We first set of parameters controls target selection, kiting,

We use genetic algorithms to evolve control tactics for groups made up of two types of units in real-time strategy games. The space of possible tactics consists of two sets of parameters. The first set of parameters controls target selection, kiting, and fleeing in a  control algorithm while the second set of parameters specifies potential fields directing unit movement and an influence map specifying attack location. Results indicate that the genetic algorithm was able to evolve control tactics that performed better against a baseline control algorithm and against random values of these parameters.Artificial intelligence(AI) has a long history. In the past, scientists used programs to amplify their ideas in checkers and chess. In 1951, the earliest working AI programs were written to run on the Ferranti Mark 1 machine of the University of Manchester – a checkers-playing program written by Christopher  Strachey and a chess-playing program written by Dietrich Prinz. In 1956, the name artificial intelligence is used for the first time. In 1990s, John Laird started his research in video games. More and more researchers have put their efforts in the area of AI since Michael Buro introduced A New AI Research Challenge~cite{buro2003real}. The main advance over the past sixty years have been advanced in search algorithms which are widely used in games. Especially many studies have been done at the StarCraft AI Competition in Artificial Intelligence and Interactive Digital Entertainment (AIIDE, held since $2010$) and Computational Intelligence and Games (CIG, held since $2011$)~cite{ontanon2013survey}. Real-time strategy (RTS) is a time-based video game that mainly about using resources to build units and defeat opponents. In real-time strategy games, players must attempt to build their resources, defend their bases and launch attacks while knowing that the opponent is trying to do the same things. Real-time strategy games introduce new pressures into strategy and war games because they require players to make quick decisions as far as how to use resources and time attacks. The StarCraft community classifies the RTS games’ tasks as two main categories: micro which is the ability to control your units individually, in order to maximize the units’ value, and macro which as a gaming technique is an application of economic theory.  M. ?ertický used Answer Set Programming (ASP) to deal with early aggression~cite{certicky2013implementing} by blocking the entrance of friend base.In this thesis, we will focus on micro. A player with good micro usually can beat a player with bad micro if they have identical units. A player with good macro typically either has a large army or high technology, and the player may also have a good defense. However, micro itself is too complicated, so we simplified the scenario in our previous work. We used Meta-Search to control a type of ranged unit called a Vulture to fight against a type of melee unit called a Zealot~cite{liucomparing}. As the results show, the Vultures with kiting ability can execute a near-perfect performance against the Zealots with a hard-coded baseline AI which only has the ability to chase the closest enemy. Since one type of unit against one type of unit is not common in games, so we extend our previous work to control a group with two types of units fighting against an identical heterogeneous group in our FastEcslent~cite{liucomparing} simulation in Figure~
ef{gameScene}. In this paper, we doubled the length of the string representation which is a set of micro algorithms’ parameters, the search space for our genetic algorithms is $2^{102}$. Our results show that the GA can consistently find a decent fitness which is $85.44$\% of the theoretical maximum fitness. The future work is we want to find a robust micro in many different positions so that we can generalize our method to the battle between multiple types of units.