Abstract—This paper presents a genetic algorithmic approach to the shortest path (SP) routing problem. Variable-length chromosomes (strings) and their genes (parameters) have been used for encoding the problem. The crossover operation exchanges partial chromosomes (partial routes) at positionally independent crossing sites and the mutation operation maintains the genetic diversity of the population. The proposed algorithm can cure all the infeasible chromosomes with a simple repair function. Crossover and mutation together provide a search capability that results in improved quality of solution and enhanced rate of convergence. This paper also develops a population-sizing equation that facilitates a solution with desired quality. It is based on the gambler’s ruin model; the equation has been further enhanced and generalized, however. The equation relates the size of the population, the quality of solution, the cardinality of the alphabet, and other parameters of the proposed algorithm. Computer simulations show that the proposed algorithm exhibits a much better quality of solution (route optimality) and a much higher rate of convergence than other algorithms. The results are relatively independent of problem types (network sizes and topologies) for almost all source–destination pairs. Furthermore, simulation studies emphasize the usefulness of the population-sizing equation. The equation scales to larger networks. It is felt that it can be used for determining an adequate population size (for a desired quality of solution) in the SP routing problem. Index Terms—Gambler’s ruin model, genetic algorithms, population size, shortest path routing problem. I.

Ms. Pac-Man is a challenging, classic arcade game with a certain cult status. This paper reports attempts to evolve a Pac-Man player, where the control algorithm uses a neural network to evaluate the possible next moves. The evolved neural network takes a handcrafted feature vector based on a candidate maze location as input, and produces a score for that location as output. Results are reported on two simulated versions of the game: deterministic and nondeterministic. The results show that useful behaviours can be evolved that are frequently capable of clearing the first level, but are still susceptible to making poor decisions. Currently, the best evolved players play at the level of a reasonable human novice.

In this work, we describe the evolutionary training of artificial neural network controllers for competitive team game playing behaviors by teams of real mobile robots. This research emphasized the development of methods to automate the production of behavioral robot controllers. We seek methods that do not require a human designer to define specific intermediate behaviors for a complex robot task. The work made use of a real mobile robot colony (EVolutionary roBOTs) and a closely coupled computer-based simulated training environment. The acquisition of behavior in an evolutionary robotics system was demonstrated using a robotic version of the game Capture the Flag. In this game, played by two teams of competing robots, each team tries to defend its own goal while trying to `attack' another goal defended by the other team. Robot neural controllers relied entirely on processed video data for sensing of their environment. Robot controllers were evolved in a simulated environment using evolutionary training algorithms. In the evolutionary process, each generation consisted of a competitive tournament of games played between the controllers in an evolving population. Robot controllers were selected based on whether they won or lost games in the course of a tournament. Following a tournament, the neural controllers were ranked competitively according to how many games they won and the population was propagated using a mutation and replacement strategy. After several hundred generations, the best performing controllers were transferred to teams of real mobile robots, where they exhibited behaviors similar to those seen in simulation including basic navigation, the ability to distinguish between different types of objects, and goal tending behaviors.

We have been exploring the potential for a coevolutionary process to learn how to play checkers without relying on the usual inclusion of human expertise in the form of features that are believed to be important to playing well. In particular, we have focused on the use of a population of neural networks, where each network serves as an evaluation function to describe the quality of the current board position. After only a little more than 800 generations, the evolutionary process has generated a neural network that can play checkers at the expert level as designated by the U.S. Chess Federation rating system. The current effort reports on a competition between the best-evolved neural network, named "Anaconda," and commercially available software. In a series of six games, Anaconda scored a perfect six wins. 1 Introduction Checkers is played traditionally on an 8 8 board with squares of alternating colors of red and black (see Fig. 1). There are two players, denoted as "red" and "wh...

Abstract not found

Abstract. In this paper we present a novel method, called Evolutionary

Abstract not found

In this research, we describe the evolutionary training of artificial neural network controllers for competitive team game playing behaviors by teams of real mobile robots (The EvBots). During training (evolution), performance of controllers was evaluated based on the results of competitive tournaments of games played between robots (controllers) in an evolving population. Competitive tournament fitness evaluation does not require a human designer to define specific intermediate behaviors for a complex robot task. Intermediate behavior selection and evaluation becomes an implicit part of winning or losing games in a tournament. The acquisition of behavior in this evolutionary robotics system was demonstrated using a robotic version of the game `Capture the Flag'. In this game, played by two teams of competing robots, each team tries to defend its own goal while trying to `attack' another goal defended by the other team. Robot controllers were evolved in a simulated environment using evolutionary training algorithms and were then transferred to real robots in a physical environment for validation. Evolutionary robotics makes use of several distinct types or levels of performance evaluation. The work presented here focuses on the competitive relative tournament ranking metric used to drive the evolutionary process. After a population has been evolved, a second metric is needed to evaluate the quality of acquired game-playing skills. We use a post training evaluation method that compares the evolved controllers to hand coded knowledge-based controllers designed to perform the same task. In particular, a very poor controller, and high quality controller give us two points on a continuum that can be used to rank the evolved controller quality.

Abstract — This paper compares the use of temporal difference learning (TDL) versus co-evolutionary learning (CEL) for acquiring position evaluation functions for the game of Othello. The paper provides important insights into the strengths and weaknesses of each approach. The main findings are that for Othello, TDL learns much faster than CEL, but that properly tuned CEL can learn better playing strategies. For CEL, it is essential to use parent-child weighted averaging in order to achieve good performance. Using this method a high quality weighted piece counter was evolved, and was shown to significantly outperform a set of standard heuristic weights.

In this paper we investigate the evolution of a blackjack player. We utilise three neural networks (one for splitting, one for doubling down and one for standing/hitting) to evolve blackjack strategies. Initially a pool of randomly generated players play 1000 hands of blackjack. An evolutionary strategy is used to mutate the best networks (with the worst networks being killed). We compare the best evolved strategies to other well-known strategies and show that we can beat the play of an average casino player. We also show that we are able to learn parts of Thorpe's Basic Strategy.