Co-Evolutionary Robot Soccer Show
SHORT EDUCATIONAL DOCUMENT
|
||||||||||||
|
This document tries to explain the principles underlying the co-evolutionary robot soccer show, and what kind of things might be learnt from playing with this game. Especially, the educational aspects regarding Darwinian evolution will be explained along with the aspects regarding technology (robotics) and modern artificial intelligence. It is believed that this game might help in understanding of the important biological concept of Darwinian evolution, since it gives the user hands-on experience with the concept, and hopefully makes the concept less abstract for the user. Such hands-on experiments are often impossible in ``traditional" biology classes (an exception might be work with the fruit fly, drosophila, which however is a very long process), and is therefore believed to be an important supplement to the more traditional biology classes. INDEX
|
||||||||||||
|
The concept of evolution is used by the natural sciences to explain how life on earth develops. In the 19th century, Charles Darwin put forward his theory of evolution to explain the origin of species, and with today’s knowledge of genetics, we can understand how the development and differentiation of genotypes might lead to the emergence of new species. The coding in the genotype of an organism together with the influence from the surrounding environment is deciding how the organism develops. The genotype encoding is interpreted and results in the phenotype, the actual traits of the organism. However, even though the organism might change during life, it is believed that such changes are not inherited by the offspring of this organism (otherwise, we would be speaking of what is known as Lamarckian evolution), since it is the genotype that is transferred to the offspring when an organism reproduce. Darwin puts forward the principle of ``survival of the fittest", so that only the most fit organisms (e.g. the strong ones that are able to find food and mates) are able to survive, and hence pass on their genotypes to the next generations. We might say that the most fit organisms are selected for reproduction. When two organisms reproduce through sexual reproduction, their two genotypes are combined to make a new genotype: the genotype of the offspring. This happens using a biological operator named cross-over. This operator takes one part of the genotype from one parent (e.g. the father) and combines it with one part of the genotype of the other parent (e.g. the mother). So, the child gets some characteristics from the father and some from the mother. Further, the biological operator called mutation will apply. Mutation makes small changes to part of the genotype of the child, so that the child will be similar to the parents, but also slightly different because of the mutations. If there exist a whole population of organisms, then they might compete for some limited food sources, and only the most fit ones (e.g. fastest, strongest, etc.) may survive to reproduce. So a sub-part of the population will reproduce and make a new generation of organisms. These new organisms (the new generation) will also compete for survival, the best ones will live to reproduce, and make yet a new generation. And so the evolution continues to make generation after generation in what is known as the evolutionary process. It is believed that diversification might happen during the evolutionary process, and that the organisms might become more and more fit during the evolutionary process. The Co-Evolutionary Robot Soccer Show allows us to study whether this is the case, and it allows us to get hands-on experience with the concepts of evolution, cross-over, mutation, fitness, etc. by looking at how they influence the evolutionary process. This is done by setting different parameters (e.g. selection pressure, mutation rate, cross-over rate, fitness) and then allowing the evolution to run and observe the outcome. Afterwards, one is encouraged to used to experience gained from the first run to modify the parameters and run other evolutionary experiments for obtaining the best possible result. Please click on the names (e.g. elitism, mutation rate, etc.) to get short explanation of what the single parameters mean.
Co-evolutionary robotics Here, we use evolution to develop robots rather than living organisms as in nature. This approach to the development of robots is called evolutionary robotics in the research community. The approach allows the roboticist to focus on the evaluation of robots (e.g. specify what a fit robot is) rather than having to hand-code the whole control program for the robot to solve the desired task. Evolutionary robotics can be viewed as part of the modern artificial intelligence approaches to robotics which arise from the field of artificial life. These approaches also include artificial neural network controllers and reinforcement learning. All these approaches works towards the automatic development of robot controllers, where the user should be able to train a robot rather than hand-coding the robot through a traditional programming language. The Co-Evolutionary Robot Soccer Show is an instance of off-line evolutionary robotics, in which the evolution takes place in simulation before the result is transferred to the physical robot. In this case, the physical robot is a Khepera miniature robot, which is circular and measures 55mm in diameter. The robot has two independent motors connected to small wheels (one on each side of the robot), and eight infra-red sensors that can sense objects in a very short distance. Further, the robot is equipped with a simple, linear camera that can sense object in 36 degrees in front of the robot in different grey levels. The controllers developed for playing soccer in the simulator can be downloaded to the physical Khepera robot that can then play the soccer game in a physical arena with a yellow tennis ball and an opponent. For instance, downloading and physical games with the evolved robots will take place in Amsterdam during RoboCup European Championships 2000. In the Co-Evolutionary Robot Soccer Show, there are two populations (the red population and the blue population) that compete against each other, and who are dependent on each other. This is known as co-evolutionary robotics. When one population of robots gets better (e.g. scores more goals), the other gets worse (e.g. more goals are scored against it). This is like the natural phenomenon of predators and prey. If the predators starts running faster, then the prey must find a new strategy in order to survive. If the prey switch to a new strategy, then the predator must find another, new strategy as well, in order to catch the prey. And so the co-evolution can continue with one strategy after another. In our case, we hope that the populations of robot soccer players can use the co-evolution to find new robot soccer strategies.
Artificial Intelligence Landmark Project The task of robot soccer is viewed by numerous researchers as the new landmark project in artificial intelligence. Indeed, some researchers claim that robots will/should be able to win a match against the current human world champion team (e.g. Brazil) in soccer in the year 2050. In the 1950’s, computer chess was put forward as an artificial intelligence landmark project. On May 11, 1997, IBM’s chess computer called Deep Blue won the match against the grand master Kasparov. However, most artificial intelligence researchers do not view the IBM chess computer as intelligent. This is due to the fact, that the chess computer has complete information of the environment, does not work in a dynamic environment, and does not have embodiment. On the contrary, robot soccer is a dynamic game with embodied agents that have incomplete information about their environment. Therefore, numerous researchers find this project a much better platform for studying and trying to understand intelligence. The Co-Evolutionary Robot Soccer Show is part of this challenge, and especially part of RoboCup Jr. RoboCup Jr. aims at providing knowledge and hands-on experience regarding the new robotics and artificial intelligence developments to the youth.
Evolutionary Parameters
- Number of generations:
the number of generations to run the evolution.
|
||||||||||||
HOME
|