ICWM 2027
Call for Papers
Important dates
Paper submission deadline
Phase 1 rejection notifications
Reviews released
Author rebuttal
Final decisions
Camera-ready deadline
Conference
Scope and definition
A world model is a computational model of an environment that represents relevant aspects of its state, structure, or dynamics and supports predicting, simulating, or reasoning about how that environment evolves or responds to actions, interactions, or interventions.
World models may be generative or non-generative, deterministic or probabilistic, action-conditioned or action-free, and may operate in observation, latent, symbolic, or geometric spaces.
World modeling must be the central scientific contribution of the paper, rather than an incidental component of a larger system.
In scope
- Latent dynamics and predictive world-model architectures
- Model-based reinforcement learning, planning, and control
- Learned simulators and models of physical interactions
- World-model evaluation, benchmarks, and theory
Contribution-dependent
- Video generation that models dynamics, rather than visual quality alone
- 3D reconstruction supporting reasoning about an evolving world
- Vision-language-action models in which world modeling is central
- Predictive representation learning that directly advances world modeling
Out of scope
- Generic LLM reasoning or instruction tuning
- Standard classification or object detection
- Model-free RL without a world-model contribution
- Generation work focused solely on visual quality
- Robotics papers where world modeling is incidental
Submission format
ICWM does not require a paper template. Authors may use an established conference format or another clear academic layout, subject to the requirements below.
- Main paper
- Maximum 9 pages, including figures and tables
- Camera-ready
- Maximum 10 pages for accepted papers
- References and appendices
- Excluded from the page limit
- Anonymity
- Single-blind review; include author names and affiliations in the manuscript
- Submission limit
- No limit on the number of submissions per author
- Number of authors
- No limit on the number of authors per submission
- Reviewing requirement
- At least one author of each submission must serve as a reviewer
- Archival status
- Accepted papers will be published in archival conference proceedings
- Concurrent review
- ICWM permits submission of work that is simultaneously under review at another venue. Authors remain responsible for complying with the policies of every other venue involved; ICWM's policy does not override or waive any restriction imposed elsewhere
- Use of AI tools
- Unrestricted; authors may use AI tools in any part of the research or writing process, and no disclosure is required
- Author responsibility
- Authors remain fully responsible for the accuracy, originality, citations, claims, and compliance of their submission. At rebuttal, use of an AI tool will not excuse errors, fabricated references, unsupported claims, or other deficiencies in the paper
- Font size
- Minimum 10pt for main text
- Page size
- A4 or US Letter with reasonable margins and spacing
- File format
-
Searchable, machine-readable PDFHow to check
Check your PDFCopy a paragraph from the PDF into a plain-text editor and confirm that it remains readable.
pdftotext paper.pdf - - Self-contained paper
- Essential contributions and evidence must appear in the main paper; reviewers are not required to read appendices
Review process
Reviews will be conducted on OpenReview using a single-blind, open process.
- Submissions, including author names and affiliations, are public immediately upon submission.
- Authors should not anonymize their manuscripts.
- Official reviewers and Area Chairs remain anonymous.
- Public comments may be posted by identifiable OpenReview users and are distinct from official reviews.
- Phase 1 uses AI-assisted analysis with mandatory human approval. Papers continuing to Phase 2 receive full human peer review.
Submission areas
Authors will select the primary area that best matches their paper during submission.
- Theory and foundations: learning principles, guarantees, computational limits, and fundamental questions.
- Identifiability, causality, and counterfactual reasoning: causal discovery, interventions, causal dynamics, and structural understanding.
- Representations and architectures: latent states, object-centric and compositional models, multimodal representations, and predictive architectures.
- Dynamics, prediction, and simulation: temporal prediction, generative world models, learned simulators, physical consistency, and long-horizon simulation.
- Planning, control, and decision-making: model-based RL, planning with learned models, imagination-based learning, and embodied decision-making.
- Generalization, scalability, and robustness: transfer, out-of-distribution generalization, scaling, uncertainty, reliability, and adaptation.
- Benchmarks, datasets, and evaluation: datasets, evaluation methods, physical and causal consistency, and downstream utility.
- Applications: robotics, autonomous systems, scientific discovery, physical and biological systems, and virtual environments.