AI2-THOR

AI2-THOR (The House Of inteRactions) is an interactive 3D environment framework for visual AI research, developed by the Allen Institute for AI (AI2). The framework provides 120+ photorealistic indoor scenes across multiple room types — kitchens, living rooms, bedrooms, bathrooms — each populated with hundreds of physics-enabled interactable objects. AI2-THOR supports a wide range of embodied AI tasks including visual navigation, object interaction, question answering, task planning, and instruction following. Objects can be picked up, pushed, opened, closed, and placed in realistic physics interactions. The environment includes a procedural generation system for creating unlimited scene variations. Since its release in 2017, AI2-THOR has been used in hundreds of research papers across computer vision, robotics, and embodied AI. It runs on Unity and is fully open-sourced under Apache 2.0. Version 4.0+ added multi-agent support and enhanced physics realism.