OptionGAN: Learning Joint Reward-Policy Options using Generative Adversarial Inverse Reinforcement Learning

TL;DR

OptionGAN introduces a novel adversarial inverse reinforcement learning approach that learns joint reward and policy options from expert demonstrations, improving transfer learning in complex tasks.

Contribution

It extends the options framework to recover reward and policy options simultaneously using generative adversarial methods from observed states.

Findings

Effective in simple and complex continuous control tasks

Significant performance improvements in one-shot transfer learning

Successfully learns joint reward-policy options from expert demonstrations

Abstract

Reinforcement learning has shown promise in learning policies that can solve complex problems. However, manually specifying a good reward function can be difficult, especially for intricate tasks. Inverse reinforcement learning offers a useful paradigm to learn the underlying reward function directly from expert demonstrations. Yet in reality, the corpus of demonstrations may contain trajectories arising from a diverse set of underlying reward functions rather than a single one. Thus, in inverse reinforcement learning, it is useful to consider such a decomposition. The options framework in reinforcement learning is specifically designed to decompose policies in a similar light. We therefore extend the options framework and propose a method to simultaneously recover reward options in addition to policy options. We leverage adversarial methods to learn joint reward-policy options using only observed expert states. We show that this approach works well in both simple and complex continuous control tasks and shows significant performance increases in one-shot transfer learning.