Humans have developed an intriguing ability to understand how the world will behave around them. This will work regardless if it is a common situation experienced multiple times, such as pushing a glass past the edge of a table, or a new unprecedented one - driving an interplanetary rover through a sand dune. We have evolved to anticipate and predict the changes in the world and respond reactively.
Thus, there is interest to investigate the cognitive underpinnings for using approximate physical simulations as an inference tool to reason about different properties of the scene. Those methods are tested in order to examine their applicability combined with robotic manipulation to understand the mechanics of the observations in the world.