kejia98/pnp_sponge_franka

该数据集是通过本地多模态记录器导出创建的，专为机器人任务设计，特别是针对Franka机器人。数据集包含5个训练集，总计1389帧，无视频数据，采用分块存储（每块1000帧，帧率为15fps）。数据结构包括图像特征（主图像和次图像，尺寸为240x320x3）、机器人状态（如末端执行器位置、姿态和夹爪状态）、动作数据（如控制命令），以及其他元数据（如时间戳、帧索引、集索引等）。这些多模态数据可用于机器人学习、控制算法开发和仿真任务，支持强化学习或模仿学习等应用场景。数据集以Parquet格式存储，便于高效处理和分析。

This dataset was created using the local multimodal recorder exporter. It is designed for robotics tasks, specifically for the Franka robot. The dataset includes 5 training episodes, totaling 1389 frames, with no video data, and is stored in chunks (each chunk contains 1000 frames at 15fps). The data structure comprises image features (primary and secondary images, sized 240x320x3), robot states (such as end-effector positions, orientations, and gripper states), action data (e.g., control commands), and other metadata (like timestamps, frame indices, episode indices). This multimodal data can be used for robot learning, control algorithm development, and simulation tasks, supporting applications such as reinforcement learning or imitation learning. The dataset is stored in Parquet format for efficient processing and analysis.