Learning Policies with External Memory

Leonid Peshkin, Nicolas Meuleau, Leslie Kaelbling

In order for an agent to perform well in partially observable domains, it is usually necessary for actions to depend on the history of observations. In this paper, we explore a {\i…