Learning Geometric Representations of Objects via Interaction

TL;DR

This paper introduces a framework for learning geometric object representations from agent-object interactions using only action-based supervision, enabling accurate localization and improved downstream task performance.

Contribution

It provides a theoretical foundation for isometric representation learning from interaction data, outperforming vision-based methods and facilitating reinforcement learning tasks.

Findings

Outperforms vision-based keypoint extractors in localization accuracy.

Guarantees isometric and disentangled representations of agent and object.

Enables efficient reinforcement learning for downstream tasks.

Abstract

We address the problem of learning representations from observations of a scene involving an agent and an external object the agent interacts with. To this end, we propose a representation learning framework extracting the location in physical space of both the agent and the object from unstructured observations of arbitrary nature. Our framework relies on the actions performed by the agent as the only source of supervision, while assuming that the object is displaced by the agent via unknown dynamics. We provide a theoretical foundation and formally prove that an ideal learner is guaranteed to infer an isometric representation, disentangling the agent from the object and correctly extracting their locations. We evaluate empirically our framework on a variety of scenarios, showing that it outperforms vision-based approaches such as a state-of-the-art keypoint extractor. We moreover demonstrate how the extracted representations enable the agent to solve downstream tasks via reinforcement learning in an efficient manner.