Policy Gradients Incorporating the Future

TL;DR

This paper introduces PGIF, a novel reinforcement learning method that enables agents to utilize future trajectory information during training without explicit future prediction, improving learning efficiency in complex environments.

Contribution

We propose PGIF, a versatile policy gradient method that incorporates future trajectory information during training, enhancing performance across various RL settings.

Findings

PGIF achieves higher rewards faster in multiple RL domains.

The method improves learning efficiency in sparse-reward environments.

PGIF is compatible with existing policy gradient algorithms.

Abstract

Reasoning about the future -- understanding how decisions in the present time affect outcomes in the future -- is one of the central challenges for reinforcement learning (RL), especially in highly-stochastic or partially observable environments. While predicting the future directly is hard, in this work we introduce a method that allows an agent to "look into the future" without explicitly predicting it. Namely, we propose to allow an agent, during its training on past experience, to observe what \emph{actually} happened in the future at that time, while enforcing an information bottleneck to avoid the agent overly relying on this privileged information. This gives our agent the opportunity to utilize rich and useful information about the future trajectory dynamics in addition to the present. Our method, Policy Gradients Incorporating the Future (PGIF), is easy to implement and versatile, being applicable to virtually any policy gradient algorithm. We apply our proposed method to a number of off-the-shelf RL algorithms and show that PGIF is able to achieve higher reward faster in a variety of online and offline RL domains, as well as sparse-reward and partially observable environments.