System, method and product for utilizing prediction models of an environment

A method comprising:

• obtaining a first scene of an environment from one or more sensors, wherein the one or more sensors observe scenes of the environment, wherein the environment comprises a pile of physical objects;
• determining a stimuli force to be applied on a physical object of the pile of physical objects;
• predicting, using a model of the pile of physical objects and based on the first scene of the environment, a predicted second scene of the environment
• in response to application of the stimuli force on the physical object;
• wherein after said predicting, applying the stimuli force on the physical object;
• and
• obtaining an observed second scene from the one or more sensors after said applying;
• comparing the predicted second scene with the observed second scene;
• and
• performing an action based on differences determined in said comparing between the predicted second scene and the observed second scene, wherein said performing the action comprises updating, based on the differences between the predicted second scene and the observed second scene, the model of the pile.