Modeling human intuitions about liquid flow with particle-based simulation.

Area of Science:

• Cognitive Science
• Physics
• Computer Science

Background:

• Humans possess an intuitive understanding of fluid dynamics, enabling prediction of diverse liquid behaviors.
• This understanding is remarkable given the complex material and dynamical properties of liquids.

Purpose of the Study:

• To propose and test a computational model explaining human perception and prediction of liquid dynamics.
• To investigate if approximate simulations, similar to video game engines, underlie this human ability.

Main Methods:

• Developed a computational model simulating fluids as interacting particles, optimized for efficiency and natural appearance.
• Conducted two behavioral experiments to assess the model's accuracy against human predictions.
• Compared the model's performance against heuristic and deep neural network alternatives.

Main Results:

• The proposed model accurately predicted human judgments of liquid flow among obstacles.
• The model significantly outperformed heuristic and deep neural network-based predictions.
• The model successfully explained variations in human predictions based on liquid properties like viscosity and stickiness.

Conclusions:

• Human physical scene understanding for fluid dynamics may rely on approximate probabilistic simulations.
• This extends previous findings on solid object dynamics to the complex domain of fluids.
• The cognitive model provides a framework for understanding intuitive fluid dynamics prediction.