allenai/ai2thor

Health v2 · maintenance only

Adoption not part of the score

1785 stars · 298 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

AI2-THOR is an open-source platform from the Allen Institute for AI providing near photo-realistic, interactable 3D environments (iTHOR, ManipulaTHOR, RoboTHOR) for embodied AI and visual reasoning research. It is built on Unity with a Python controller API, supporting physics-based object interaction, multiple agents, and headless cloud rendering.

Use cases

• train embodied AI agents in simulated 3D indoor environments
• research visual navigation and object manipulation with a robotic arm
• run sim2real experiments matching simulated and physical apartments
• generate synthetic images and metadata for computer vision models
• simulate multi-agent interaction in kitchens, bedrooms, and living rooms
• run headless rendering on GPU compute clusters for RL training

When to choose

• you need photo-realistic, physics-based interactable 3D scenes for embodied AI research
• you want a Python API to control agents and query object states in simulation
• you need sim2real counterparts or multi-agent support out of the box

When to avoid

• you need lightweight 2D or tabular RL environments like Gym
• you require real-time game production rather than research simulation
• you cannot run Unity-based rendering or lack GPU resources

Facets

library · maturity active

simulation computer-vision machine-learning reinforcement-learning robotics game-engine artificial-intelligence computer-vision robotics simulation reinforcement-learning python windows cloud embodied-ai unity3d sim2real 3d-environments visual-ai physics-simulation research linux macos

Member repositories

For agents

markdown · JSON · MCP: product_card(name="allenai/ai2thor")