Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
COLLABORATIVE TYPE METHOD AND SYSTEM OF MULTISTATE CONTINUOUS ACTION SPACE
Document Type and Number:
WIPO Patent Application WO/2020/024172
Kind Code:
A1
Abstract:
The present invention provides a collaborative type method and a system of a multistate continuous action space, belonging to the reinforced learning field. The method comprises the following steps: initializing an action set for states of an arbitrary state collection; initializing related parameters for states in an arbitrary state collection and actions in the action set; and constructing a corresponding collaboration mechanism in a correction layer and a strategy evaluation updating layer of the action set respectively until the return of an intellectual body i in state s is converged. The invention further provides a system for implementing the collaborative type method of multistate continuous action space. The invention has the beneficial effect as follows: the cooperative problem of multiple intellectual bodies in continuous action space can be well solved.

Inventors:
HOU HANXU (CN)
HAO JIANYE (CN)
ZHANG CHENGWEI (CN)
Application Number:
PCT/CN2018/098103
Publication Date:
February 06, 2020
Filing Date:
August 01, 2018
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
UNIV DONGGUAN TECHNOLOGY (CN)
International Classes:
G06N99/00
Foreign References:
CN107734579A2018-02-23
CN105959353A2016-09-21
US8948499B12015-02-03
Other References:
ZHANG, CHENGWEI ET AL.: "SCC-rFMQ Learning in Cooperative Markov Games with Continuous Actions", IN PROC. OF THE 17TH INTERNATIONAL CONFERENCE ON AUTONOMOUS AGENTS AND MULTIAGENT SYSTEMS (AAMAS 2018, 15 July 2018 (2018-07-15), XP055681267
YANG, YUJUN ET AL.: "Cooperation Learning for Autonomous Micro-mobile Robot", COMPUTER ENGINEERING, vol. 29, no. 10, 30 June 2003 (2003-06-30), ISSN: 1000-3428
Attorney, Agent or Firm:
SZ KINDWALF INTELLECTUAL PROPERTY FIRM (CN)
Download PDF: