data_pickup_tire
收藏资源简介:
grasp是一个由Robot Data Studio整理和发布的机器人学习数据集,采用Lance格式存储,具体版本为rllab_published_lance_dataset_v2。数据集的核心数据存储在`data/episodes.lance`表中,记录了机器人操作的序列。该数据集包含55个操作片段(episodes),共计18030帧数据,并关联了165个视频文件。此外,数据集中还提供了在`meta/stats/`目录下的状态和动作的归一化统计信息,便于模型训练时进行数据标准化处理。该数据集适用于机器人抓取等强化学习或模仿学习任务的研究与开发。
grasp is a robot learning dataset organized and published by Robot Data Studio, stored in Lance format with the specific version rllab_published_lance_dataset_v2. The core data of the dataset is stored in the `data/episodes.lance` table, recording robot operation sequences. The dataset includes 55 episodes, totaling 18,030 frames of data, and is associated with 165 video files. Additionally, the dataset provides normalized statistics for states and actions in the `meta/stats/` directory, facilitating data standardization during model training. This dataset is suitable for research and development in reinforcement learning or imitation learning tasks such as robot grasping.
根据您提供的数据集详情页面README文件内容,以下是对该数据集的详细总结:
数据集概述
该数据集名为 grasp,由 Robot Data Studio 策划整理,采用 Lance 格式存储。
基本格式信息
- 格式标识符:
rllab_published_lance_dataset_v2 - 主训练表:
data/episodes.lance
数据规模
- 总 Episode 数:55 个
- 总帧数:18,030 帧
- 总视频数:165 个
附加数据
- 状态/动作归一化统计信息:存放在
meta/stats/目录下
数据地址
- 数据集页面地址:https://huggingface.co/datasets/rllab-postech/data_pickup_tire




