ThreeDWorld-House

ThreeDWorld-House数据集由麻省理工学院-IBM沃森人工智能实验室创建，包含15个模拟的物理房屋环境，每个环境包含6至8个相连的房间，如卧室、客厅和厨房，并配备有家具和其他物品。数据集设计模块化，包括多种地板规划几何形状和墙面及地板材料变化，以及各种家具和道具布局。数据集旨在支持ThreeDWorld Transport Challenge，评估AI代理在物理真实环境中的推理和规划能力，特别是在处理复杂物理场景理解和长期任务方面的能力。

ThreeDWorld-House dataset was developed by the MIT-IBM Watson AI Lab. It contains 15 simulated physical household environments, each with 6 to 8 connected rooms such as bedrooms, living rooms and kitchens, and is furnished with furniture and other objects. The dataset features a modular design, incorporating diverse floorplan geometries, variations in wall and floor materials, as well as various furniture and prop layouts. It is designed to support the ThreeDWorld Transport Challenge, aiming to evaluate the reasoning and planning capabilities of AI Agents in physically realistic environments, especially for complex physical scene understanding and long-term task processing.