BEHAVIOR-1K

BEHAVIOR-1K是由斯坦福大学等机构联合开发的机器人行为基准数据集，旨在推动以人类为中心的机器人技术发展。该数据集包含1,000种日常活动，涵盖50种场景（如房屋、花园、餐厅、办公室等）和超过9,000个带有丰富物理与语义属性的对象。数据集基于大规模调查结果构建，反映了人类对机器人完成日常任务的真实需求。其内容包括基于谓词逻辑的活动定义、高保真3D场景和对象模型。BEHAVIOR-1K通过OMNIGIBSON模拟环境实现，支持刚体、柔性体和流体的物理模拟，为机器人学习提供了高度逼真的训练平台。该数据集可用于开发和评估机器人在复杂环境中的长期任务规划和物理操作能力，助力机器人技术从实验室走向实际应用。

BEHAVIOR-1K is a robotic behavior benchmark dataset co-developed by Stanford University and other research institutions, aimed at advancing human-centric robotic technology. This dataset encompasses 1,000 daily activities, covering 50 scenarios such as houses, gardens, restaurants, offices and more, along with over 9,000 objects endowed with rich physical and semantic attributes. Constructed based on large-scale survey results, the dataset reflects the real-world demands of humans for robots to complete daily tasks. Its contents include predicate logic-based activity definitions, high-fidelity 3D scene and object models. BEHAVIOR-1K is implemented via the OMNIGIBSON simulation environment, which supports physical simulation of rigid bodies, flexible bodies and fluids, providing a highly realistic training platform for robot learning. This dataset can be used to develop and evaluate robots' long-term task planning and physical manipulation capabilities in complex environments, facilitating the transition of robotic technology from laboratories to real-world applications.

搜集汇总

背景与挑战

背景概述

BEHAVIOR-1K是一个基于真实人类需求构建的模拟基准数据集，包含1000个家庭活动，这些活动源自调查，反映了人们实际希望机器人协助完成的任务。数据集在50个完全交互式场景中实例化，涉及超过10,000个对象，并通过OmniGibson平台提供高度真实的感知、物理和交互模拟，包括流体、变形和热效应等特性，旨在解决长期移动操作挑战，以弥合研究与现实应用之间的差距。