DriveDNA benchmarks driving style with 4,121 drives and 465 drivers
MOTIF-Lab · hf · 2026-07-28
- DriveDNA is a large-scale multimodal naturalistic driving dataset and benchmark for driving style identification.
- It contains 4,121 drives from 465 drivers across 115 vehicle models, totaling 975 hours of human-controlled driving at 10 Hz with forward video.
- The benchmark tests three tasks: few-shot driver re-identification, personalized behavior prediction, and condition-matched comparison, with additional maneuver annotations and 276,248 rule-generated events.
- Results show learned representations outperform classical descriptors on unseen drivers, while video-only models can suffer from route leakage, meaning high accuracy may come from context shortcuts rather than true driving behavior.
More from Research
- TMLR paper proves stability guarantees for selective SSMs with discontinuous gating — burny_tech · 2026-07-28
- New in ML workshop sets NeurIPS 2026 submissions for Paris with Aug. 29 deadline — niloofar_mire · 2026-07-28
- Kimi K3 report says its 2.8T MoE gains come from better gradient flow — doodlestein · 2026-07-28
- A simple Google Docs trick makes NeurIPS rebuttals easier to export as Markdown — JeanKossaifi · 2026-07-28
- Survey maps progress reward modeling across robotic learning and benchmarks — northwestern-university · 2026-07-28
- Embodied manipulation gets a five-layer data pyramid for robot alignment — PekingUniversity · 2026-07-28