tape_online_buffer
收藏资源简介:
该数据集由LeRobot项目创建,是一个用于机器人学研究的离线强化学习数据集,采用Apache 2.0许可证。数据集包含60个完整的任务执行序列(episodes),共计7281个时间步(帧),专注于单一任务场景。数据以分块形式组织(chunk size: 1000),包含两种模态:结构化数据(存储在parquet文件中,总计约100MB)和视频数据(存储在mp4文件中,总计约200MB),视频帧率为10fps。所有数据均划分为训练集。每个数据样本包含丰富的多模态观测和动作信息:观测部分包括来自两个相机的RGB图像(分辨率均为128x128)和一个8维的机器人状态向量;动作部分是一个7维的连续控制向量。此外,每个样本还包含强化学习所需的奖励信号、回合结束标志、时间戳、索引信息以及一个补充的离散惩罚信号。该数据集适用于机器人视觉运动控制、离线强化学习、模仿学习以及多模态表示学习等研究任务。
This dataset is created by the LeRobot project and is an offline reinforcement learning dataset for robotics research, licensed under Apache 2.0. It contains 60 complete task execution episodes, totaling 7281 timesteps (frames), focusing on a single task scenario. The data is organized in chunks (chunk size: 1000) and includes two modalities: structured data (stored in parquet files, approximately 100MB in total) and video data (stored in mp4 files, approximately 200MB in total), with a video frame rate of 10fps. All data is divided into the training set. Each data sample contains rich multimodal observations and action information: the observations include RGB images from two cameras (both with a resolution of 128x128) and an 8-dimensional robot state vector; the action is a 7-dimensional continuous control vector. Additionally, each sample includes reinforcement learning-related signals such as reward, episode termination flag, timestamp, index information, and a supplementary discrete penalty signal. This dataset is suitable for research tasks in robot visual motor control, offline reinforcement learning, imitation learning, and multimodal representation learning.




