Graph networks as learnable physics engines for inference and control

TL;DR

This paper introduces graph network-based learnable models that serve as physics engines, enabling accurate predictions, system identification, and planning in complex dynamical systems, advancing toward human-like world understanding.

Contribution

The paper presents a novel graph network architecture that acts as a learnable physics engine supporting prediction, inference, and planning for diverse physical systems.

Findings

Supports accurate predictions across eight physical systems

Enables system identification through inference model

Facilitates online and offline planning with differentiable models

Abstract

Understanding and interacting with everyday physical scenes requires rich knowledge about the structure of the world, represented either implicitly in a value or policy function, or explicitly in a transition model. Here we introduce a new class of learnable models--based on graph networks--which implement an inductive bias for object- and relation-centric representations of complex, dynamical systems. Our results show that as a forward model, our approach supports accurate predictions from real and simulated data, and surprisingly strong and efficient generalization, across eight distinct physical systems which we varied parametrically and structurally. We also found that our inference model can perform system identification. Our models are also differentiable, and support online planning via gradient-based trajectory optimization, as well as offline policy optimization. Our framework offers new opportunities for harnessing and exploiting rich knowledge about the world, and takes a key step toward building machines with more human-like representations of the world.