MCTS蒙地卡羅樹於模擬賽車之應用

Application of Monte-Carlo Tree Search on Simulation Car Racing

侯嘉豪

指導教授 ： 王才沛

國立交通大學/資訊學院/多媒體工程研究所/碩士(2016年)

摘要

本篇論文是實做在TORCS 平台上，以蒙地卡羅樹搜尋作為尋找最佳賽車路線的核心演算法，製作賽車AI。蒙地卡羅樹搜尋演算法大致上分為四個部分：選點、展開、模擬和更新，文中會介紹如何把蒙地卡羅樹搜尋運用在模擬賽車的AI上，分別為各個步驟量身打造。賽車行進時，程式會在一定時間內自行模擬大量賽車路線，並透過時間差比較優劣，從中找出較佳的賽車路線來當作實際行走的依據。最後透過TORCS內建的AI比較其結果，利用一個比較基礎的AI證明可行性，再與另一個高等純熟的AI比較其差異性。

關鍵字

蒙地卡羅 ； 模擬賽車

並列摘要

This paper is concerning about how to find the best route of the car racing game on TORCS by Monte-Carlo Tree Search, an algorithm includes four parts: choice of site, expansion, simulation and update. While simulating the car racing AI, the program will generate a great number of routes and find out one way through the difference in time to work in reality. Finally, compare the results between two AI to choose which is more feasibility.

並列關鍵字

Monte-Carlo Tree Search ； simulation racing

參考文獻

[3] “Evolving the Optimal Racing Line in a High-End Racing Game,” 2012 IEEE Conference on Computational Intelligence and Games

[7] F. Braghin, F. Cheli, S. Melzi, and E. Sabbioni. Race driver model. Comput. Struct., 86(13-14):1503–1516, 2008.

[13] “An Antilock-Braking Systems (ABS) Control: A Technical Review,” Intelligent Control and Automation, 2011, 2, 186-195

[14] D. Loiacono, L. Cardamone, and P. L. Lanzi, “Simulated Car Racing Championship Competition Software Manual", April 2013.

[1] “The open racing car simultaor website, ” [Online] http://torcs.sourceforge.net/