CONE Builds Work Zones for Autonomous-Driving Tests

Field notes

The project uses ROADWork scenes to derive five work-zone archetypes, then combines rule-based MUTCD templates with a VAE-LDM learned from real layouts. Those layouts are inserted into CARLA with lane closures, props, traffic, and routes before planners are tested end to end. The page reports success-rate changes from 17.9% to 71.4% in-distribution and from 12.5% to 87.0% on held-out geometric and prop variants, while noting that transfer between archetypes remains limited.