Learning With Options That Terminate Off-Policy

Authors

  • Anna Harutyunyan Vrije Universiteit Brussel
  • Peter Vrancx PROWLER.io
  • Pierre-Luc Bacon McGill University
  • Doina Precup McGill University
  • Ann Nowé Vrije Universiteit Brussel

Keywords:

Reinforcement Learning

Abstract

A temporally abstract action, or an option, is specified by a policy and a termination condition: the policy guides the option behavior, and the termination condition roughly determines its length. Generally, learning with longer options (like learning with multi-step returns) is known to be more efficient. However, if the option set for the task is not ideal, and cannot express the primitive optimal policy well, shorter options offer more flexibility and can yield a better solution. Thus, the termination condition puts learning efficiency at odds with solution quality. We propose to resolve this dilemma by decoupling the behavior and target terminations, just like it is done with policies in off-policy learning. To this end, we give a new algorithm, Q(beta), that learns the solution with respect to any termination condition, regardless of how the options actually terminate. We derive Q(beta) by casting learning with options into a common framework with well-studied multi-step off policy learning. We validate our algorithm empirically, and show that it holds up to its motivating claims.

Downloads

Published

2018-04-29

How to Cite

Harutyunyan, A., Vrancx, P., Bacon, P.-L., Precup, D., & Nowé, A. (2018). Learning With Options That Terminate Off-Policy. Proceedings of the AAAI Conference on Artificial Intelligence, 32(1). Retrieved from https://ojs.aaai.org/index.php/AAAI/article/view/11740