Modeling human intuitions about liquid flow with particle-based simulation

Humans can easily describe, imagine, and, crucially, predict a wide variety of behaviors of liquids—splashing, squirting, gushing, sloshing, soaking, dripping, draining, trickling, pooling, and pouring—despite tremendous variability in their material and dynamical properties. Here we propose and test a computational model of how people perceive and predict these liquid dynamics, based on coarse approximate simulations of fluids as collections of interacting particles. Our model is analogous to a “game engine in the head”, drawing on techniques for interactive simulations (as in video games) that optimize for efficiency and natural appearance rather than physical accuracy. In two behavioral experiments, we found that the model accurately captured people’s predictions about how liquids flow among complex solid obstacles, and was significantly better than several alternatives based on simple heuristics and deep neural networks. Our model was also able to explain how people’s predictions varied as a function of the liquids’ properties (e.g., viscosity and stickiness). Together, the model and empirical results extend the recent proposal that human physical scene understanding for the dynamics of rigid, solid objects can be supported by approximate probabilistic simulation, to the more complex and unexplored domain of fluid dynamics.Author summary: Although most people struggle to learn physics in school, every human brain is a remarkable “intuitive physicist” when it comes to the quick, unconscious judgments we make in interacting with the world. Almost effortlessly, and with surprisingly high quantitative accuracy, we can judge when a plate placed near the edge of a table might be at risk of falling, or how far a partially filled glass of water can be tipped before the water is in danger of spilling. What kinds of computations in the brain support these abilities? We suggest an answer based on probabilistic inference operating over particle-based simulations, the same class of approximation methods used in video games to simulate convincing real-time interactions between objects in a virtual environment. This hypothesis can potentially account for people’s quantitative, graded judgments in diverse and novel situations including a wide array of materials and physical properties, without positing a large number of separate systems or heuristics. Here, we build on previous evidence that a system of approximate probabilistic simulation supports judgments about rigid objects (e.g. judging the stability of towers of blocks, as in the game Jenga), and ask whether people can also make systematic and accurate predictions about flowing and splashing liquids, such as water or honey. We show that it is possible to capture people’s quantitative predictions using a computational model that approximates the true underlying fluid dynamics to varying degrees of coarseness, and find that people’s responses are most consistent with a very coarse approximation; while typical engineering applications might use tens or hundreds of thousands of particles to simulate a fluid, the brain might get by with roughly a hundred particles. Furthermore, we find that people consistently underestimate the potential energy of a splashing liquid in our virtual scenes, and that our model captures this behavior.