Inferring mass in complex scenes by mental simulation.

By Jessica B. Hamrick, Peter W. Battaglia, T. Griffiths, Josh Tenenbaum. 2016

103 citations recorded by the source corpus.

After observing a collision between two boxes, you can immediately tell which is empty and which is full of books based on how the boxes moved. People form rich perceptions about the physical properties of objects from their interactions, an ability that plays a crucial role in learning about the physical world through our experiences. Here, we present three experiments that demonstrate people's capacity to reason about the relative masses of objects in naturalistic 3D scenes. We find that people make accurate inferences, and that they continue to fine-tune their beliefs over time. To explain our results, we propose a cognitive model that combines Bayesian inference with approximate knowledge of Newtonian physics by estimating probabilities from noisy physical simulations. We find that this model accurately predicts judgments from our experiments, suggesting that the same simulation mechanism underlies both peoples' predictions and inferences about the physical world around them.

Subfields: Cognitive Neuroscience, Social Psychology

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Cognitive Neuroscience

In Social Psychology

Citations in the corpus

Visual Interaction Networks: Learning a Physics Simulator from Video (2017, background): “Overcoming this challenge will help close the gap between human and machine performance on important classes of behavior that depend on physical reasoning, such as model-based decision-making [3], physical inference [13], and counterfactual reasoning [10, 11].”
The Tools Challenge: Rapid Trial-and-Error Learning in Physical Problem Solving (2019, methodology): “…tuning learns better estimates of object and world properties (such as gravity or density) from action observations, using Bayesian inference to match simulated outcomes to observations (similar to Hamrick, Battaglia, Grifﬁths, & Tenenbaum, 2016; Yildirim, Smith, Belledonne, Wu, & Tenenbaum, 2018).”
ShapeStacks: Learning Vision-Based Physical Intuition for Generalised Object Stacking (2018, background): “Humans are very apt at predicting structural stability [10], inferring relative masses [2] and extrapolating trajectories of moving objects [1].”
Interpretable Intuitive Physics Model (2018, background): “Humans demonstrate the profound ability to understand the underlying physics of the world [9,10] and use it to predict the future.”
Visual Grounding of Learned Physical Models (2020, background): “…in Figure 1, humans can automatically identify the separation between liquid (water) and solid (ﬂoating cube), estimate their properties such as gravity, density, and viscosity, and predict key features of their future motion through mental simulation (Battaglia et al., 2013; Hamrick et al., 2016).”
Learning A Physical Long-term Predictor (2017, background): “However, the nature of these mental models remains unclear and is being actively investigated (Hamrick et al., 2016).”
Model Identification via Physics Engines for Improved Policy Search (2017, background): “Several cognitive models that combine such Bayesian inference with approximate knowledge of Newtonian physics have been proposed recently [22]–[24].”