ETRI and GIST introduced SceneQueryPlanner, which plans information retrieval alongside actions for embodied AI agents. It stores object locations and properties in a 3D scene graph and queries only needed information to reduce failures caused by guessing. The researchers report higher success rates than ReAct in simulation and have released implementation and evaluation data. This does not establish long-duration performance on real robots in everyday environments; separately licensed components of the included simulator also require attention.