Arti4D-Semantic
收藏资源简介:
Arti4D-Semantic是由弗莱堡大学、苏黎世联邦理工学院等联合构建的开放世界移动操作数据集,包含62段真实场景RGB-D序列,标注了600次物体交互动作及层级化语义标签。数据集创新性地融合了物体运动轴注释与父子关系标签,涵盖三种观测范式(第一人称、第三人称及机器人视角),支持语义-运动联合理解。其通过人类演示记录铰接物体(如柜门、抽屉)的开启状态变化,旨在解决移动机械臂在动态环境中的长程操作问题,为机器人提供物体运动预测的先验知识。
Arti4D-Semantic is an open-world mobile manipulation dataset jointly constructed by the University of Freiburg, ETH Zurich, and other institutions. It includes 62 real-world RGB-D sequences, annotated with 600 object interaction actions and hierarchical semantic labels. This dataset innovatively integrates object motion axis annotations and parent-child relationship tags, covering three observation paradigms: first-person, third-person, and robot-centric views, to support joint semantic-motion understanding. It records the opening state changes of articulated objects such as cabinet doors and drawers via human demonstrations, aiming to address the long-range manipulation challenges of mobile robotic arms in dynamic environments and provide prior knowledge for robot object motion prediction.



