Goal Grounding
A core concept in embodied intelligence referring to the ability of an agent to establish correspondences between targets in natural language instructions and object instances in visual observations. It comprises three levels: object detection, recognition and localization.
Timeline (last 90 days)
StopWatch finds that object grounding failure is the main bottleneck constraining the success rate of object-goal navigation.
Object grounding refers to the agent's ability to establish correspondences between the target in a natural language instruction and specific object instances in visual observations, comprising three levels: object detection, object recognition, and object localization.
Object grounding in embodied settings needs to be performed online and continuously, and is affected by viewpoint changes, occlusion, and motion, making it significantly harder than visual localization in static images.
In embodied navigation, embedding object grounding into the policy as an explicit intermediate representation is equivalent to injecting a structural inductive bias of 'locate the target first, then decide'.
All Facts (4)
StopWatch finds that object grounding failure is the main bottleneck constraining the success rate of object-goal navigation.
50%UnverifiedObject grounding refers to the agent's ability to establish correspondences between the target in a natural language instruction and specific object instances in visual observations, comprising three levels: object detection, object recognition, and object localization.
50%UnverifiedObject grounding in embodied settings needs to be performed online and continuously, and is affected by viewpoint changes, occlusion, and motion, making it significantly harder than visual localization in static images.
50%UnverifiedIn embodied navigation, embedding object grounding into the policy as an explicit intermediate representation is equivalent to injecting a structural inductive bias of 'locate the target first, then decide'.
50%