AWhat changed
An image can show a closed refrigerator, but it cannot reveal what changes when a robot opens it. AI2-THOR connects a simulated indoor world to a Python controller: an agent requests an action, the world applies its rules, and the agent receives a new image and structured information. The 2017 account describes 120 rooms across four household settings, with objects that can be opened, picked up or otherwise manipulated. This makes sequences of perception and action repeatable without requiring a physical robot for every experiment. Its contribution here is the documented simulation resource and interface, separate from the earlier image-target navigation algorithm developed with the platform.
Scientific scope
An image can show a closed refrigerator, but it cannot reveal what changes when a robot opens it. AI2-THOR connects a simulated indoor world to a Python controller: an agent requests an action, the world applies its rules, and the agent receives a new image and structured information. The 2017 account describes 120 rooms across four household settings, with objects that can be opened, picked up or otherwise manipulated. This makes sequences of perception and action repeatable without requiring a physical robot for every experiment. Its contribution here is the documented simulation resource and interface, separate from the earlier image-target navigation algorithm developed with the platform. The room and object counts describe the cited 2017 version. Camera visibility, distance limits and object-action preconditions restrict what the agent can do; simulator metadata can expose information unavailable in an ordinary image. Reported execution speeds depend on hardware and settings. Simulation realism does not establish reliable transfer to arbitrary real homes. This resource record does not award a second copy of the navigation algorithm, Unity physics or subsequent AI2-THOR extensions.
Evidence
The linked original works support the scoped statement and contribution attribution. WRS has not independently replayed experiments or reconstructed a full mathematical proof. The room and object counts describe the cited 2017 version. Camera visibility, distance limits and object-action preconditions restrict what the agent can do; simulator metadata can expose information unavailable in an ordinary image. Reported execution speeds depend on hardware and settings. Simulation realism does not establish reliable transfer to arbitrary real homes. This resource record does not award a second copy of the navigation algorithm, Unity physics or subsequent AI2-THOR extensions.
Contributors
- The named authors in the research record · SCIENCE: Source-listed human researchers and engineers for the exact method or resource. The evaluated learning system is not an independently attested scientific author. Earlier methods, funding and data-collection roles remain separate. Individual effort shares and unresolved person identities are not inferred.
Provisional WRS assessment, edition 2026-10-23.6. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.
Full scientific record · Browse contribution grades · Request a correction