Unsupervised Learning of Latent Physical Properties Using Perception-Prediction Networks

TL;DR

This paper introduces an unsupervised framework called perception-prediction networks (PPN) that learns latent physical properties of objects from their interactions, enabling accurate dynamics simulation and interpretable property extraction without labeled data.

Contribution

The paper presents a novel end-to-end unsupervised learning approach using perception-prediction networks that generalize across systems and translate learned representations into human-interpretable properties.

Findings

PPNs accurately simulate dynamics of unseen objects

PPNs can extract human-interpretable physical properties

PPNs generalize to different numbers of objects and scenarios

Abstract

We propose a framework for the completely unsupervised learning of latent object properties from their interactions: the perception-prediction network (PPN). Consisting of a perception module that extracts representations of latent object properties and a prediction module that uses those extracted properties to simulate system dynamics, the PPN can be trained in an end-to-end fashion purely from samples of object dynamics. The representations of latent object properties learned by PPNs not only are sufficient to accurately simulate the dynamics of systems comprised of previously unseen objects, but also can be translated directly into human-interpretable properties (e.g., mass, coefficient of restitution) in an entirely unsupervised manner. Crucially, PPNs also generalize to novel scenarios: their gradient-based training can be applied to many dynamical systems and their graph-based structure functions over systems comprised of different numbers of objects. Our results demonstrate the efficacy of graph-based neural architectures in object-centric inference and prediction tasks, and our model has the potential to discover relevant object properties in systems that are not yet well understood.