Offline reinforcement learning in high-dimensional stochastic environments

Offline reinforcement learning (RL) has emerged as a promising paradigm for real-world applications since it aims to train policies directly from datasets of past interactions with the environment. The past few years, algorithms have been introduced to learn from high-dimensional observational states in offline settings. The general idea of these methods is to encode the environment into a latent space and train policies on top of this smaller representation. In this paper, we extend this general method to stochastic environments (i.e., where the reward function is stochastic) and consider a risk measure instead of the classical expected return. First, we show that, under some assumptions, it is equivalent to minimizing a risk measure in the latent space and in the natural space. Based on this result, we present Latent Offline Distributional Actor-Critic (LODAC), an algorithm which is able to train policies in high-dimensional stochastic and offline settings to minimize a given risk measure. Empirically, we show that using LODAC to minimize Conditional Value-at-Risk (CVaR) outperforms previous methods in terms of CVaR and return on stochastic environments.