Beta DVBF: Learning State-Space Models for Control from High Dimensional Observations

TL;DR

This paper introduces Beta DVBF, a method for learning state-space models directly from high-dimensional images, addressing the challenge of dimensionality mismatch in latent-variable models for control tasks.

Contribution

The paper proposes solutions to improve latent-variable models for high-dimensional observations, enabling effective dynamics learning from high-resolution images.

Findings

Addresses the dimensionality gap between observations and latent space.

Proposes novel methods to enhance model learning from high-dimensional data.

Improves control performance with high-resolution image-based models.

Abstract

Learning a model of dynamics from high-dimensional images can be a core ingredient for success in many applications across different domains, especially in sequential decision making. However, currently prevailing methods based on latent-variable models are limited to working with low resolution images only. In this work, we show that some of the issues with using high-dimensional observations arise from the discrepancy between the dimensionality of the latent and observable space, and propose solutions to overcome them.