A Multi-Objective Deep Reinforcement Learning Framework

  • 2018-03-08 04:50:21
  • Thanh Thi Nguyen
  • 3

Abstract

This paper presents a new multi-objective deep reinforcement learning (MODRL)framework based on deep Q-networks. We propose linear and non-linear methods todevelop the MODRL framework that includes both single-policy and multi-policystrategies. The experimental results on a deep sea treasure environmentindicate that the proposed approach is able to converge to the optimal Paretosolutions. The proposed framework is generic, which allows implementation ofdifferent deep reinforcement learning algorithms in various complexenvironments. Details of the framework implementation can be referred tohttp://www.deakin.edu.au/~thanhthi/drl.htm.

 

Quick Read (beta)

loading the full paper ...