AI2-THOR
E366086
AI2-THOR is an interactive 3D simulation platform designed for training and evaluating embodied AI agents in visually rich, physics-enabled environments.
All labels observed (1)
| Label | Occurrences |
|---|---|
| AI2-THOR canonical | 1 |
Statements (66)
| Predicate | Object |
|---|---|
| instanceOf |
3D simulation environment
ⓘ
AI research tool ⓘ embodied AI benchmark ⓘ simulation platform ⓘ |
| affiliation | Allen Institute for Artificial Intelligence projects ⓘ |
| developer |
Allen Institute for Artificial Intelligence
ⓘ
surface form:
Allen Institute for AI
|
| goal |
bridge simulation and real-world robotics
ⓘ
enable reproducible embodied AI experiments ⓘ |
| hasDimension | 3D ⓘ |
| hasEnvironmentType |
bathrooms
ⓘ
bedrooms ⓘ household environments ⓘ indoor scenes ⓘ kitchens ⓘ living rooms ⓘ offices ⓘ |
| hasFeature |
continuous control support
ⓘ
depth observations ⓘ discrete action space ⓘ domain randomization options ⓘ evaluation metrics ⓘ high-level actions API ⓘ interactive objects ⓘ low-level control API ⓘ metadata for objects ⓘ multi-room scenes ⓘ object affordances ⓘ physics-enabled environment ⓘ procedural scene variations ⓘ segmentation masks ⓘ task definitions ⓘ visual observations ⓘ |
| hasInterface |
Python API
ⓘ
Unity environment build ⓘ |
| license | open-source ⓘ |
| programmingLanguage | Python ⓘ |
| provides |
RGB images
ⓘ
agent pose information ⓘ depth maps ⓘ instance segmentation ⓘ object bounding boxes ⓘ object states ⓘ semantic segmentation ⓘ |
| renderingEngine |
Unity game engine
ⓘ
surface form:
Unity
|
| shortName |
Thor
ⓘ
surface form:
THOR
|
| supports |
computer vision research
ⓘ
embodied AI agents ⓘ imitation learning experiments ⓘ interactive manipulation tasks ⓘ multi-scene training ⓘ multi-task learning ⓘ reinforcement learning experiments ⓘ visual navigation research ⓘ |
| supportsIntegrationWith |
OpenAI Gym-style interfaces
ⓘ
reinforcement learning libraries ⓘ |
| supportsPlatform |
Linux
ⓘ
Windows ⓘ macOS ⓘ |
| usedFor |
evaluating embodied agents
ⓘ
navigation tasks ⓘ object manipulation tasks ⓘ object search tasks ⓘ robotics simulation ⓘ task planning research ⓘ training embodied agents ⓘ vision-and-language tasks ⓘ |
Referenced by (1)
Full triples — surface form annotated when it differs from this entity's canonical label.