ALAR: Artificial Life and Adaptive Robotic Lab
Computer Science Faculty, UNSW@ADFA
Canberra, Australia
MiroSot Robot Soccer Team
Introduction
Robot soccer is a domain that provides a challenging and competitive atmosphere for making advances in “AI” research. The game is highly dynamic and team oriented which puts an enormous strain on control algorithms, which must utilise teamwork and make decisions based on relatively slow and imprecise information in order to score goals, whilst preventing the opposition scoring goals against them.
This domain is one that is geared towards pushing both hardware and software boundaries, but the focus here is with the software. In the small robot teams, the robots are controlled via a host computer mainly dedicated to vision processing. In these leagues, successful software control of the robots is most important, as differences in hardware have only a minor effect on the overall outcome of a game. With these smaller robots and also the simulation games, winning is never decided by the cost of its hardware, it is decided on the successfulness of the soccer playing strategies, meaning the contest is accessible and even winnable at relatively low cost compared to some of the more sophisticated hardware solutions.
The Players
The School of Computer Science at ADFA has 2 complete teams of that conform to FIRA’s MiroSot (small league) rules. Each side consists of 3 robots. They are cubic with a side length of approximately 7.5 cm. Movement is controlled by adjusting the speed of each wheel. Control is achieved only from the host computer setting speed values for the left and right wheels of each robot. Communication is achieved via an RF system, which transmits on the 418Mhz or 433Mhz bands. Although the rules allow robots to have manipulators, these robots do not. These robots can only manipulate the ball by colliding with it.
The Host System
The host system is a Pentium 3 - 1000Mhz system with 256mb of RAM. The camera is a colour NTSC camera, which runs at 640 x 480 @ 30hz. The frame capture card is a Matrox Meteor II. The development environment is Microsoft Visual C++ 6.
The Game
The soccer game takes place on a 150 x 130 cm field. The field is enclosed so neither the robots nor the ball can escape. The goal is 40cm in size and the ball is a standard orange golf ball. The rules are kept as similar as possible to the real game of soccer, however the game is constrained to only 2 dimensions.
The Research
The research is strongly focused on increasing knowledge in the field of computational intelligence, as opposed to just winning the game of soccer (although the latter is quite beneficial to our egos). Non-traditional methods of control are used including swarm intelligence and neural networks. These are improved with machine learning & optimisation. The bulk of the effort however, is focussed on using machine intelligence techniques based around biological learning. This means neural networks, evolutionary techniques, and swarm intelligence are core tools. The idea is that robots, like humans, can learn and adapt without “hard coding” exact solutions.
The Future
The future of robot soccer is well mapped out. The ultimate plan for Robocup is by 2050 they can submit a robot team that will defeat the human world championship team at soccer. The real future is not in robot soccer however, it is the utilisation of the research into real world problems. Robots will no longer be restricted to perfectly defined tasks in intrinsically known environments. They will be able to achieve tasks that are seen as impossible or unbelievable today.
Intelligence machines will have an increasing role in the future. The power and complexity of hardware increases daily. Software is struggling to keep up as we hand over more and more complex tasks to the computer. If programs are continued to be developed with the ideology, “computers can only do EXACTLY what you tell them to do”, then computers will become constrained in what they can do. Computers must learn to adapt to situations that have not being considered during the software design stage.
Computational intelligence is looking for these answers, however, they won’t be solved without a long term commitment to research and robot soccer provides a framework for worldwide commitment towards a problem whose answers are potentially useful for a wide range of applications.
By using one powerful machine, most tasks can be done, however, there are cases where multiple robots must be used together to achieve an aim. The cost of one powerful robot is also usually much more than if many smaller robots are used to do the same task. By using multiple robots that can work as a team, many advantages are attained, including increased success probability and more efficiency. A wider range of work tasks can be also be undertaken as the team of robots are not specialised to only one specific problem. The successful application of teamwork will be important in the future of robotics and robot soccer allows the development of these teamwork techniques, which are vital to play well in this game.
Links
The Artificial Life and Adaptive Robotic (ALAR) LAb