Please use this identifier to cite or link to this item:
|Title: ||Applying Hybrid Learning Approach to RoboCup's Strategy|
|Authors: ||Jong Yih Kuo;Fu-Chu Huang;Shang-Pin Ma;Yong-Yi Fanjiang|
|Contributors: ||NTOU:Department of Computer Science and Engineering|
|Keywords: ||RoboCup;Case-based reasoning;Strategy;Genetic algorithm|
|Issue Date: ||2015-05-19T06:34:55Z
|Publisher: ||Journal of Systems and Software (JSS)|
|Abstract: ||RoboCup (Robot world cup tournament) soccer game is a competitive game that has become a popular research domain in recent years since it involves a complex system for the behavior of multiple agents. In this paper, a hybrid approach, case-based reasoning genetic algorithm (CBR-GA) is applied to the soccer game for providing better strategies. By using CBR-GA, the soccer robots can obtain the suitable strategies for different conditions and store the related experiences, which may be reused in the future. Rule-based reasoning (RBR) will be employed to create a new strategy for the soccer robots when CBR-GA cannot provide a suitable one. A multi-agent learning system, constructed by combining case-based reasoning genetic algorithm with RBR strategy (CGRS), is implemented on the latest WrightEagle simulation platform that is released in 2011. In the CGRS system, two kinds of agent, namely “coach agent” and “movement agent”, are designed for the soccer game. The coach agent is responsible for deciding on the strategy goal and assigning tasks to the movement agents. Every movement agent will then execute its respective task for achieving the strategy goal. Better basic skills will facilitate the movement agents to execute more effectively the assigned tasks or plans; hence, many basic skills are designed for training the movement agents. To increase learning efficiency, the strategy cycle time is reduced with a suitable case base. To validate the effectiveness of the proposed approach, our soccer team played with the WrightEagle soccer team which has remained in the top two positions in simulation 2d in recent years. Our team gradually gets higher winning frequency in 50 rounds. Furthermore, a comparison experiment shows that the proposed approach has higher winning frequency than other methods including CBR-GA, CBR-RBR and RBR. Finally, the proposed approach is also found to have better learning mechanisms than other learning approaches in soccer game.|
|Relation: ||86(7), pp.1933-1944|
|Appears in Collections:||[資訊工程學系] 期刊論文|
Files in This Item:
There are no files associated with this item.
All items in NTOUR are protected by copyright, with all rights reserved.