Asynchronous Deep Q-network in Continuous Environment Based on Prioritized Experience Replay
Download as PDF
DOI: 10.23977/meimie.2019.43075
Author(s)
Hongda Liu, Hanqi Zhang and Linying Gong
Corresponding Author
Hongda Liu
ABSTRACT
Deep Q-network is a classical algorithm of reinforce learning, which is widely used and has many variants. The research content of this paper is to optimize and integrate some variant algorithms so that it has the advantage of running in the continuous environment, and improve the learning efficiency by Prioritized Experience Replay and multiple agents' asynchronous parallel method, and establish the asynchronous Deep Q-network framework based on priority Experience Replay in the continuous environment. This paper uses some games in the Atari 2600 domain to test our algorithm framework, which achieved good results, improved stability, convergence speed and improved performance.
KEYWORDS
Deep Q-network, Continuous Environment, Prioritized Experience Replay, Asynchronous