REASONING_ABOUT_PHYSICAL_INTERACTIONS_WITH_OBJECT_ORIENTED_PREDICTION_AND_PLANNING
-
3 separate components, perception, physics and rendering are jointly trained.
-
Assume availability of the segmentation mask for object in a scene.
-
The trained module can successfully be used to predict action sequence for block stacking.
A_Compositional_Object_Based_Approach_to_Learning_Physical_Dynamics