Object geometry serves humans' intuitive physics of stability.

Area of Science:

• Cognitive Science
• Developmental Psychology
• Physics Perception

Background:

• The prevailing theory suggests humans use mental simulation via an intuitive physics engine to judge physical stability.
• This study investigates the role of perceptual object geometry in predicting falling direction.

Purpose of the Study:

• To determine if object geometry alone is sufficient for judging falling direction.
• To compare human performance with computational models based on geometric features and deep learning.

Main Methods:

• Adults and children judged the falling direction of tilted objects with varying geometric properties.
• Performance was compared against computational models using geometric features (centroid, height, base, aspect ratio).
• A deep convolutional neural network (ResNet-50) was used as a comparative model.

Main Results:

• Human performance, for both adults and children, was accurately predicted by models based on object geometry, especially the geometric centroid.
• A deep convolutional neural network (ResNet-50) also effectively modeled human judgments.
• Findings indicate geometric features are sufficient for predicting falling direction.

Conclusions:

• Object geometry appears sufficient for judging the falling direction of tilted objects.
• This suggests that mental simulation may not be necessary for these physical stability judgments.
• Perceptual cues related to object shape play a critical role in intuitive physics.