ROBOT SOCCER STRATEGY – BIOMIMETIC APPROACH

Nicu George Bîzdoacă, Coman Daniela

University of Craiova, Craiova, Romania,

Hani Hamdan

SUPELEC, Department of Signal Processing and Electronic Systems, Gif Sur Yvette, France

Khalid Al Mutib

King Saud University, College of Computer and Information Sciences, Riyadh, Kingdom of Saudi Arabia

Keywords: Robotics, Biomimetic Approach, Control Strategy, Cooperative Tasks.

Abstract: Biomimicry is the right pattern for a successful algorithm or control strategy. The present paper applies

biomimetic approach to robotic soccer game, using as model the human soccer team’s strategy. For

implementation the authors use a team of MIROSOT soccer robot, with three robots. Using software Simi

Scout a suitable tactics analysis can be extract from the games. Analyzing the real soccer game, a number of

attributes are specified and are embedded at different levels. The specified levels are interconnected and the

game analysis is processed for optimization. The authors approach offer a flexible and more complex

situation: all the robots can be attacking, defending or goalkeeper player. In this case the role changing

strategy is based on the distance between player and the ball, the payer and the goal area and the player and

the attack area. The controller for the attacking robot is designed using Petri nets. The Petri net model is

implemented in Petri Net Toolbox under MATLAB environment. Using this information the robot program

is adapted and the tests/games are experimented. The results are commented and improved control

architecture, based on practical results, is proposed.

1 INTRODUCTION

The nature’s inventions have inspired researchers in

developing effective algorithms, methods, materials,

processes, structures, tools, mechanisms, and

systems. Biomimetics is a new multidisciplinary

domain that includes not only the uses of animal-like

robots – biomimetic robots as tools for biologists

studying animal behavior and as research frame for

the study and evaluation of biological algorithms

and applications of these algorithms in civil

engineering, robotics, aeronautics. Multi-agent

system has emerged as a subfield of AI and helps to

understand and provide with theory and principles

for constructing complex systems with multiple

agents and their coordination/competition in

dynamical environments. The robot soccer game is

an interesting benchmark problem for the multi-

agent systems.

Generally spoken in robot soccer regarding the

division of labor between the components of a

soccer team, namely between the host computer

system and the autonomous mobile robots, three

system configurations are defined: 1. Remote

brainless system; 2. Robot-based system; 3. Vision-

based system.

The vision-based system can be described as the

step from the remote brainless to the robot based

system, as some of the intelligence is transferred

from the main computer to the single agents, but the

control of the vision system and the strategic

coordination still remain tasks of the host unit.

The Micro-Robot’ World Cup Soccer

Tournament (MiroSot) is the brainchild of Jong-

Hwan Kim (Coman, 2008) of KAIST, Korea and

was initiated in 1995. A mobile robot soccer team

consists of up to three micro mobile robots. Two

teams play soccer according to the rules similar in

433

George Bîzdoac

a N., Coman D., Hamdan H. and Al Mutib K. (2010).

ROBOT SOCCER STRATEGY – BIOMIMETIC APPROACH .

In Proceedings of the 7th International Conference on Informatics in Control, Automation and Robotics, pages 433-436

DOI: 10.5220/0002954004330436

 SciTePress

nature to the real soccer game. The Federation of

International Robot Soccer Association (FIRA) has

established these rules.

The mechanical construction of the micro mobile

robot is based on a duralumin frame, on which two

DC motors with gear boxes and a controller board

are mounted.

Robot has two parallel wheels and in addition is

suspended by two slipper elements mounted at the

front and back part of the frame.

2 POLLING ROBOT SOCCER

STRATEGY FROM THE REAL

SOCCER GAME

The soccer robot game structure has 6 steps:

Step 1. Image acquisition and primary calculation

(distance, velocity, relative angle calculation etc.)

Step 2. Decide which posture is suitable for player

offensive or defensive.

Step 3. Determination of team strategy and

player profile assignment.

Step 4. Determination of the target position.

Step 5. Path planning.

Step 6. Calculation of the wheels’ moving

direction, velocity and displacement.

In order to have a successful team the secret are:

good player’s profile, good team strategy.

The player profiles are:

- Attack

- Midfield

- Defender

- Goalkeeper

Using Simi Scout a real game is analyzed and the

player profile is extracted (for this example is used

Romania Columbia, 1994, soccer game, and for

players profile, is used Romanian player profiles).

Using the real game information, the player

situations can be extracted. Let’s analyse , in detail,

the strategy for Attack player (Bîzdoacă et al.,

2005).

Assuming that the left half of the playground is

the opponent side, it is reasonable that the attacking

player must move to the right side of the ball as soon

as possible.

Figure 1: SimiScout analise for soccer game Romania-

Columbia.

3 PETRI NET MODEL FOR

ATTACKING SOCCER ROBOT

Petri nets, developed by Carl Adam Petri in his

Ph.D. thesis in 1962, are generally considered as a

tool for studying and modeling of systems. A Petri

net is foremostly a mathematical description, but it is

also a visual or graphical representation of a system.

A Petri net can be analysed in order to reveal

important information about the structure and

behaviour of the modeled system. The information

may for instance suggest improvements to the

system.

A Petri Net consists of a number of places and

transitions with tokens distributed over places. Arcs

are used to connect transitions and places. When

every input place of a transition contains a token, the

transition is enabled and may fire. The result of

firing a transition is that a token from every input

place is consumed and a token is placed into every

output place. The firing of transitions represents

causality and inferencing relations. Petri nets have

been generalized by allowing multiple token arcs,

inhibitor arcs, place capacity, colored tokens etc.

(Peterson, 1981).

Formally, the structure of a Petri net, defined by

its places, transitions, input function and output

function (Coman, 2008), (Kim et al., 2004).

The Petri net model (Kim et al., 2004) for the

attacking soccer robot modeling the real situation,

has been designed so that the topology to contain the

following situations in which the attacking robot

can be, as follows:

a) The attacking robot is behind the ball;

b) The attacking robot kicks the ball;

c) The attacking robot is in offside position;

d) The attacking robot is in contact with ball and

ICINCO 2010 - 7th International Conference on Informatics in Control, Automation and Robotics

434

the ball is situated behind the robot.

In state (a) the attacking robot is in a probable

position to kick, in state (b) it is kicking the ball, in

state (c) it is in front of ball, so should be careful to

avoid the offside position, and in state (d) it is in

contact with ball.

With these four states, the Petri-net for the

attacking robot controller is formed. “Angle” is used

to refer to the angle between the heading direction of

the attacking robot and ball. “Distance” is used to

refer to the distance between the attacking robot and

ball in pixels. In state (a), the attacking robot is

ordered to move to ball and kick it. In state (b), if

“Angle” is above 45° , or “Distance” is more than 20

pixels, the attacking defense robot goes to state (a).

In state (a), if the ball is on the right side of attacking

robot (offside position), it should go to state (c). In

state (b), if the attacking robot fails to kick the ball,

the robot goes to state (c). In state (c), the attacking

controller orders the robot to move sideways and it

comes behind the ball without touching it and goes

to state (a). In state (c), if “Distance” is below 10

pixels, the robot goes to state (d), so it should move

away from the ball till “Distance” is above 20 pixels

and then can get into state (c).

Analyzing the situations for Goalkeeper player,

for Defender and for Midfield we can generalize the

situations as:

) The robot is behind the ball;

) The robot kicks the ball in the opponent

direction;

) The robot is in unwanted position;

) The robot is in contact with ball and the ball is

situated behind the robot.

Using Simi Scout a set of attributes are assured.

Statistically, after analyzing the all attributes

proposed, only 7 are retained and become transition

for our algorithm:

. Tries to kick the ball, though it is not in a good

position to kick,

. In front of the ball and at the following instant it

is in a good position to kick,

. In front of the ball, and moving to an unwanted

position,

. In unwanted position and escaping from that,

. Misses the ball, and is in front of the ball,

. In unwanted position, and then in contact with the

ball and behind, and

. Away from the ball and behind, but still in an

unwanted position.

The incidence matrix has rank 3, so that the Petri net

model shown in Fig. 2 will have:

m - rank (A) = 1 ⇒ at most 1 P-invariants are

linearly independent,

n − rank(A) = 4 ⇒ at most 4 T-invariants are liniarly

independent.

Figure 2: Petri net model for the soccer robot.

As every robot has the same strategy, the role for

attack, defender, midfield or attack can be easily

switched between the robots. The roll will be

assigned dynamically, be image analyze.

A role assigned it is means that the intervention

area of the robot is in:

• Attack Area

• Midfield Area

• Defense Area

• Goalkeeper Area

The ball position will switch the team strategy.

The team strategy extracted (from Romania –

Columbia 1994 - soccer game - using Simi Scout

software) is:

- If the ball is in attack area then 2 player are

attack player, and the nearest player to midfield

area is midfield

- If the ball is in the midfield area, then 2 player

are midfield player, and the nearest player to

defense area is defender

- If the ball is in defense area, then 2 players are

defense player, and the nearest player to goal

area is goalkeeper.

5 NUMERICAL SIMULATION

AND EXPERIMENTS

The results from the mathematical method of

checking through the invariants method associated

transitions and the corresponding positions after

calculating the incidence matrix of the net have been

validated through the simulations using Petri Net

Toolbox in Matlab environment.

ROBOT SOCCER STRATEGY - BIOMIMETIC APPROACH

435

Table 1: Global Statistics Places.

It was validated in this way that the net topology, the

evolution of (their dynamics), as well as structural

and behavioral properties. The following two tables

(Table 1- Global Statistics Places and Table 2 -

Global Statistics Transitions) present the complete

lists of global indices associated with the places and

the transitions considered in the architecture of Petri

net that modeling the controller for attacking robot.

Table 2: Global Statistics Transitions.

Also, the special options of Petri Net Toolbox,

which confers a high capacity of analysis, has made

possible a synthesis of this Petri net model which

allows exploring the dependences of global

performance indicators associated with the net

positions/transitions on two “Design Parameters”.

6 CONCLUSIONS

The research proposes a biomimetic algorithm based

on analyzing the real situation (Peterson, 1981). In

this paper, a Petri net model is used for designing a

low level controller for the soccer robots. The

presented controller did not use the information of

opponent team. A future development of this

research will add the opponent predicted strategy

and a more dynamical switching strategy. Using a

single model robot controller offer advantages in

terms of implementation, but the team strategy can

be easily indentified by the opponent and

annihilated. The Petri net model presented is

implemented in MATLAB environment. The

simulations studies was validated that the net

topology, the evolution of (their dynamics), as well

as structural and behavioral properties and was

provided the global performance indices associated

with the places and the transitions and also the

whole set of global indices associated with all the

nodes of the net. Finally, the feasibility of the

proposed architecture is demonstrated by the

experimental results.

Figure 3: Dependency of the Queue length index

associated p

position.

Figure 4: Dependency of the Queue length index

associated p

position.

ACKNOWLEDGEMENTS

This report is part of the ROMANIAN NATIONAL

UNIVERSITY RESEARCH COUNCIL (CNCSIS)

contribution to the project PNCDI – II - 289/2008.

REFERENCES

Coman D..2008. Using Petri Nets in the Soccer Robot

Control Architecture, Annals of DAAAM. pp. 295 –

296.

Bîzdoacă N. G., Degertu S., Diaconu I. 2005. Behavior

based control for robotics demining. International

Symposium on System Theory, SINTES 12, ISBN 973-

742-148-5, 973-742-152-3, pp. 249-254.

Kim J. H., Kim D. H., KimY.J., Seow K.T.,2004. Soccer

Robotics, Springer-Verlag.

Peterson J. L.,1981. Petri net theory and modelling of

systems, Prentice-Hall Inc.

ICINCO 2010 - 7th International Conference on Informatics in Control, Automation and Robotics

436