BATON
收藏资源简介:
BATON是由南佛罗里达大学与同济大学联合构建的多模态自然驾驶数据集,旨在研究驾驶员与自动驾驶系统间的双向控制转移行为。该数据集包含380条路线、127名驾驶员的136.6小时驾驶数据,同步采集了前视视频、车内视频、车辆CAN信号、雷达交互及GPS上下文等多模态信息,涵盖1,460次控制权移交(handover)和1,432次接管(takeover)事件。数据通过真实道路环境下的车载设备采集,经时序对齐和事件标注处理,主要用于驾驶行为理解、控制权转移预测等任务,为智能驾驶系统的人机交互设计提供实证支持。
BATON is a multimodal natural driving dataset jointly constructed by the University of South Florida and Tongji University, aiming to investigate the bidirectional control transfer behavior between drivers and autonomous driving systems. This dataset contains 136.6 hours of driving data from 127 drivers across 380 routes, and synchronously collects multimodal information including forward-looking video, in-cabin video, vehicle CAN signals, radar interactions, GPS context and other related data. It covers 1,460 control handover events and 1,432 control takeover events. The data is collected via on-board equipment in real road environments, and processed through temporal alignment and event annotation. It is mainly used for tasks such as driving behavior understanding and control transfer prediction, providing empirical support for the human-computer interaction design of intelligent driving systems.




