ember-lab-berkeley/robocasa365-target-composite-unseen
收藏资源简介:
该数据集名为Target (Human) — composite-unseen,是RoboCasa365集合的一部分,包含人类远程操作数据,用于16个复合任务(每个任务500个演示),记录在10个保留的目标厨房中。这些任务在预训练数据集中没有出现,是一个保留的泛化基准。数据集包含8,104个片段,6,724,287帧(20 fps,约93小时),608个任务(自然语言表述,基础RoboCasa任务类别为16个)。数据采集使用了3个256×256 h264视频摄像头,机器人设备为PandaOmron(Panda 7-DoF臂 + Omron移动底座 + 躯干升降)。
The dataset is named Target (Human) — composite-unseen and is part of the RoboCasa365 collection. It contains human teleoperation data for 16 composite tasks (500 demos/task) recorded in 10 held-out target kitchens. These tasks are NOT represented in the pretraining datasets (composite-unseen) — held-out generalization benchmark. The dataset includes 8,104 episodes, 6,724,287 frames (20 fps → 93 h), and 608 tasks (natural-language phrasings; underlying RoboCasa task classes: 16). Data was captured using 3 × 256×256 h264 video cameras, with a PandaOmron robot (Panda 7-DoF arm + Omron mobile base + torso lift).




