CoRL runs one cycle per year on a spring-submit / autumn-decide / November-meet rhythm. This skill turns the 2026 calendar (verified 2026-07-08 via corl.org, OpenReview, and the conference's public timeline) into a working plan, including what a team should be doing right now in each phase.

The 2026 cycle as a reference timeline

Every date above is a single cycle's snapshot. The 2027 cycle will publish its own timeline at the corl.org year-site; nothing here transfers automatically.

Phase playbooks

Now: the submitted-paper quiet period (June–July)

• Do not update the OpenReview PDF — the submission is frozen; focus energy on experiments you may need for the rebuttal instead.
• Pre-build rebuttal ammunition: the most predictable reviewer asks at CoRL are more seeds, more evaluation episodes, a missing robot-learning baseline, and a real-robot check for a sim-heavy paper. Run these before reviews land; the window between reviews and the August 11 rebuttal deadline is days, not weeks.
• Keep the anonymized project site (if any) untouched and unlinked from any identity; review is double-anonymous until decisions.
• Decide the fallback venue now (see corl-topic-selection), so a first-round rejection in early August converts to a resubmission plan the same week.

Reviews land (early August)

• Triage against the first-round rejection rule first: in the 2026 process, a paper with no score of weak-accept-or-above from any reviewer or the AC is a candidate for rejection without rebuttal. If you are past that gate, the rebuttal is live — switch to corl-author-response immediately.
• Map each concern to: (a) already answered in the paper, (b) answerable with a prepared experiment, (c) requires concession. One page forces this triage anyway.

Rebuttal through decision (mid-August – September)

• Submit the one-page PDF by August 11, 11:59 PM AoE; then remain reachable — the reviewer–AC discussion (Aug 12–19) can surface follow-ups, and reviewers are asked to update scores when a rebuttal warrants it.
• Decisions are finalized in SAC–AC meetings and a final SACs–PCs meeting; there is no author-visible activity in that stage, so use September for the camera-ready or resubmission branch prep.

Acceptance branch (September – November)

• Camera-ready: 9-page main text (one page more than submission, explicitly to absorb review feedback), \usepackage[final]{corl_2026}, appendix merged into the same PDF, due October 12, 23:59 AoE — see corl-camera-ready.
• Remember that reviews and rebuttals of accepted papers are made public: write the camera-ready knowing the exchange will be readable next to it.
• Book Austin early; plan the poster/talk material and the public code release (corl-artifact-evaluation) in parallel, not sequentially.

Team calendar template

# corl-cycle-plan.yaml — assign an owner and a date to every row
paper_id: ""
phase_now: post-submission          # update as the cycle advances
rebuttal_prep:
  extra_seeds_run:        {owner: , status: }   # started before reviews arrive
  missing_baseline:       {owner: , status: }
  real_robot_spotcheck:   {owner: , status: }
  one_page_template_ready: {owner: , status: }  # LaTeX skeleton compiled
dates_to_watch:
  reviews_released: "early Aug 2026 (待核实 — watch OpenReview + corl.org)"
  rebuttal_due: "2026-08-11 23:59 AoE"
  discussion: "2026-08-12..19"
  decision: "Sep 2026 (待核实)"
  camera_ready: "2026-10-12 23:59 AoE"
fallback:
  venue: ""                          # decided BEFORE decisions arrive
  required_rework: ""

Planning a CoRL 2027 submission from scratch

Working backward from a late-May deadline (the 2026 pattern; 2027 dates 待核实):

1. T minus 9–12 months — lock the task suite and data-collection plan; hardware time and teleop data are the long poles in robot-learning projects.
2. T minus 6 months — baselines running end to end; if a claim needs sim-to-real evidence, the real-robot rig must exist by now, not in April.
3. T minus 3 months — full evaluation protocol frozen (seeds, episodes, task splits — corl-experiments); start the paper with the Limitations section drafted early, since it is mandatory and counts inside the 8 pages.
4. T minus 1 month — supplementary video storyboarded and shot (corl-supplementary); anonymization sweep of paper, video, and any site.
5. Deadline week — abstract registration a few days before the paper deadline (2026 spacing: May 26 vs May 29); final OpenReview form checks per corl-submission.

Risks specific to this venue's shape

• The rebuttal window is short and fixed; teams that start experiments after reviews arrive routinely miss it. Prepare in the quiet period.
• The single annual deadline means a miss costs a full year — hold the routing discussion (corl-topic-selection) before investing deadline-week effort.
• Hardware fails on deadline schedules: keep one robot-day of slack per claimed real-robot table in the final month.

Reconfirm every date at https://www.corl.org/ (2026 cycle) or the 2027 year-site before acting; this file records the 2026-07-08 state of knowledge.