Intro

For last 10 years development of telecommunication technologies has led to occurrence of concepts of cross-platform, distributed and intellectual program systems. Such systems can be realized by different ways, but multiagentnye systems (MAS) concentrate all properties necessary for such technologies with the greatest expressiveness and completeness. Results from introduction agentnyh technologies have confirmed the predicted perspectivity to this direction. The technology and the theory of agents continue to develop within the limits of research and commercial projects. The special attention is given integration of methods of an artificial intellect which till now found application mainly in research works, in real commercial appendices (personal assistants, obrabotchiki mails, programs for electronic commerce, computer games, control systems and the control by complex processes in medicine, the industry, systems for search and data processing).

How briefly to formulate basic idea MAS? < The key element of these systems becomes the program agent, capable to perceive a situation, to make of the decision and kommunitsirovat with other agents. These opportunities considerably distinguish MAS from the organized systems existing "rigidly", providing with it such important new property, as self-organizing. Thus separate "parts" of the program for the first time have an opportunity to agree about how the problem should be solved, these "parts" get own activity and can initiate dialogue with the user during in advance not ordered moments of time, they can work in conditions of uncertainty and offer specifications and pereformulirovki problems, etc.

The concept of agents developed within the limits of multiagentnyh of technologies and multiagentnyh of systems (MAS), assumes presence of activity, that is abilities of the program independently to react to external events and to choose corresponding actions. Today agentnye technologies offer various types of agents, models of their behaviour and property, family of architecture and the libraries of components focused on modern requirements, such, for example, as distribution, autonomy, etc.

Practically in all works where the agent and what its basic properties is defined, that such, a platitude became the remark on absence of a common opinion in this occasion. Actually, using concept "agent", each author defines the agent with a concrete set of properties. The concept of the agent is used in different areas, for example, on manufacture by the agent the robot, and in the field of telecommunications - the program etc. As consequence, depending on an inhabitancy agents possess different properties. Therefore during development and realization of systems within the limits of the given direction there was a set of types of agents, for example: independent agents, mobile agents, personal assistants, intellectual agents, social agents, etc.

Thus, instead of unique definition of the base agent, there is a set of definitions of derivative types. Considering the aforesaid, the concept of the agent is expedient for treating as a meta-name or a class which includes set of subclasses. It is possible will stop on the following neutral concept:

Agents are described as by a number of properties which characterize concept of the agent. Usually the agent possesses a set from following properties:

• forecasting of a condition;
• adaptibility: the agent possesses ability to be trained;
• autonomy: the agent works as the independent program, putting itself the purpose and carrying out actions for achievement of these purposes;
• collaborativity: the agent can cooperate with other agents in several ways, for example, playing a role of the supplier/consumer of the information or simultaneously both these roles;
• ability to reasonings: agents can possess partial knowledge or mechanisms of a conclusion, for example, knowledge how to cite data from various sources to one kind. Agents can specialize on a concrete subject domain;
• communicativity: agents can communicate with other agents;
• mobility: ability to transfer of a code of the agent from one server on another.

Agents can be considered as development of a paradigm of object-oriented programming: from commands to functions, from functions - to objects, from objects - to agents.

Now methods and strategy of management by agents within the limits of MAS and methods of interaction of agents among themselves are actively investigated.

MAS and agents find wide application at work with the distributed systems, the centralized data processing in which is complicated or in general is impossible by virtue of their distribution. Use of the independent agents cooperating among themselves, allows to make parallel data processing in all parts of the distributed system. Also, due to decentralization of process of processing of the information, reliability of system as in case of the centralized system, refusal of the central unit leads to crash of system as a whole raises. In case of the decentralized system refusal of one or even several units of system does not break working capacity of other units and systems as a whole.

Obvious spheres of application of the such decentralized systems are the distributed databases, computer networks and klasternye systems.

Separately it is necessary to allocate such environment for functioning MAS and agents, as global computer network Internet. In a kind of heterogeneity of network Internet and absence of strictly certain format of a data storage on servers of a network, on the agents working in Internet, additional requirements are imposed ? they should work on various program and hardware platforms and "be able" to process data presented in various formats that demands from them some "intellectuality". Additional means can be applied to maintenance of intellectuality.

As an example multiagentnoj systems it is possible to result game in the robotized football, the championships on which are spent since 1997 by federation RoboCup.

As agents in this system act it is hardware the realized mechanical robots-football players operated by specially written program modules. Depending on mechanical realization of robots-football players (wheel robots, 4-legged robots and robots-humanoids), and also schemes of the organization of competitions the circle of problems solved by program modules can widely vary, as well as the list of controllable parameters as robot-agent and environment surrounding it.

For the reasons specified above now development of operating algorithms for the robots participating in competitions of federation RoboCup, is not represented possible because of complexity of mechanical realization of the robot-agent and a plenty of controllable parameters of system.

Nevertheless, in 2001 in the Moscow State University the system on the organization of competitions of virtual robots-football players received the name "Virtual football" has been developed. Its purpose - creation and working off in a mode of modelling of such algorithms of management of robots-football players which in the further can be transferred and on real robots. Thus group (command) algorithms of strategic and tactical levels of management are considered by a command of robots.

The analysis of the specified schemes of competitions of robots-football players shows, that, despite of distinction of designs of such robots, and the technical decisions accepted at their construction, in schemes of competitions of robots much in common in top, strategic, levels of management of players that allows to set the task of their working off by means of uniform means of modelling. As such system in the given work by development of algorithm of management system "Virtual football" will be used.

The analysis and characteristics of existing algorithms

Below the brief characteristic of two algorithms of management by robots football players is resulted.

Initial codes of algorithms and the detailed description of their functioning can be found on sites of developers. I shall result the positive and negative parties of the given realizations:

Algorithm of management of robots-football players of a command «Dnepr»

• Improvement of a prediction. The account of finiteness of the sizes of the player and a ball; otskoka from a wall; bars; the account of initial speed of the player in the beginning of a prediction; a turn and dispersal in a direction of a ball, and also time spent for a turn before impact; position of enemy players, etc.
• Improvement of technics of impact. The account of finiteness of the sizes of the player and a ball; physics of collision of the player and a ball (final weights of objects); variable accuracy of impact depending on a situation, etc.
• Specialization. It is possible to try to pick up numerous parameters of a binding to a straight line a gate - a ball for defenders; to add the third halfback; does not prevent to experiment with allocated attacking.
• Command game. In the version of a server 2 it will be the basic tactics-strategic aspect of game. However also more vigorous attempts in this direction are now possible. Actually komandnost games, apparently, also should become a corner stone of a scientific part of movement of virtual football.
• The account of actions of the opponent. Here are possible as the elementary receptions (for example to not beat there where near to linej flight of a ball there will be an enemy player), and refined enough (pressure of the goalkeeper, the analysis of the purpose of group actions of the opponent, etc.)
• Strategy. Since version 1.5, there was an opportunity to learn current account, time of game, etc. It already allows to change strategy on a course of game. Besides it is necessary to realize more precisely various lines of actions - protection, poluzashchita, an attack. These strategy should possess the certain inertia and to depend not only on position of a ball, but also from the general distribution of players to a field, the one who supervises a ball, etc.

Expert system "Virtual football" DIAI

• the System allows to model the processes approached to real conditions. As in it sobljudeny some laws of physics (preservation of an impulse, friction, inertia).
• In expert system various ways of modelling of rules are realized. Visual programming of rules, and their writing manually.
• Relative simplicity of system.
• Various knowledge bases can be stuck together, that is to unite lists of rules written by various people.
• Realization of commands of virtual football allows to make their self-trained that it is impossible to tell about expert system. Visual programming does not allow to do a command self-trained, therefore after of some games to win against the opponent it is impossible.
• At increase or reduction of scales of a field, there are displacement of the allocated zones and the purposes concerning a field.
• the User cannot create probable zones, that is hit of all players in corresponding zones because of it probably imposing is obligatory corrected.
• the User cannot adhere a concrete zone to the concrete player.

The purposes and problems

The present work is directed on the decision of problems of the robotized football and has for an object development of algorithms and programs of management of robots-football players, and also creation of means of program modelling of football of robots.

Object of research in the given work are algorithms and the programs operating commands of robots-football players, in a mode of modelling of their game on a computer.

Methods of carrying out of researches

Researches on a theme magisterskoj works will be spent in the specialized environment of modelling of game of virtual robots-football players developed in the Moscow State University, "Virtual football".

The principles incorporated in the given environment of modelling, and also the requirement to developed algorithms of management of group of robots-football players.

Stages of carrying out of work

As a result of the analysis of existing algorithms of management of group of robots-football players, and also systems of modelling of their game, it is possible to allocate following stages:

1. Development of strategy of an arrangement of robots-football players on a game floor. As criteria of an estimation of position of agents of system on a game floor minimal time of transfer of a ball, the account of average speed of players, the account of distance up to players of a command of the opponent, a direction of carrying out of attack will be used the minimal distance between agents of system. Corresponding systems of the equations and inequalities, and also their decisions will be deduced at a following development cycle magisterskogo the project.
2. Realization of impact on a ball. First of all it is necessary to consider, that impact on a ball should be productive. That is its result should be either the hammered goal, or successful transfer of a ball to a member of a command, or prevention of goal situations at own gate. For realization of that it is necessary to consider a trajectory of movement of a ball after collision with the player.
3. Specialization of players. It is necessary to consider such specific problems for football, as protection of own gate, poluzashchita and attack of a gate of the opponent.
4. The account of position and actions of the opponent. Any of the considered algorithms of management of agents of system did not consider position and actions of the opponent, hence their account will allow to raise productivity of game of a command considerably.