Project · Active
BATON: bidirectional automation transition benchmark
What it is
BATON is a multimodal benchmark that observes the full loop of human–automation control transitions — both directions, not just takeovers. It draws on 136.6 hours of naturalistic driving from 127 drivers across 380 routes, with multiple synchronized sensor streams captured throughout each transition.
Why it matters
Prior work on automation transitions has focused almost entirely on the takeover direction — the moment a driver must retake control. BATON’s bidirectional framing captures the handoff into automation as well, which matters because how a driver disengages shapes how ready they are to re-engage. That symmetry is what makes BATON useful as a shared benchmark rather than a single-purpose dataset.
Connections
BATON is one of the fleet sources behind DriveMotion’s in-cabin motion corpus, and its transition events complement ADAS-TO and TriDrive’s driver-state forecasting. It sits in the lab’s AI for mobility and Vehicle technologies threads.
Project team
- Hao Zhou (PI)
- Yuhang Wang
- Yiyao Xu
- C. Yang
- L. Li
- Jingran Sun
Outputs & releases
- Paper: Wang, Xu, Yang, Li, Sun, Zhou. BATON: a multimodal benchmark for bidirectional automation transition observation in naturalistic driving. arXiv:2604.07263 — see also the publication page
- Project page: wangyuhang-cmd.github.io/baton
- Code: github.com/OpenLKA/BATON
- Dataset: huggingface.co/datasets/HenryYHW/BATON