Survey companion site
From Generation to Guarantee
A Survey on World Models, Vision-Language-Action Models, and Safety Verification for Long-Tail Autonomous Driving
Citation
For use in publications and presentations please cite this survey as follows:
Que, H., Wu, Q., Yao, H. “From Generation to Guarantee: A Survey on World Models, Vision-Language-Action Models, and Safety Verification for Long-Tail Autonomous Driving.” Manuscript under review, 2026.
The Core Thesis
We read three research families as one coordinated response to the open-world long tail: world models generate the tail, vision–language–action models understand it, and safety verification guarantees behaviour on it. The catch: closing the loop from guarantee back to generation needs a shared scenario representation that is at once generatable, interpretable, and checkable — an interface that does not yet exist. Hover or tap a stage.
Generation, understanding, and guarantee form a cycle. The survey’s thesis is that the cycle is held open by a missing interface — a scenario object every stage can both produce and consume. Hover a stage to see its role.
Statistics
1989~2026Papers per research area
Papers per year
Paper Explorer
1989~2026Hover a point to preview its in-corpus citation neighbourhood (solid = papers it cites, dashed = papers citing it); click a point to pin its citation neighbourhood; Esc or a click on the chart background releases it. Double-click a point to open the paper. Markers: shape = venue type, colour = research area.
Paper List
1989~2026All references from the survey bibliography, grouped by research area. Lists respect the filters and search above.