Programming the Goalkeeper of 20 DOF for FIRA Cup

Transcription

1 Programming the Goalkeeper of 20 DOF for FIRA Cup Mauro Santoyo Mora 1,Víctor Hugo Cacique Borrego 1,KarlaA. Camarillo Gómez 1, and J. Manuel Ibarrra Zannatha 2 1 Instituto Tecnológico de Celaya, Av. Tecnológico and García Cubas, Celaya, Guanajuato, México 2 Centro de Investigación y de Estudios Avanzados del Instituto Politécnico Nacional Av. Instituto Politécnico Nacional 2508, Col. San Pedro Zacatenco 07360, México, D. F. karla.camarillo@itcelaya.edu.mx Abstract. In this paper is proposed an algorithm for an autonomus goalkeeper capable of the interaction with the environment of an AndroSot(Android Soccer Tournament) game. The algorithm is based on the behavior of the actuation and the vision systems acting as a trajectory sensor. Through a proposed set of trajectories, the goalkeeper es capable to determinate the trajectory of the soccer-player. The experimental results shows that the goalkeeper stops any ball shooted by the soccer-player, it was the goalkeeper of the DotMX of the FIRA World Cup Key words: Humanoid Robot, FIRA Cup, Robot Soccer, Vision Control. 1 Introduction The robotics itself relates with the desire of synthesize natural functions from a human body by the use of mechanisms, sensors, actuators and computer systems. Obviusly this represents a huge commitment that evidently seems require a multitude of ideas from a lot of classic fields. It seems reasonable to divide Robotics in four main areas: mechanical handling, locomotion, computer vision and artificial intelligence[1]. There exist different Robot-Soccer World Cups all over the world, from which the most remarkable tournaments are the FIRA (Federation of International Robot-Soccer Association) World Cup and RoboCup. In both cases, the humanoid robot has to have anthropomorphic characteristics, but also a certain level of autonomy depending of the environment where the humanoid will unfold. Robot-soccer can be portrayed as a competition of advanced robot technology within a confined space. It offers a challenging arena to the young generation Corresponding author

2 2 Programming the Goalkeeper of 20 DOF for FIRA Cup and researchers working with autonomous mobile robotic systems. It is hoped that FIRA s flagship event, called the FIRA Robot World Cup, will help to generate interests in robotics in the young minds. Through this events, FIRA hopes to help them better understand and appreciate, with interests, the scientific concepts and technological developments involved. FIRA believes that some of these interests will fuel scientific and engineering skills that ultimately develop into research outcomes to serve mankind in a variety of ways [2]. The FIRA Cup is organized into several categories, one of which is AndroSot, were the size dimension of the android is proportional to that of a human being. For this category, the match is played by two teams, each consisting for at most three robots, including one goalkeeper. All the robots are controlled by off-board computers, where the goalkeeper could posses its own camera and be fully autonomous. The team DotMX is composed of research centers as the Department of Automatic Control of CINVESTAV-IPN and the Institute of Research in Applied Mathematics and Systems of UNAM. The team also includes to the Universidad Politécnica de Sinaloa, the Instituto Tecnológico de Celaya and the company Verstand Labs. DotMX integrates a large number of students, researchers and contributors from these institutions, who carry the research projects in parallel. One of the most important achievments of the team has been the First Place in United Soccer obtained in the FIRA Robot World Cup 2012 celebrated in Bristol, UK in August 2012[3]. The application of the Bioloid plattform in its different kit presentations for Robot Soccer World Cups has been considerable because of its versatility and compatibility with other systems like the vision module HaViMo 2.0. The main bases considered for this work consists in the application of the Bioloid humanoid robot in different Robot Soccer Cups with the integration of a vision module compatible with the architecture of the robot. The first reference considered presents an overview of a humaniod soccer robot, build out of readily available hardware. It is described a vision algorithm to detect goal posts, used for Self-Localisation of the robot. The performance of the robot in the FIRA World Cup 2007 is analized and some improvements suggested[4]. With the develop of a cheap and lightweight integrated camera and image processor by the Computer Science Department of Freie Universität Berlin, was possible to integrate a vision module capable of delivering the results from the image processing via a serial interface and further processed by a low power CPU like a microcontroller[5]. It is cited the first results of hook-up a HaViMo 2.0 in a Bioloid belonging to the DotMX team, in order to develop a vision system capable of execute search and track tasks[6].

3 2 Description Programming the Goalkeeper of 20 DOF for FIRA Cup Structure of the Humanoid Robot The humanoid robot assembly is based in the Humanoid Type A of the Bioloid Premium Kit. The robot is of 20 DOF, from which 18 DOF are used for the locomotion of the robot. The 2 DOF remaining were integrated with the HaViMo 2.0 (Hamid s Vision Module 2.0) to create the head of the robot, generating in this way an exploration system. The humanoid s body can be seen in the Fig. 1. Fig. 1. Humanoid robot. Basically the head of the humanoid possesses a pitch and a yaw rotations (Fig. 2), allowing the robot to explore the environment where he is found. This exploration is performed between the HaViMo 2.0 and the actuators conforming the head, where the HaViMo module gives us information about the location of objects in the image space then it is requiered an alteration in the positions taken by the actuators to move the head and frame the ball in the image. Notice that the boundaries of the values that could take each actuator for the pitch and yaw angles, shown in the Fig. 3. This angles were calculated in base to the running degree of the Dynamixel AX-12 actuator. This running goes through 0 to 300 with a resolution of 0.29 [7], implying an 8-bit word that is used to define the desired position. The computational system consists of the CM-510 controller, that is an 8- bit AVR microcontroller with 256K Bytes in-system programmable flash, model ATmega2561. The CM-510 controller (Fig. 4) offers an architecture of buttons, LED indicators, communication buses (mainly used for actutation and vision systems), ADC channels and some other characteristics.

4 4 Programming the Goalkeeper of 20 DOF for FIRA Cup Fig. 2. Head rotations. a) b) Fig. 3. Values and boundaries for pitch and yaw angles. (a) Pitch angle boundaries. (b) Yaw angle boundaries. Fig. 4. Controller CM-510 [7].

5 Programming the Goalkeeper of 20 DOF for FIRA Cup Hamid s Vision Module 2.0 The main objective of robotics is synthesize tha natural body functions into an artificial system. One of the more applied areas is artificial vision, making use of images got from the visual environment, to identify surrounding objects and efects. HaViMo, shown in Fig. 5, is a visual perception system for microprocessors of low consumption, which consists of a microlens and a signal processor with two basic algorithms of image processing precharged. Fig. 5. HaViMo 2.0. The usage of this vision module requires two fundamental operations: calibration and implementation. The calibration phase requires a GUI (Graphical User Interface) named HaViMoGUI, which allows the user define the colors by each object is identified. The definition of object s color is made by selecting the desired color in the RGB(Red, Green, Blue) space or by selecting a pixel from an acquired image as seen in the Fig. 6. Fig. 6. HaViMoGUI.

6 6 Programming the Goalkeeper of 20 DOF for FIRA Cup It is possible to modify several characteristics of the module that will be considered at any frame captured by the camera. The main characteristics are: gain of color(red, green or blue), luminance, white balance, and so others. The desired configuration as well as the colors of interest are written in an EEPROM memory integrated in the HaViMo, allowing us to turn off and on the module without the needing of a recalibration, all this under the same conditions of ilumination. The algorithms contained in the signal processor are based in the color processing. The first develops a color segmentation based on the calibration required by the user, which results are presented in form of a Look-Up Table (LUT). The second algorithm generates a low resolution image by the gridding process. In this process each byte contents the information of a 5 5 region extracted from the original image saving it in a new image location (x, y). Four bits codify the color of the region and the rest four bits represent the number of pixels containing that color. At the end of the execution of this two algorithms the results are presented in a table (Fig. 7) containing the next information[5]: Fig. 7. Captured image information [5].

7 Programming the Goalkeeper of 20 DOF for FIRA Cup 7 Region index: Contains the value 0 if the region is invalid and nonzero otherwise. Region color: Color category of the detected region. Number of pixels: Number of detected pixels inside the region. Sum of X coordinates: Result of addition of the X coordinates of all detected pixels. Can be divided by number of pixels to calculate average X. Sum of Y coordinates: Result of addition of the Y coordinates of all detected pixels. Can be divided by number of pixels to calculate average Y. Max X: Bounding box right margin. Min X: Bounding box left margin. Max Y: Bounding box bottom margin. Min Y: Bounding box top margin. 3 Strategies of the Goalkeeper 3.1 Searching and Tracking the Ball As first step, was necesary calibrate our vision module to define the adecuated LUT to recognize the ball from the visual environment, performing a selection of the orange color. In base to this calibration, was possible to know the actual coordinates of the ball in the image space which are usefull to calculate the track positions of the head. Before we can perform a track motion it is indispensable to find the ball in the environment. This search process consists in perform an exploration of the frontal visual field of the humanoid, by taking 6 different positions divided in two pitch positions. With this search the robot was capable to see in a radius of 2 meters near him. It was propossed a division of the visual field (see Fig. 8) for the goalkeeper in order to discretize the different possibilities. As result of this idea, the visual field was created as follows: Fig. 8. Visual field for the goalkeeper

8 8 Programming the Goalkeeper of 20 DOF for FIRA Cup 3.2 Ball Trajectories Analysis It is important to recognize the possible ways in which our opponent could kick the ball in our goal direction. In this way, it was developed a study of the possible different trajectories that would take the ball once it is kicked by another robot. With this, the experiment was based in the trajectories shown in the Fig. 9. Fig. 9. Trajectories defined for experiment. The positions registered by each actuator, for pitch and yaw motions, are shown in the Table 1. This route consisted of a straight line direct to the goalkeeper, taking an initial position in front of him at 150 cm. It is important to remark that every route was tested with 20 samples to homogenize every trajectory and the experiment. The Fig. 10 shows the results of Table 1. As can be seen in the figure there is present a noticeable change in the pitch angle, by the other hand, in the yaw angle the registered changes are very small. Henceforth, for this kind of behavior we will assume that the ball is advancing in straight line to our goal. The second case that will be presented consists in a trajectory that starts its trail at 60 cm from the left of the robot and 150 cm in front of him, ending this trail on the opposite side and with the attempt of scoring a goal in a cross path. This route corresponds to R3 in the Fig. 9. The results are shown in the Table 2 and the Fig. 11. All the remaining routes were tested in a similar way, concluding all the experiment in an algorithm with the characteristics and cases presented below,

9 Programming the Goalkeeper of 20 DOF for FIRA Cup 9 Table 1. Position values for route 1 (R1). Number Position Pixels of Ball Centering Pitch Yaw Cx Cy X X Fig. 10. Actuators behavior for route 1.

10 10 Programming the Goalkeeper of 20 DOF for FIRA Cup Table 2. Position values for route 3 (R3). Number Position Pixels of Ball Centering Pitch Yaw Cx Cy X X Fig. 11. Actuators behavior for route 3.

11 Programming the Goalkeeper of 20 DOF for FIRA Cup 11 for this, it was considered the initial position were the robot would find the ball, dividing the visual field of the robot in three sections: Left Visual Field: The possible cases considered for this field are: Cross Path Shot (Left to Right): This condition will take place when the desired position of the yaw angle results in a sign change in the next equation: (Angle yaw 512) < 0 (1) It is required as well a decrement in the pitch angle to take this case as a possibility. Direct Shot: A small threshold is considered in this case, meaning this that it is present a noticeable change in the pitch angle and almost a constant angle in yaw. This case would be similar to route 1 in the Fig. 10. Left Side Shot: It will be considered that the yaw variable should be larger than the threshold stipulated in the case of a direct shot. This case also requires the position of the ball s centroid, which would be located in the left side of every captured image. This coordinates must be less than 60 for the X axis and larger than 30 for the Y axis. (Angle yaw 512) > 100 Cx < 60 Cy > 30 (2) Central Visual Field: There exist five cases when this condition takes place during a match. For this case is defined a new threshold with origin in the central position of the yaw rotation. The boundaries of this threshold are defined by an angle of ±11.4 from its origin. Shot to the Left Side: This condition is susceptible to be present because of a cross path shot when is accomplished: (Angle yaw 512) > 0 (3) Requiring the goalkeeper to stop the ball at his right side. In case the ball runs a trail to the same left side crossing the left boundary of the threshold, it will be truth: (Angle yaw 550) > 10 (4) Shot to the Threshold Center: This shot will be defined as a direct shot to the goalkeeper position, where the pitch angle will describe a function with decrements to track the ball. If this case presents the initial position of the ball in the left side of the threshold, the condition to be accomplished for the trajectory must be: (Angle yaw 512) > 15 (Angle yaw 550) <= 10 Angle pitch > 10 (5) By the other hand, for the right side of the threshold: (Angle yaw 512) < 15 (474 Angle yaw ) <= 10 Angle pitch > 10 (6)

12 12 Programming the Goalkeeper of 20 DOF for FIRA Cup Shot to the Right Side: The goalkeeper will consider this condition viable as a cross path shot going from the right to the left side when: (Angle yaw 512) < 0 (7) In the case the ball has been found in the right side of the threshold and runs a trail to the same side, the condition that describes it is: (474 Angle yaw ) > 10 (8) Right Visual Field: The last three cases present for this visual field are listed below: Cross Path Shot (Right to Left): For this case the side change condition will be traduced as a sign change in the next operation: (Angle yaw 512) > 0 (9) In this way the goalkeeper will notice the cross path in base to this change and the decrement in the pitch angle. Direct Shot: It is considered a small threshold, where if the ball remains along its trail inside of the boundaries, the pitch angle wil decrement its position while tracking the ball. Right Side Shot: The yaw angle described while the system tracks the ball must be lesser than the right boundary of the threshold, with this condition, also it is necesary that the centroid of the ball remains in the image with coordinates larger than 100 in the X axis and larger than 30 in the Y axis. (474 Angle yaw ) > 10 Cx > 100 Cy > 30 (10) 4 Results In a cualitative way the results that has been gotten are listed below: The vision system HaViMo 2.0 was applied satisfactorily in two different programming languages (RoboPlus and Embedded C for AVR microcontrollers). In base of the objects recognition, as the ball, was possible to develop a tracking system with the exploration system. With the research developed to identify the behavior of the exploration system as a trajectory sensor, was possible identify the trajectories of the ball in different cases. So that, it allowed to define an algorithm to control the reactions of the goal keeper once the ball was kicked to our goal. One of the most remarcable result of these research was that for the first time, a goalkeeper maintain its own goal with no goals received. This result was gotten by the team dotmx in the FIRA 2012 Tournament celebrated in Bristol, United Kingdom, which won the first place in the AndroSot category.

13 5 Conclusions Programming the Goalkeeper of 20 DOF for FIRA Cup 13 It is possible to achieve a competitive robot under the Bioloid architecture and the hook-up of a HaViMo 2.0 module. Although the main controller was not dedicated in execution for the image processing, the results were satisfactory in the processes of search, recognition and track of the ball during the estimation of its trail. The next step in this research will be the substitution of the CM-510 by the controller CM-530 and the hook-up of a LynkSprite Camera with the objective of developing a digital image processing that will predict the ball trail overall in base to the image information. References 1. John J. Craig: Robótica. Editorial Pearson Educación, México Federation of International Robot-soccer Association, Sitemap, net/?mid=firaoverview 3. Verstand Labs, Website, 4. Joerg Christian Wolf, Phil Hall, Paul Robinson, Phil Culverhouse: Bioloid based Humanoid Soccer Robot Design, Technical Report, Centre for Robotics and Intelligent Systems, University of Plymouth. 5. Computer Science Department of Freie Universität Berlin: Embedded Vision Module for Bioloid Quick Start. Technical Report. 6. Luis Samahí García González, Juan M. Ibarra Zannatha: Visión Artificial con HaViMo 2.0 para el Humanoide Bioloid. In: Sexto Workshop GTH, México Robotis, e-manual,