A Software for Game Strategy for Robot Soccer based on Acting Areas

Transcription

1 Anais do XXVI Congresso da SBC EnRI l III Encontro de Robótica Inteligente 14 a 20 de julho de 2006 Campo Grande, MS A Software for Game Strategy for Robot Soccer based on Acting Areas Juliana G. Denipote 1,2, Simone D. Francisco 1, Sérgio Ricardo G. Salazar 3 1 Cientistas Associados Desenvolvimento Tecnológico Ltda. Rua Alfredo Lopes, 1717 Vila Elisabeth CEP: São Carlos, SP Brasil. 2 Departamento de Engenharia Elétrica EESC USP Universidade de São Paulo São Carlos, SP Brasil. 3 Instituto de Ciências Matemáticas e de Computação ICMC USP Universidade de São Paulo São Carlos, SP Brasil. juliana.gouveia@cientistasassociados.com.br, sd francisco@yahoo.com.br, ssalazar@icmc.usp.br Abstract. This article describes the Software of Strategy for Robot Soccer implemented by the company Cientistas Associados. Although the software was developed for F180 of ROBOCUP robots category, it can be used for other categories as well. In order to allow the robots play soccer, the software uses the strategy of Acting Area that determines the action each robot should execute basing on the area it is positioning in the field. The software uses the method of potential fields to avoid collisions among the robots during navigation. 1. Introduction In 1997, during a conference on Artificial Intelligence in Japan, the first ROBOCUP (Robot World Cup) happened: a competition among robots or simulators, recognized by IEEE (Institute of Electrical and Electronics Engineers), in which the goal is to play soccer. Since then, ROBOCUP has been happening annually in internationally recognized conferences on Multi-agents Systems and Artificial Intelligence, transforming the robots soccer in a tradition. In a first moment, the system could be seeing as just for entertainment. However, the elements involved in the process of doing a team of robots working harmoniously can be considered a challenge, a stimulating research in several areas, as in computer science, as in electric and mechanical engineering. The goal of Cientistas Associados Robot Project is the construction of robot soccer kits, beginning from robots elaboration until the team s control in field. This kit is composed by 5 robots of the F180 ROBOCUP category [RoboCup 2006], a vision system and a game strategy system. The robots were developed in accordance with the norms of ROBOCUP, with a diameter of 180mm and 150mm height. The robot communicates wirelessly with the central server, using a protocol to manage collisions and allowing the server to communicate with several robots simultaneously. During a robot soccer game, the computational vision system is responsible for tracking all of the players and the ball in the field, supplying the strategy system with their respective positions and orientations. With such players and ball information, the strategy 1

2 system determines a new action for each robot. As soon as the indication of the new action arrives to the player, a communication protocol transmits a message from the computer to the robots, and the task is performed by the electronic circuitry and mechanical actuators. This article describes the game strategy software, explaining how it manages and send the actions to be executed by the robots team with the purpose of win the game. To reach this objective, the system look for the best form of, possessing the ball, carry it to the goal, without colliding the robots each other and prevent the opponent team to reach the goal as well. This article is structured in sections as follows: the second section describes a general vision of the robots soccer system through its architecture, and gives a brief description of architecture modules; the third section focus in the so-called Acting Area soccer game strategy; the fourth section describes how the robots navigation and crashes avoidance are made using Potential Fields; finally, the fifth section has the conclusions. 2. Robotic Soccer System - General Description The Software for Game Strategy for Robot Soccer is responsible for the interface between the user and the robots team, allowing them to play football under a game strategy. The architecture of the software has five modules, as can be observed on Figure 1: Interface, Vision System, Strategy, Navigation and Communication. Figure 1. Software Architecture. The Communication Module is responsible for the interaction between the robots and the software. It sends orders for the robots to play. The Navigation Module is responsible for each robot s autonomous locomotion along the field, trying to avoid collisions 2

3 among them. The Strategy Module contains the strategy for the game, in other words, at each instant, it determines the action that each robot should execute. The Vision System is responsible for finding the robots and ball s positions in the field through image processing; the images are captured by a camera located above the field. Finally, the Interface Module is a graphic designed interface from which the user can simulate and visualize the game. Although each module is detailed in the following subsections, a deeper emphasis will be given to the Strategy Module and Navigation, which will be detailed in the sections 3 and Communication The communication between software and the robots is carried through by communication protocol that transfers the commands informed by the software to the robot kernel. These commands are the movements that the robots must do in field. The communication protocol is structuralized to guarantee consistency in the sending and reception of data, informing error messages when the requested command is not been properly executed by a robot. By the protocol, the robot receives only commands that belong to the software primitives set and these commands are mapped in an action in the hardware Vision System The vision system is a system of objects search and track. For the robots soccer, the track is made in some objects (robots and ball), using a camera that captures the image of the field. To track objects, it is initially necessary to mark the robots and the ball and calibrate the robot s colors. The marking defines the corresponding colors standard of an object (robots and ball) and differentiates the robots of each team. By marking it is also possible to define the orientation, which defines the player s front. In the case of the opponent team, it is not possible to determine the orientation; because of each team have its proper player s identification form and its orientations. The whole field, the centre circle, and goalposts are also marked due to orientation. After the calibration and marking, the track module has the necessary information to follow objects in a video User Interface The Software for Game Strategy for Robot Soccer is responsible for doing the interface between the user and the robots team, playing de game with a strategy. For this purpose, a friendly visual interface was created enabling to locate the robots in field, as well as begin the game. The software is also able to simulate games without the robots in field in order to check the strategy. Depending on the game type chosen by the user, some information in the interface will be exhibited or omitted, but regardless of the game type, this information will be displayed on screen in five different locations, as can be seen in Figure 2: menu options (where the Strategy options are available, or can be imported or exported to a file), field (where the match is visualized - when playing a real game, a video of the field is exhibited, and when playing a simulated game, a moving picture is shown), default options (the basic options for the game like Close, Play the game and Pause the game), the teams formation (in which one can access each robot information, like) and message box (where is exhibited the information of the other areas selected by the user). 3

4 Figure 2. Software Screen. The software was developed for robots of the category ROBOCUP F180, however it can be used for robots of other categories by the entrance of each robot s specific information in the interface, as for instance, its radius, its marks, etc. 3. Game Strategy - Acting Area The Software for Game Strategy for Robot Soccer is responsible for determining each robot s actions during the match. The strategy created for the game is called Acting Area and defines the robot s actions according to the area they are located in the field, their category (fullback, striker or goalkeeper), the other players positions as well as the position of the ball. The strategy for the goalkeeper robot is always stay in positions where the abscissa coordinate corresponds to the defensive line of the goal area(x gol ), and the ordinate is calculated according to the ball position, but always respecting the limits of the goalposts (y gmínimo y robô y gmáximo ). When the ball is in a certain distance considered far away from the goal, the robot is positioned in the center of the goal (Figure 3). Figure 3. Goalkeeper actions when it is far from the goal. 4

5 When the ball is in a distance considered close to the goal, but is moving apart from the goal (Figure 4 a), the goalkeeper is positioned in the corresponding abscissa of the ball (y bola ). When the ball is in a distance considered close to the goal and approaching it, the robot is positioned in the abscissa that corresponds to the estimated intersection of the ball path and the goal (Figure 4 b). This estimate is given by the equation: y inters = y b y b (x b x gol ) x b Where (x bola, y bola ) is the coordinate of the ball; ( Deltax bola, Deltay bola ) are the displacements of the ball in x and y, observed in the last two frames; x gol is the line in x where the goalkeeper should stay for not leaving the defensive line in the goal area, nor to invade the goal. Figure 4. Goalkeeper actions when it is near the goal. The possible actions for the fullback robot are: to give passage for the robot of the same team that has got the ball; and to go to in front of the ball in order to disturb some opposing robot to catch the ball. The attacker robot actions are: to go towards the ball; to catch the ball; to pass the ball for another robot; to carry the ball near the goal; and to kick the ball to the goal. When a fullback robot is in a favorable position to execute an attacker action, it changes the fullback category with some attacker and executes the action. For instance, when a fullback is near the ball in a favorable position to kick it for the goal, its category changes, becoming an attacker and an attacker robot becomes a fullback. The same happens when an attacker robot is in a favorable situation to become a fullback. The areas where those changes are possible are exemplified in Figure 5. Another situation, that it is necessary to verify the area where the robot is to determine its action, is when the robot has the ball and needs to give it for another robot. In other words, it will execute a pass. In that situation, the software for game strategy creates a graph whose nodes are the robots of the same team and the edges are drawn among the robot that has the ball and all the other robots of its team. The edges receive values that correspond to the distance among the robot that has the ball Dand the robot that will receive the ball multiplied by a weight P (to see Figure 6). In order to determine the pass weight, it is verified if the robot that will receive the pass is blocked, in other words, it is verified if there are robots in certain areas in 5

6 Figure 5. Changing areas between strikers and fullbacks. Figure 6. Graph for pass determination. the field where is impossible a player pass the ball to another. Figure 7 illustrates the blockade areas. In Figure 7(a), the orange circle represents the robot that has got the ball and the black circle represents the possible candidate to receive the pass. The distance between then was called R. In Figure 7(b), two circles with radius R were drawn center at these two robot. In Figure 7(c), two lines were drawn with a distance d1 among them. In addition, a circle with radius d1 is drawn with center in the robot that should receive the pass, as in Figure 7(d). Finally, Figure 7(e) has the areas where the blockades are verified. If there is robots in area A, the weight P of the edge of Figure 6 graph will be equal to 4, if there is robots in area B, P will be equal to 3, and in the area C, P will be equal to 2. If there aren t robots in those areas, the weight will be 1. After having calculated all the values of the graph edges, the smallest path is used to determine for which robot the ball will be passed. The robot, whose edge with the robot that has got the ball has the smallest value, will receive the pass. In Figure 6, the robot that will receive the pass will be the 3, in spite of its distance of the robot that has got the ball be larger than the distance of the 6

7 others, there isn t any other robot close to him that hinders the pass. A message is send to the Control Module (see subseção 2.1) for turn the robot 1 towards the robot 3 and pass the ball. Figure 7. Blockade areas to determining the weight of a pass. 4. Navigation When software determines that a robot must go to a defined position in the field, the robot must do the trajectory until the destination point without colliding with another robot. This locomotion control is made by Potential Field. The Potential Field method is based on the existence of imaginary forces that act on the robot, considering that the obstacles produce repellent forces on the robot and the destination produces attracting forces. These forces are combined by a potential function that describes the potential field. A resultant force R, obtained by the addition of all the attracting and repellent forces, is calculated for each robot position [Wijk 2001, Gh. Lazea 1997, K. Goldberg 1991]. Figure 8 shows, from the robot view point, the forces on the ball (B) and its opponets (A). In the upper left figure, A exerts repellent forces field and in upper right figure, B exerts an attracting forces field. The lower figure shows the mapping produced for the forces in this environment. The potential function supplies, at each iteration, the most promising direction for the robot. The advantage of this approach is that it is computationally cheap, and able to being operated in real time. The main difficulty that must be surpassed is the presence of local minimums in the potential, i.e., when the sum of attracting and repellent forces result zero, disabling the robot to has a direction to follow. Such problem was solved by using vortex fields. Vorticity is a property of fluids to surround the obstacles during its movement. Thus, the employed method for robots navigation is the combination of repellent and attracting forces in the Potential Field and the vortex fields. Based on these methods, a decision is taken on which side the robot will go, in order to deviate from the obstacles [Luca and Oriolo 1994, Mezencio 2004, Ribeiro et al. 2001]. The Figure 9 shows a simulation of potential field where the obstacles robots are represented by red circles, the initial and destination points are represented by blue circles 7

8 Figure 8. Attracting and repellent forces. and the determined trajectory is represented in green. It can be noted that the trajectory defined for the potential field method prevents the collision with the robots in the field. Figure 9. Path simulating using the potential fields method. 5. Conclusion This article presented the Software for Game Strategy for Robot Soccer developed by the Cientistas Associados Company for Robocup s F180 robots category. The strategy used for this software was Acting Area, which determines the actions of each robot as a function based on the area where it is located in the field. Also, the Potential Fields method was presented for the robots navigation preventing collision among them. The presented methods employ simple calculations with integer values that make possible perform the game in real time. Acknowledgements The authors thank Fapesp Fundação de Amparo à Pesquisa do Estado de São Paulo - for the financial support. 8

9 References Gh. Lazea, E. L. (1997). Aspects on path planning for mobile robots. K. Goldberg, D. Halperin, J. C. L. R. W. (1991). Algorithmic Foundation of Robotics. A K Peters Ltda, Massachussetts, USA. Luca, A. D. and Oriolo, G. (1994). Local incremental planning for nonholonomic mobile robots. IEEE Int. Conf. On Robotics and Automation, pages Mezencio, R. (2004). Implementação do método de campos potenciais para navegação de robôs móveis baseada em computação reconfigurável. Workshop de Teses e Dissertação USP São Carlos. Ribeiro, C., Costa, A. H. R., and Romero, R. A. F. (2001). Robôs móveis inteligentes: Princípios e técnicas. Anais do XXI Congresso da Sociedade Brasileira de Computação, SBC, 3: RoboCup (2006). Robocup official site. Online. Disponível em robocup.org - last access 04/04/2006. Wijk, O. (2001). Triangulation Based Fusion of Sonar Data With Application in Mobile Robot Mapping and Localization. PhD thesis, Royal Institute of Technology. 9