Learning Discrete State Abstractions With Deep Variational Inference

TL;DR

This paper introduces a deep variational inference approach to learn discrete state abstractions for large state spaces, enabling efficient planning in high-dimensional environments with image states.

Contribution

It proposes a novel end-to-end method combining neural encoders and hidden Markov models to learn discrete state abstractions from high-dimensional data.

Findings

Effective in robotic manipulation domains with image states

Outperforms previous bisimulation methods in grid-world environments

Enables planning for unseen goals using learned abstractions

Abstract

Abstraction is crucial for effective sequential decision making in domains with large state spaces. In this work, we propose an information bottleneck method for learning approximate bisimulations, a type of state abstraction. We use a deep neural encoder to map states onto continuous embeddings. We map these embeddings onto a discrete representation using an action-conditioned hidden Markov model, which is trained end-to-end with the neural network. Our method is suited for environments with high-dimensional states and learns from a stream of experience collected by an agent acting in a Markov decision process. Through this learned discrete abstract model, we can efficiently plan for unseen goals in a multi-goal Reinforcement Learning setting. We test our method in simplified robotic manipulation domains with image states. We also compare it against previous model-based approaches to finding bisimulations in discrete grid-world-like environments. Source code is available at https://github.com/ondrejba/discrete_abstractions.