Research papers
Novel methods, empirical studies, algorithmic advances, benchmarks, datasets, or system demonstrations that bridge generative pretraining and robot action. 4–9 pages + references.
Machine Learning/CORE Unranked
TAG 2026
CoRL 2026 Workshop on The Action Gap: Large Generative Models Learn Rich Knowledge
The G2A 2026 workshop at CoRL 2026 explores the 'action gap' between large-scale generative models trained on video, image, and 3D data and the executable robot policies needed to act in physical environments. It focuses on how world models can be grounded in robot actions, bridging generative pretraining with embodied control through novel methods, benchmarks, and evaluation protocols.
Paper fit
Contribution paths
A strong submission should clearly identify its contribution and evaluate it appropriately.
Position & negative-result papers
Opinionated positions on the right abstractions or evaluation criteria, reproducibility studies, and rigorous negative results that the community should know about. 4–9 pages + references.
Research areas in scope
01
Topics of interest
Generative world models for robot learningVision-language-action modelsWorld action modelsWorld-to-action groundingPredictive dynamics modelsClosed-loop control from generative modelsSim-to-real transferCross-embodiment groundingObject- and interaction-centric representationsAction supervision from diverse dataRobot data efficiency and knowledge transferEvaluation of the action gap
Policies worth checking twice
- Submissions must follow the CoRL 2026 LaTeX template.
- Papers must be 4–9 pages, excluding references and appendices.
- All papers must be anonymized for double-blind review.
- Author names, affiliations, and acknowledgments must be removed.
- Prior work must be cited in the third person.
- Proceedings are non-archival, allowing future conference or journal submissions.
- Work under review at or accepted by other venues is welcome, except CoRL 2026 main-conference papers.
Official sources
Compiled from the official call for papers. The organizers’ pages remain authoritative.
Last verified September 9, 2026