AI2-THOR: An Interactive 3D Environment for Visual AI
Eric Kolve1, Roozbeh Mottaghi1,2, Winson Han1, Eli VanderBilt1, Luca Weihs1, Alvaro Herrasti1, Matt Deitke1,2, Kiana Ehsani1, Daniel Gordon2, Yuke Zhu3, Aniruddha Kembhavi1,2, Abhinav Gupta1,4, Ali Farhadi1,2 1Allen Institute for AI, 2University of Washington, 3Stanford University, 4Carnegie Mellon University Figure 1: AI2-THOR consists of interactive 3D environments that can be used with embodied agents.
Abstract
We introduce The House Of inteRactions (THOR), a framework for visual AI research, available at http://ai2thor.allenai.org. AI2-THOR consists of near photo-realistic 3D indoor scenes, where AI agents can navigate in the scenes and interact with objects to perform tasks. AI2-THOR enables research in many different domains including but not limited to deep reinforcement learning, imitation learning, learning by interaction, planning, visual question answering, unsupervised representation learning, object detection and segmentation, and learning models of cognition. The goal of AI2-THOR is to facilitate building visually intelligent models and push the research forward in this domain.
中文速览
让AI真正"看懂"并操作现实世界中的物体，一直是视觉智能领域的核心难题，而缺乏一个高质量、可交互的仿真环境是制约研究进展的重要瓶颈。AI2-THOR（The House Of inteRactions）正是为此而生——它提供了一套接近照片级真实感的室内3D仿真框架，支持智能体在场景中自由导航，并与物体发生丰富的物理交互，例如切面包、用咖啡机煮咖啡、打开水龙头往杯子里注水。研究者可以通过简洁的Python API调用这套系统，获取RGB图像、深度图、语义分割等多模态视觉输出，并在包含手工建模场景和程序化生成的海量房间（多达10000间）中训练和评估模型。自2017年发布以来，AI2-THOR已被超过150篇论文采用、下载逾50万次，覆盖深度强化学习、视觉语言理解、机器人操作、多智能体协作等众多方向，为推动具身人工智能（Embodied AI）从游戏环境走向真实世界提供了重要的研究基础设施。
原文 arXiv:1712.05474;中英对照 + 大白话阅读 https://aha.fim.ai/paper/1712.05474v4