Get Started
Home
Topics
Search
Library
Research questionHow can domestic robots export task-useful perception without leaking household semantics, geometry, or goals?Keeping raw sensor data inside the home does not prevent exported structured representations from revealing household content. Semantics, geometry, spatial structure, and task targets may remain inferable even when the robot retains rich perception locally.
Alignment & Safety
Computer Vision
Evaluation & Benchmarks
Robotics
Latest papersRecent research connected to this question, newest first.Seeing Less Is Not Seeing Safely: Privacy Leakage from Task-Scoped Robot Perception ExportsThe evidence concerns structured exports from domestic-robot perception to planners, cloud services, logs, or learning pipelines. It covers navigation, collision checking, and object-goal execution in 120 AI2-THOR scenes, with a ProcTHOR replication and held-out representation-level attacks; the findings do not establish a universal privacy ordering among abstractions.research paper · Sep 2, 2026
Related questions
How can robots maintain geometrically grounded semantic maps as objects and spatial relations change in real time?How can mobile robots understand people and surroundings well enough to navigate safely and collaborate adaptively?How can robot perception encode action-relevant scene dynamics to improve manipulation generalization?How can robots learn robust, generalist bimanual household manipulation from limited human demonstrations?