Project · Active

ADAS-TO: naturalistic dataset of human takeovers during ADAS engagement

PI

What it is

ADAS-TO is a large-scale multimodal naturalistic dataset documenting how drivers actually take back control from ADAS-engaged vehicles — over 15,000 takeover events, recorded from 327 drivers across 22 vehicle brands, with synchronized video and vehicle telemetry for each event.

Why it matters

Takeover behavior is one of the least-instrumented links in the human–automation chain: most existing datasets come from a small number of research vehicles or simulator studies, which cannot capture how takeover timing and quality vary across the production ADAS systems drivers actually own. ADAS-TO’s scale and brand diversity make it possible to empirically characterize takeover performance the way it is encountered in the field, not just in a lab.

Connections

ADAS-TO shares instrumentation lineage with OpenLKA and feeds the same driver-state modeling questions as TriDrive, BATON, and DriveMotion. It sits in the lab’s AI for mobility and Vehicle technologies threads.

Project team

  • Hao Zhou (PI)
  • Yuhang Wang
  • Yiyao Xu
  • Jingran Sun

Outputs & releases