A policy gradient reinforcement learning algorithm with fuzzy function approximation

Dongbing Gu; Erfu Yang

doi:10.1109/ROBIO.2004.1521910

A policy gradient reinforcement learning algorithm with fuzzy function approximation

Dongbing Gu^*, Erfu Yang

^*Corresponding author for this work

Design, Manufacturing And Engineering Management

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution book

4 Citations (Scopus)

Abstract

For complex systems, reinforcement learning has to be generalised from a discrete form to a continuous form due to large state or action spaces. In this paper, the generalisation of reinforcement learning to continuous state space is investigated by using a policy gradient approach. Fuzzy logic is used as a function approximation in the generalisation. To guarantee learning convergence, a policy approximator and a state action value approximator are employed for the reinforcement learning. Both of them are based on fuzzy logic. The convergence of the learning algorithm is justified.

Original language	English
Title of host publication	IEEE International Conference on Robotics and Biomimetics, 2004. ROBIO 2004
Place of Publication	Piscataway, NJ.
Publisher	IEEE
Pages	936-940
Number of pages	5
ISBN (Print)	0780386148
DOIs	https://doi.org/10.1109/ROBIO.2004.1521910
Publication status	Published - 2004
Event	2004 IEEE International Conference on Robotics and Biomimetics, ROBIO 2004 - Shenyang, China Duration: 22 Aug 2004 → 26 Aug 2004

Conference

Conference	2004 IEEE International Conference on Robotics and Biomimetics, ROBIO 2004
Country/Territory	China
City	Shenyang
Period	22/08/04 → 26/08/04

Keywords

fuzzy Q-learning
policy gradient method
reinforcement learning
learning convergence
approximation theory
convergence of numerical methods
functions
fuzzy sets
large scale systems
learning algorithms
state space methods

Access to Document

10.1109/ROBIO.2004.1521910

Cite this

@inproceedings{3637ca12463d4ad38c878dac3930f99b,

title = "A policy gradient reinforcement learning algorithm with fuzzy function approximation",

abstract = "For complex systems, reinforcement learning has to be generalised from a discrete form to a continuous form due to large state or action spaces. In this paper, the generalisation of reinforcement learning to continuous state space is investigated by using a policy gradient approach. Fuzzy logic is used as a function approximation in the generalisation. To guarantee learning convergence, a policy approximator and a state action value approximator are employed for the reinforcement learning. Both of them are based on fuzzy logic. The convergence of the learning algorithm is justified.",

keywords = "fuzzy Q-learning, policy gradient method, reinforcement learning, learning convergence, approximation theory, convergence of numerical methods, functions, fuzzy sets, large scale systems, learning algorithms, state space methods",

author = "Dongbing Gu and Erfu Yang",

year = "2004",

doi = "10.1109/ROBIO.2004.1521910",

language = "English",

isbn = "0780386148",

pages = "936--940",

booktitle = "IEEE International Conference on Robotics and Biomimetics, 2004. ROBIO 2004",

publisher = "IEEE",

note = "2004 IEEE International Conference on Robotics and Biomimetics, ROBIO 2004 ; Conference date: 22-08-2004 Through 26-08-2004",

}

TY - GEN

T1 - A policy gradient reinforcement learning algorithm with fuzzy function approximation

AU - Gu, Dongbing

AU - Yang, Erfu

PY - 2004

Y1 - 2004

N2 - For complex systems, reinforcement learning has to be generalised from a discrete form to a continuous form due to large state or action spaces. In this paper, the generalisation of reinforcement learning to continuous state space is investigated by using a policy gradient approach. Fuzzy logic is used as a function approximation in the generalisation. To guarantee learning convergence, a policy approximator and a state action value approximator are employed for the reinforcement learning. Both of them are based on fuzzy logic. The convergence of the learning algorithm is justified.

AB - For complex systems, reinforcement learning has to be generalised from a discrete form to a continuous form due to large state or action spaces. In this paper, the generalisation of reinforcement learning to continuous state space is investigated by using a policy gradient approach. Fuzzy logic is used as a function approximation in the generalisation. To guarantee learning convergence, a policy approximator and a state action value approximator are employed for the reinforcement learning. Both of them are based on fuzzy logic. The convergence of the learning algorithm is justified.

KW - fuzzy Q-learning

KW - policy gradient method

KW - reinforcement learning

KW - learning convergence

KW - approximation theory

KW - convergence of numerical methods

KW - functions

KW - fuzzy sets

KW - large scale systems

KW - learning algorithms

KW - state space methods

UR - http://www.scopus.com/inward/record.url?scp=28344451638&partnerID=8YFLogxK

UR - http://ieeexplore.ieee.org/xpl/mostRecentIssue.jsp?punumber=10204

UR - http://www.robio.org/

U2 - 10.1109/ROBIO.2004.1521910

DO - 10.1109/ROBIO.2004.1521910

M3 - Conference contribution book

AN - SCOPUS:28344451638

SN - 0780386148

SP - 936

EP - 940

BT - IEEE International Conference on Robotics and Biomimetics, 2004. ROBIO 2004

PB - IEEE

CY - Piscataway, NJ.

T2 - 2004 IEEE International Conference on Robotics and Biomimetics, ROBIO 2004

Y2 - 22 August 2004 through 26 August 2004

ER -

A policy gradient reinforcement learning algorithm with fuzzy function approximation

Abstract

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this