家庭机器人抓取数据集

家庭机器人抓取数据集是由卡内基梅隆大学机器人学院创建的大规模数据集，专注于家庭环境中的机器人抓取任务。该数据集包含28000次抓取尝试，数据来源于多个家庭环境，通过自制的低成本移动机械臂收集。创建过程中，研究人员设计了一种能够处理由于不完美执行和校准错误导致的噪声标签的框架。该数据集旨在解决机器人学习中的泛化问题和数据集偏差，通过在真实家庭环境中收集的数据，提高机器人在非结构化视觉环境中的适应性和性能。

The Household Robot Grasping Dataset is a large-scale dataset created by the Robotics Institute at Carnegie Mellon University, focusing on robot grasping tasks in household environments. It contains 28,000 grasping attempts collected from multiple household environments using a self-developed low-cost mobile robotic arm. During the dataset creation process, researchers designed a framework capable of handling noisy labels caused by imperfect execution and calibration errors. This dataset aims to address the generalization problem and dataset bias in robot learning, and improve the adaptability and performance of robots in unstructured visual environments through data collected from real household environments.