A Compositional Object-Based Approach to Learning Physical Dynamics

TLDR

Intuitive physics simulators must generalize across varying object counts and scene configurations, a challenge addressed by symbolic engines that model generic objects and interactions. The authors introduce the Neural Physics Engine (NPE) to learn simulators that generalize across variable object counts and scene configurations. The NPE factorizes scenes into composable object-based representations, using a neural network whose compositional structure models pairwise interactions and is trained by stochastic gradient descent. Compared to less structured architectures, the NPE’s compositional representation improves prediction accuracy, generalization across object counts and scene configurations, and inference of latent properties such as 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.