A History Tree Heuristic to Generate Better Initiation Sets for Options in Reinforcement Learning

Cilden, Erkin
Polat, Faruk
Options framework is a prominent way to improve learning speed by means of temporally extended actions, called options. Although various attempts focusing on how to derive high quality termination conditions for options exist, the impact of initiation set generation of an option is relatively unexplored. In this work, we propose an effective heuristic method to derive useful initiation set elements via an analysis of the recent history of events.