UCSD Kitchen
收藏资源简介：
UCSD Kitchen数据集由加州大学圣地亚哥分校（UC San Diego）的 Ge Yan、Kris Wu 和 Xiaolong Wang 创建，是一个包含真实世界机器人交互的数据集，旨在支持机器人理解和执行厨房环境中的复杂任务。该数据集包含 150 个演示，涵盖三个不同厨房环境中的五种任务，每种任务包含十个演示。数据集提供了丰富的机器人交互信息，包括 RGB 图像、机器人关节状态、动作指令、语言指令及其嵌入表示等。数据集的创建过程利用 HTC VIVE 控制器和基站跟踪人类手部的 6 自由度运动，并通过 Triad-OpenVR 包将人类操作准确映射到 xArm 机器人上，使其能够在真实厨房环境中与物体交互。数据通过 TensorFlow Datasets 的 GeneratorBasedBuilder 类构建，包含步骤数据和每个演示的元数据。该数据集的应用领域主要集中在机器人行为学习和自然语言指令理解，旨在解决机器人在真实世界复杂环境中执行任务的挑战。研究人员可以利用该数据集开发和评估能够理解自然语言指令并执行复杂任务的机器人系统。
The UCSD Kitchen Dataset was created by Ge Yan, Kris Wu, and Xiaolong Wang from the University of California, San Diego (UC San Diego). It is a real-world robot interaction dataset designed to support robots in understanding and executing complex tasks in kitchen environments. This dataset contains 150 demonstrations, covering five tasks across three distinct kitchen environments, with ten demonstrations per task. It provides rich robot interaction information, including RGB images, robot joint states, action instructions, language instructions and their embedding representations, etc. During the dataset construction, HTC VIVE controllers and base stations are used to track the 6-degree-of-freedom movements of human hands, and human operations are accurately mapped to the xArm robot via the Triad-OpenVR package, enabling the robot to interact with objects in real kitchen environments. The dataset is built using the GeneratorBasedBuilder class of TensorFlow Datasets, and includes step-by-step data and metadata for each demonstration. Its application fields mainly focus on robot behavior learning and natural language instruction understanding, aiming to address the challenges of robots executing tasks in complex real-world environments. Researchers can use this dataset to develop and evaluate robot systems that can understand natural language instructions and perform complex tasks.
