Bound Action Policy for Better Sample Efficiency

DOI

10.2991/ncce-18.2018.131How to use a DOI?

Keywords

Reinforcement; policy gradient; action output.; locomotion policy; Gaussian distribution.

Abstract

Reinforcement learning algorithm for solving robotic locomotion control problem has achieved great progress. Use a Gaussian distribution to represent the locomotion policy of the robot is a general way. A locomotion policy means the distribution of action output. However, in real-world control problems, the actions are bounded by physical constraints, which introduces a bias when Gaussian distribution is used as the policy. This paper proposes logistic gaussian policy, can reduce both the bias introducing by Gaussian distribution and the variance between policy gradient samples.

Copyright

© 2018, the Authors. Published by Atlantis Press.

Open Access

This is an open access article distributed under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/).

Download article (PDF)

TY  - CONF
AU  - Junning Huang
AU  - Zhifeng Hao
PY  - 2018/05
DA  - 2018/05
TI  - Bound Action Policy for Better Sample Efficiency
BT  - Proceedings of the 2018 International Conference on Network, Communication, Computer Engineering (NCCE 2018)
PB  - Atlantis Press
SP  - 794
EP  - 799
SN  - 1951-6851
UR  - https://doi.org/10.2991/ncce-18.2018.131
DO  - 10.2991/ncce-18.2018.131
ID  - Huang2018/05
ER  -