The development of autonomous robot soccer players is quite important, since it requires the application of a broad spectrum of technologies, such as solving control theoretic problems or using image processing techniques. It is an excellent domain for studying the cooperation of systems in a dynamic, competitive environment. However, building a complete robot soccer team from scratch is very expensive and the researchers have to spend a lot of time and energy on the problems of actually building the physical robots, instead of concentrating on the algorithms and methods of the high-level strategy and cooperation.
It is exactly the purpose of the 2D and 3D simulation platforms, developed by the international RoboCup organization, and the world championship organised every year to solve the aforementioned problem by providing a standard software framework for the researchers of the field. The system simulates the low-level processes and state changes, the flow of the game and the environment of the players. As a result, the autonomous software agents developed on top of this framework make it possible to develop, analyze and test methods of cooperation and high level team strategy.
In my thesis I am going to present the software framework that I developed using the Java language and the tactics and strategy layer on top of it implemented in Matlab environment. The goal of the system is to support the development, testing and benchmarking of different strategies developed in Matlab for the standard RoboCup 2D platform and also to help the research of methods applicable in the field of multi-agent cooperation.
The low-level component developed in Java provides basic services, like communication with the RoboCup server application, processing of sensory information, storing the abstract representation of the environment and scheduling of the agent’s actions.
Based on this, there is the tactics layer implemented in Matlab, which is responsible for the planning of local state trajectories of the angent. Hidden, implicit information is extracted from the world model and complex tactical primitives are defined which describe the control signals required to follow the specified state trajectories. These control signals are translated to sequences of standard RoboCup actions.
The ultimate goal is of course the development of a team, which members can sense their environment and act autonomously to reach long term goals, such as winning the match. This can not be achieved without the development of high-level strategies, which leads to problems, such as formation control or calculating the optimal joint action for distributed agent systems. To investigate and tackle these problems I implemented a coordintaion graph based high-level strategy also in the Matlab environment, which is responsible of defining the long term goals of the team and translating it to complex tactical primitives.