pepijn223/robocasa_pretrain_human300_v4
收藏资源简介:
该数据集使用LeRobot创建,属于机器人学领域,旨在提供用于预训练的人类行为数据。数据集包含丰富的多模态信息,具体特征包括:三个摄像头视角的视频数据(robot0_agentview_left、robot0_agentview_right、robot0_eye_in_hand),每个视频分辨率为256x256,3通道(RGB),帧率为20fps,使用h264编解码器;动作数据为12维浮点数组,表示机器人的动作控制;状态观测数据为16维浮点数组,表示机器人的状态信息;此外,还包括时间戳、帧索引、剧集索引、索引和任务索引等元数据。数据集规模较大,总剧集数为32043,总帧数为29106226,总任务数为4740,数据文件大小约100MB,视频文件大小约200MB。机器人类型为robocasa,数据全部用于训练集分割。该数据集适用于机器人学习、行为模仿和预训练模型开发等任务。
This dataset was created using LeRobot and belongs to the robotics domain, designed to provide human behavior data for pretraining. It contains rich multimodal information, with specific features including: video data from three camera perspectives (robot0_agentview_left, robot0_agentview_right, robot0_eye_in_hand), each with a resolution of 256x256, 3 channels (RGB), and a frame rate of 20fps, using the h264 codec; action data as a 12-dimensional float32 array representing robot control actions; observation state data as a 16-dimensional float32 array representing robot state information; additionally, it includes metadata such as timestamp, frame index, episode index, index, and task index. The dataset is large in scale, with a total of 32,043 episodes, 29,106,226 frames, and 4,740 tasks, with data files approximately 100MB in size and video files approximately 200MB. The robot type is robocasa, and all data is used for the training split. This dataset is suitable for tasks such as robot learning, behavior imitation, and pretrained model development.



