A UCT agent for Tron: Initial investigations

A UCT agent for Tron: Initial investigations
复制标题

Tron 的 UCT 代理：初步调查

DOI：
--
复制
发表时间：
2010
期刊：
Proceedings of the 2010 IEEE Conference on Computational Intelligence and Games
影响因子：
--
通讯作者：
S. Lucas
S. Lucas
中科院分区：
--
文献类型：
--
作者：
Spyridon Samothrakis;David Robles;S. Lucas

文献摘要

被引文献

相似文献

蒙特卡洛树搜索（MCTS）在人工智能领域引起了极大的兴奋。这主要归功于它在围棋上的成功。在本文中，我们测试这种方法在Tron，同时移动两个代理游戏。虽然创建的代理能够发挥良好的标准，但在相同的场景中，他们的决策存在一定程度的随机性。这表明MCTS的成功在很大程度上取决于每个单独游戏对蒙特-卡罗模拟的适用性。
Monte Carlo Tree Search(MCTS) has generated a great deal of excitement in the A.I. community, mainly due to its success in Go. In this paper we test this approach in Tron, a simultaneous move two-agent game. Although the agents created are able to play to a good standard, there is a degree of randomness in their decisions in identical scenarios. This suggests that the success of MCTS is heavily dependent on the suitability of each individual game for Monte-Carlo Simulations.