In the physical world, every object has different affordances or ways in which we can interact with them. Through this interaction, we can get a sense of their inner properties and subsequently use this information to discover new ways in which we can interact with them on them or imagine how they will behave in different scenarios.
For example, when we prod, stir or shake a liquid we get information about how much it resists to the applied movement giving us cues about its viscosity. This useful reasoning skill allows us to interact with our everyday environment in a seamlessly way.
Robots on the other are traditionally programmed to perform a task for a specific scenario and are not able to generalize beyond that. In this work, we argue that it is possible to embed a similar reasoning model as the one used by humans in a general purpose robot and use it to get estimates of the latent properties of a fluid.
We used the Baxter robot to test our hypothesis and get estimations of three different liquids by incorporating physical knowledge of fluid dynamics using an exact physical model based on the working principle of a rotational viscometer.