A Compositional Object-Based Approach to Learning Physical Dynamics

TL;DR

The paper introduces the Neural Physics Engine (NPE), a neural network framework that models physical interactions with compositional object-based representations, enabling better generalization and property inference in simulated physics scenarios.

Contribution

It proposes a novel neural architecture that factorizes physical scenes into object interactions, improving generalization and interpretability over less structured models.

Findings

NPE outperforms less structured architectures in predicting object movement.

NPE generalizes across different object counts and scene configurations.

NPE can infer latent object properties like mass.

Abstract

We present the Neural Physics Engine (NPE), a framework for learning simulators of intuitive physics that naturally generalize across variable object count and different scene configurations. We propose a factorization of a physical scene into composable object-based representations and a neural network architecture whose compositional structure factorizes object dynamics into pairwise interactions. Like a symbolic physics engine, the NPE is endowed with generic notions of objects and their interactions; realized as a neural network, it can be trained via stochastic gradient descent to adapt to specific object properties and dynamics of different worlds. We evaluate the efficacy of our approach on simple rigid body dynamics in two-dimensional worlds. By comparing to less structured architectures, we show that the NPE's compositional representation of the structure in physical interactions improves its ability to predict movement, generalize across variable object count and different scene configurations, and infer latent properties of objects such as mass.