RLJ Submission Guide
The RLJ peer review process prioritizes rigorous methodology over perceived importance, aiming to foster scholarly discussions on both well-established and emerging topics in RL.
Submission Procedure
RLJ accepts submissions through OpenReview on a rolling basis, with a new volume published four times a year. Papers intended for presentation at RLC must be accepted by that year's RLC deadline.
A submission is a single anonymized PDF typeset with the RLJ style files. See the Authors Guide for how to prepare one.
Scope
We invite submissions presenting original research of interest to the reinforcement learning community. For example, research on topics including, but not limited to:
- RL algorithms (e.g., new algorithms for existing settings and new settings, including agentic systems)
- Hierarchical RL (e.g., skill discovery, hierarchical representations and abstractions)
- Exploration (e.g., intrinsic motivation, curiosity-driven learning, exploration–exploitation tradeoff)
- Theoretical RL (e.g., complexity results, convergence analysis)
- Social and economic aspects (e.g., safety, fairness, interpretability, privacy, trustworthiness, human-AI interaction)
- Bandit algorithms (e.g., theoretical contributions, practical algorithms)
- Planning algorithms (e.g., decision-making under uncertainty, model-based approaches)
- Foundations (e.g., showing relationships between methods, formalizations, unifying theory, clarifying misconceptions in the literature)
- Evaluation (e.g., methodology, meta-studies, replicability, and validity)
- Benchmarks (e.g., specific RL challenges, real-world applications)
- Applied RL (e.g., medical, operations, traffic)
- Deep RL (e.g., analysis of the interplay between RL and deep learning models)
- Multi-agent RL (e.g., cooperative, competitive, self-play)
- RL from human feedback (e.g., reward learning from human data, human-in-the-loop learning)
- Imitation learning (e.g., learning from demonstrations, apprenticeship learning, inverse RL)
- Neuroscience, cognitive science, and philosophical work targeting a computational RL audience
Content
Submissions must be concise, complete, and polished. Manuscripts should be carefully proofread and written with clarity and precision. Authors planning to submit to RLJ should consult the RLJ Reviewers Guide, which describes expectations for presentation quality and common issues that may hinder evaluation.
Note that these instructions call for the rejection of work with even minor errors in the scientific process that prevent reviewers from properly assessing the paper's claims (e.g., omitting the number of trials, using phrases that result in over-claiming), even if these errors could be corrected by changing a few words. Furthermore, rejection typically comes with a 6-month restriction from resubmitting. We therefore strongly encourage you to carefully read the instructions for reviewers to ensure that you are familiar with how your work will be evaluated. Submissions that do not meet these standards may be returned without review.
Originality
Authors should not submit papers that are substantially similar to versions that have been published, accepted for publication, or submitted in parallel to other conferences or journals. Any such submissions may be rejected or retracted after publication. Authors may submit substantially similar versions to workshops and may publish substantially similar versions on arXiv. Notably, dual submission with RLDM is explicitly allowed.
Preprints
Authors may publish substantially similar versions on arXiv.
Disclosure of Funding
Authors must disclose all sources of funding and any relationships or interests that could reasonably be perceived as supporting or influencing the reported work. This disclosure is not solely intended to ensure that publications are attributed to the appropriate funding organizations; it is also an important part of maintaining transparency for readers and enabling them to evaluate the work in its full context.
Accordingly, authors should report all grants, contracts, fellowships, gifts, or other financial support that contributed to the research, as well as any competing interests, including additional sources of income or financial relationships related to the subject of the manuscript. Disclosures should cover support received or interests held within the 18 months preceding submission.
Review Process
The review process for RLJ is double-blind. As such, authors are responsible for ensuring that their submissions do not contain any identifying information. This applies to any materials linked from the submission, such as code. Papers violating this double-blind policy may be rejected without further review.
The RLJ peer review process prioritizes rigorous methodology over subjective perceived importance, aiming to foster scholarly discussions on both well-established and emerging topics in RL.
We use OpenReview for the review process. However, reviews and papers that are rejected or withdrawn will not be made public.
This section summarizes how your submission will be handled. The RLJ Reviewers Guide is the authoritative version, and it is public precisely so that authors can read it. We strongly encourage you to do so before submitting.
Who reviews your paper
Every submission is evaluated by at least two reviewers, supervised by an Editor. The two reviewers play deliberately different roles, and this differs notably from the typical NeurIPS/ICLR/ICML process:
- A Technical Reviewer produces what is essentially a list of technical errors in the paper. There is no commentary on novelty, significance, or impact. This is not an oversight; it is intentional.
- A Senior Reviewer writes the type of review authors traditionally expect, focused on whether claims are matched by evidence—which includes technical correctness—and on how the work is situated in the literature.
Both are asked to work through your cover page contribution by contribution, stating for each one what evidence was offered and whether it was sufficient.
Timeline
The stated duration of each step is a goal, not a guarantee. If a stage runs long, the Editor is expected to leave a comment in OpenReview with an updated timeline.
| Stage | Target |
|---|---|
| Editors-in-Chief desk reject or assign an Editor | 1 week |
| Editor desk rejects or assigns a Senior and a Technical Reviewer | 2 weeks |
| Reviewers write their reviews | 3 weeks |
| Editor writes the meta-review and drafts the decision | 1 week |
| Editors-in-Chief approve the decision | 1 week |
When the reviewing pipeline is over capacity, we prioritize the timely review of initial submissions.
Note that the review process may be delayed around holidays and certain events (such as RLC). In such cases, the email you receive when submitting will indicate the delay.
Decisions
Reviewers recommend Accept, Revision, or Reject. The decision itself rests with the Editor, and is one of two:
- Accept as is: The paper is accepted in its current form; only minor changes are required, to the point that no response to the Editor is needed.
- Reject: The paper is not accepted in its current form. The meta-review states which of the two resubmission terms below applies.
A rejection carries one of two resubmission terms:
| The meta-review says | When you may resubmit |
|---|---|
| The work is not yet ready for publication | As soon as you have addressed every concern the Editor raised |
| The work does not meet the acceptance criteria, or requires significant changes | After 6 months |
Where resubmission is invited, the sooner you resubmit, the more likely it is that some of the same editors and reviewers can be secured for it.
Note that a failure to support a claim typically results in rejection with the 6-month wait, rather than an invitation to resubmit. See Supporting Claims.
If you are invited to resubmit
The meta-review will state what needs to be addressed. Your resubmission should come with a letter to the Editor discussing the primary concerns raised and how your changes address them. You may disagree with a concern, and that disagreement should still be discussed in the letter even if you chose not to change the manuscript.
The Editor may then accept or reject the paper immediately, consult the previous reviewers, or initiate an entirely new review.
Desk rejection
Papers may be desk rejected by the Editors-in-Chief or by the assigned Editor, on the basis of issues identifiable in a short skim. There are two kinds, and the difference to you is substantial:
| Type | Examples | Consequence |
|---|---|---|
| Typographical and formatting | Template violations, text spilling into margins, messy references, unreadable figures, multiple typos | Fix and resubmit immediately |
| Major errors | Missing essential related work or baselines, vacuous theoretical statements, overclaiming with insufficient empirical evidence | 6-month wait before resubmitting |
See Desk Rejection Criteria and Process for the full criteria.
How papers are evaluated
Acceptance is based only on the submission, not on the potential of the paper with further edits. A paper that requires even minor changes may be desk rejected or rejected with an invitation to resubmit. The five acceptance criteria reviewers apply are set out in How Papers are Evaluated.
Two consequences are worth stating plainly for authors. First, presentation alone can sink a paper: if a submission needs a nontrivial editing pass, restructuring, notation cleanup, figure redesign, or reference cleanup, it will not be accepted as is. Second, RLJ deliberately avoids tastemaking: acceptance is not based on perceived impact, and clearly scoped incremental work should be accepted. Reviewers are explicitly instructed not to argue that your environments are too simple, that your method is not state of the art, that your method is too simple, or that you should have compared against work published in the last six months.
Use of Large Language Models (LLMs)
The use of LLMs and other writing tools is allowed in preparing submissions. However, 1) all listed authors should correspond to humans, and 2) the authors are responsible for ensuring that the content of the paper is correct and original. The authors are responsible for ensuring that plagiarized text does not occur, even if the LLM is the source.
Note that senior reviewers are instructed to verify that all cited papers exist, that the listed authors are correct, and that publication venues are accurately reported—generative AI tools can introduce citation errors and fabricated references.
Publication
All papers are published open access under the Creative Commons Attribution (CC BY) license. Copyright remains with the authors, who are free to share the final version of their work.
Accepted papers require a final version prepared with the [accept] option in the style
files, together with a signed publication agreement—see
Preparing the Final Version.
Publication Fees
This journal does not charge publication fees, and all articles are freely accessible to readers.