Giacomo Aldegheri¹ (), Roland W. Fleming^1,2; ¹Department of Experimental Psychology, Justus Liebig University Giessen, ²Centre for Mind, Brain and Behaviour (CMBB), University of Marburg and Justus Liebig University, Giessen

Humans are able to successfully interact with objects with vastly different physical characteristics. For example, hard objects, such as boxes, and soft objects, such as cloth, behave in entirely different ways when manipulated. Humans might achieve such flexibility in two ways: either by having a unified internal model that is general enough to make predictions in such diverse scenarios, or a variety of domain-specific models, with the ability to switch between them. Domain-specific models (such as rigid-body and particle-based simulation) are commonly used in video game physics engines: being tailored to solve specific problems, they can make more efficient use of computational resources. Do humans also possess multiple internal models of the physical world, and the ability to switch between them? To answer this question, we designed a task to probe, in a controlled yet ecological way, how human participants interact with objects with diverse physical characteristics. Participants are shown, on a computer screen, an ‘arena’ including an object, a simple manipulator (a rectangular pusher) and a goal indicated by a shape and location within the arena. Using the keyboard, they can rotate and translate the pusher, and their task is to use it to push the object into the goal. On each trial the object belongs to one of four categories: box, rope, cloth and granular (e.g. a pile of coffee beans). The task varies widely depending on the object category: for example, a box can only be pushed into a specific position, while granular objects can be pushed into a given shape as well. We collect a large-scale online dataset of human responses on this task, and present preliminary results assessing systematic differences in behavior between object categories.

Acknowledgements: This research was funded by the European Research Council (ERC) Advanced Award ‘STUFF’ (ERC-ADG-2022- 101098225)