Stochastic Latent Actor-Critic: Deep Reinforcement Learning with a Latent Variable Model

TL;DR

SLAC is a novel reinforcement learning algorithm that learns compact latent representations from high-dimensional images, enabling efficient and high-performing control in complex tasks by combining stochastic models with RL.

Contribution

The paper introduces SLAC, a new method that unifies stochastic latent models with reinforcement learning to improve sample efficiency and performance in image-based control tasks.

Findings

SLAC outperforms existing methods in complex control tasks.

It achieves higher sample efficiency in learning from images.

The approach effectively combines stochastic models with RL.

Abstract

Deep reinforcement learning (RL) algorithms can use high-capacity deep networks to learn directly from image observations. However, these high-dimensional observation spaces present a number of challenges in practice, since the policy must now solve two problems: representation learning and task learning. In this work, we tackle these two problems separately, by explicitly learning latent representations that can accelerate reinforcement learning from images. We propose the stochastic latent actor-critic (SLAC) algorithm: a sample-efficient and high-performing RL algorithm for learning policies for complex continuous control tasks directly from high-dimensional image inputs. SLAC provides a novel and principled approach for unifying stochastic sequential models and RL into a single method, by learning a compact latent representation and then performing RL in the model's learned latent space. Our experimental evaluation demonstrates that our method outperforms both model-free and model-based alternatives in terms of final performance and sample efficiency, on a range of difficult image-based control tasks. Our code and videos of our results are available at our website.