HABIT

HABIT是一个大规模机器人演示数据集，专为人类存在环境设计，旨在教授机器人策略人类感知行为。该数据集填补了现有数据集几乎完全在人类缺席环境中收集的空白，每个片段都包含一个与机器人共享工作空间的共同存在的人类。任务按照三种交互角色组织，捕捉不同的人机交互模式，并包含丰富的人类和机器人子任务级注释。HABIT通过引入人类存在作为数据集多样性的新维度，扩展了机器人策略到与人类共享的环境。HABIT is a large-scale robotic demonstration dataset designed for human-present environments, aiming to teach robots human-aware behaviors. This dataset addresses the gap in existing datasets that are mostly collected in human-absent environments, with each clip featuring a co-present human sharing the workspace with the robot. Tasks are organized according to three interaction roles, capturing diverse human-robot interaction patterns, and include rich human and robot sub-task level annotations. By introducing human presence as a new dimension of dataset diversity, HABIT extends robotic policies to environments shared with humans.