Furniture Bench
收藏资源简介:
FurnitureBench 是由韩国科学技术院(KAIST)和加州大学伯克利分校(UC Berkeley)联合开发的真实世界家具组装基准数据集,旨在推动复杂长时域机器人操作任务的研究。该数据集包含超过 219 小时的演示数据,涵盖 8 种家具模型的 5000 多个演示,数据来源为两名人类操作员通过 Oculus Quest 2 控制器和键盘进行的远程操作。数据集创建过程遵循严格的标准化流程,采用 3D 打印的家具模型,确保实验的可复现性,并提供详细的环境搭建指南和任务初始化工具。其应用领域主要集中在复杂长时域机器人操作任务的研究,如家具组装,旨在解决机器人在真实世界中的长期规划、灵巧操作和视觉感知等挑战性问题。此外,该数据集还提供了 FurnitureSim 模拟器,用于快速迭代实验。
FurnitureBench is a real-world furniture assembly benchmark dataset co-developed by the Korea Advanced Institute of Science and Technology (KAIST) and the University of California, Berkeley (UC Berkeley), aimed at advancing research on complex long-horizon robotic manipulation tasks. This dataset contains over 219 hours of demonstration data, covering more than 5,000 demonstrations across 8 furniture models. The data was collected via teleoperation performed by two human operators using Oculus Quest 2 controllers and keyboards. The dataset is constructed following a strict standardized workflow, adopting 3D-printed furniture models to ensure experimental reproducibility, and provides detailed environment setup guides and task initialization tools. Its main application areas focus on research of complex long-horizon robotic manipulation tasks such as furniture assembly, with the goal of solving challenging problems encountered by robots in real-world scenarios, including long-term planning, dexterous manipulation and visual perception. Additionally, this dataset also offers the FurnitureSim simulator for rapid experimental iteration.




