Household Objects Dataset

该数据集包含了38个对象，覆盖了12个通用类别和49项任务，主要用于在现实世界实验中评估抓取方法。该数据集的灵感来源于LERF-TOGO数据集中的对象和任务，并通过部分识别、部分选择和提升成功率等指标进行评估。规模上，该数据集涉及38个对象，分为12个类别，包含49项任务，其研究重点是面向任务的抓取技术。

This dataset comprises 38 objects spanning 12 general categories and 49 tasks, and is primarily designed for evaluating grasping methods in real-world experiments. It draws inspiration from the objects and tasks in the LERF-TOGO dataset, and is evaluated using metrics such as partial recognition, partial selection, and success rate improvement. In terms of scale, this dataset encompasses 38 objects grouped into 12 general categories and includes 49 tasks, with its core research focus being task-oriented grasping techniques.

搜集汇总

背景与挑战

背景概述

该数据集研究机器人对陌生物体的任务导向抓取，通过几何分解和大型语言模型实现零样本抓取。数据集包含多种任务场景和不同阈值下的分解结果，展示了抓取的实用性和有效性。