Papers
Up to 8 pages, excluding references, using the NeurIPS 2026 paper template. Submissions are non-archival and managed on OpenReview.
Machine Learning/CORE Unranked
WM PAI Workshop 2026
World Models in Physical AI Workshop
The World Models in Physical AI Workshop at NeurIPS 2026 brings together researchers from generative modeling, reinforcement learning, robotics, computer vision, and simulation to advance learned models that simulate, plan, and act in the physical world. The workshop focuses on making world models actionable, physically grounded, and deployable for embodied AI systems like robots and autonomous vehicles.
Paper fit
Contribution paths
A strong submission should clearly identify its contribution and evaluate it appropriately.
Research areas in scope
01
Representations and Architectures
Latent versus pixel or video modelsJEPA-style embeddingsIdentifiable latent representationsOmnimodal models3D dynamicsContact dynamics
02
World Models for Action
Model-based reinforcement learningPlanning and controlLearning in imaginationForward and inverse dynamicsWorld-action modelsAction-conditioned prediction
03
Generative Simulation
World models as data generatorsInteractive and closed-loop simulation for robotics and autonomous drivingSim-to-realReal-to-simGenerative simulators and digital twins
04
Evaluation
Physical correctnessCausal faithfulnessDownstream control utilityPolicy-ranking preservationRobustnessGeneralizationBenchmarks and evaluation protocols
05
Scaling and Foundation World Models
Data and compute scalingLarge pretrained world modelsMultimodal dataTransfer across embodimentsAction spaces and domains
06
Safety, Reliability and Broader Impact
UncertaintyControllabilityLong-horizon consistencyFailure analysisSafe deploymentBroader impact of world models that drive physical systems
Policies worth checking twice
- Submissions must be anonymized for double-blind review: remove author names, affiliations, identifying acknowledgements, and funding information.
- Authors must avoid links or supplementary materials that reveal identity.
- At least one author per submission must agree to serve as a reviewer.
- Work already published or presented at the NeurIPS 2026 main conference is not eligible.
- Accepted submissions are non-archival, allowing future submission to archival venues.
- Authors may use AI tools for assistance (e.g., spell checking, grammar, code) but remain fully responsible for all content.
- If AI agents or LLMs are an important, original, or non-standard component of the research, their use must be explicitly described in the paper.
- AI agents or LLMs may not be listed as authors.
Official sources
Compiled from the official call for papers. The organizers’ pages remain authoritative.
Last verified September 9, 2026