FT-LLM-2026-RAMEN/fractal

该数据集是一个机器人相关数据集，包含87212个episodes，3786400帧数据，涵盖599个不同任务。数据包括256x320分辨率的RGB图像（3fps）、8维状态观测（电机x,y,z,rx,ry,rz,rw和夹持器状态）、7维动作（x,y,z,roll,pitch,yaw和夹持器动作）以及时间戳、帧索引等元数据。数据集采用分块存储（每块1000个episodes），总数据量100MB，视频文件500MB。使用Google机器人采集，适用于机器人学习研究。

This is a robotics dataset containing 87,212 episodes and 3,786,400 frames across 599 different tasks. It includes 256x320 resolution RGB images (3fps), 8-dimensional state observations (motors x,y,z,rx,ry,rz,rw and gripper status), 7-dimensional actions (x,y,z,roll,pitch,yaw and gripper action), along with timestamps, frame indices and other metadata. The dataset is stored in chunks (1000 episodes per chunk) with total data size of 100MB and video files of 500MB. Collected using Google robot, its suitable for robotics learning research.