INTERDRIVE
收藏资源简介:
INTERDRIVE是一个大规模的交通场景交互行为数据集,由加州大学伯克利分校、NEC Labs America和加州大学圣地亚哥分校共同构建。该数据集通过精心设计的人类标注流程,确保高质量地捕获真实世界驾驶场景中细腻的agent-agent交互行为,如并线、跟车、让行等。数据集涵盖了来自Waymo Motion和NuPlan的两个数据集的注释,共包含150k个交互行为标注和单agent行为标注,旨在为语言条件下的交通模拟提供训练基础,推动自动驾驶车辆的安全和可靠测试。
INTERDRIVE is a large-scale interactive behavior dataset for traffic scenarios, co-developed by the University of California, Berkeley, NEC Labs America, and the University of California, San Diego. Leveraging a carefully designed human annotation pipeline, this dataset ensures high-quality capture of fine-grained agent-agent interactive behaviors in real-world driving scenarios, including lane merging, car-following, yielding, and other typical driving interactions. The dataset incorporates annotations from two existing datasets, Waymo Motion and NuPlan, and contains a total of 150k annotated instances of interactive behaviors and single-agent behaviors. It aims to provide a training foundation for language-conditioned traffic simulation, and advance the safe and reliable testing of autonomous vehicles.




