: taking service robots to play soccer

Transcription

1 : taking service robots to pay soccer Larena Adaberto, Escaante Boris, Torres Luis, Abad Verónica, Vázquez Lauro Bio-Robotics Laboratory, Department of Eectrica Engineering Universidad Naciona Autónoma de México, UNAM adaarena@ao.com Abstract. This paper describes the efforts to use the VirBot architecture for controing the operation of robotic biped soccer payers. The ViRbot architecture is designed to dea with a wide range of probems for the service robots, incuding navigating, motion panning, goba ocaization, speech recognition and task panning using a rue based system. This paper discusses what modifications must be done when using devices with imited resources and how decisions can be taken. 1 Introduction The ViRbot system [1] has been tested in the Robocup@Home [2] category in the Robocup competition at Bremen, Germany in 2006 with the robot TX8 and in the same competition in Atanta, USA in 2007 with the robot TPR8, getting the 3rd pace. In the humanoid category we have participated in the Mexican Robotics Open 2007 with a demonstration of our system. The ViRbot system divides the operation of a mobie robot in severa subsystems: Fig. 1. The ViRbot Architecture Fig. 2. RoboNova I Biped Robot

2 Each subsystem has a specific function that contributes to the fina operation of the robot. When trying to contro an autonomous biped robot such as Robonova I [3] with imited processing capabiities, severa adaptations must be done to the architecture in order to simpify the compexity of operation, ocaization and knowedge representation, see figure 2. This paper discusses how the VirBot modues can be modified to aow biped robots to pay soccer. 2 Modue Description This section expains the different ViRbot modues and how they can be modified. 2.1 Interna Sensors In the case of our biped robots, basicay the interna sensors are a gyroscope and tit sensors, digita compass and Tripe Axis Aceerometer. They serve to contro the baance of the robot and detect faings. The biggest difference in biped robots is that there are no whees and whee encoders, so the ocaization probem becomes more difficut to address. 2.2 Externa Sensors The ony externa sensor currenty used in our robots is a CMU Cam 3 camera [4], since there are imposed restrictions when using sonar and IR sensors. This camera is equipped with a Phiips LPC2106 processor, 64 Kb RAM, 128 Kb ROM, seria port and SD card sot. It can grab RGB and YCrCb images up to 352 by 288 pixes at rates of 26 FPS. We do not use any other contact, refective, infrared or microphone sensor. For this reason the amount of information the robot can extract from the environment is imited to visua information. 2.3 Human-Robot Interface As mentioned above, due to space and processing imitations in sma biped robots, currenty they are not equipped with microphones and speech recognition. In the future some speech recognition in Digita Signa Processors DSP s wi be impemented. 2.4 Robot Contro Modue Like the TX8 and TParacho robots, our robots are equipped with on-board controing card. In this case a Micom MR-C3000 with an ATmega 128 MPU with 32K fash

3 memory. A the movements the robot does, ike waking or kicking, are programmed as singe routines in the Robonova I board, using RoboBasic [5]. Once the program is downoaded and stored in the robot, the commands are fired through the robot s standard RS-232 seria port connected directy to the CMUCam 3 seria port (our controing device). The interna sensors are directy attached to the robot s board, so it uses their reading to correct position and standing-up the robot when faing down. 2.5 Perception Modue Unike the service robots, the visua processing and robot s inteigence is done in the same board, in this case the CMUCam 3. Despite we have recenty migrated successfuy entirey our ViRbot system to the Pocket PC architecture, our tests on the robot demonstrated it is sti too heavy to be carried out for the Robonova. We are currenty using basic YCrCb [6] coor segmentation on the camera to find the ba, enemies and goa areas. 2.6 Goba Locaization Modue Since there is no map representation and the environment is aways changing, due to the sef, others and ba movements, a vision based ocaization method is necessary. Currenty MCL Monte Caro Locaization [7] is being impemented but it becomes more inaccurate because there are no map, aser and sonar readings, and the ony fixed characteristics are the fied ines, goa areas and two andmark poes. By the moment our ocaizing agorithm uses the goa areas coor and the amount of viewabe pixes of such coor to determine its distance and current orientation. The movement estimation is obtained by previousy measuring the movements the robot does after each command and getting its probabiity density. When turning, for exampe, each turn has a fixed amount and successive turns are accumuated. The same appies for waking and running. Obviousy a good robot caibration becomes fundamenta. 2.7 Cartographer Since the ony eements in the fied are the robots and the ba, and the fied is a fixedsize rectange, the map consists on a probabiistic occupancy grid. The fied is divided into regions of 5 per 5 centimeters. One byte per ocation indicates the probabiity (from zero to +-128) of being occupied and the sign indicates if it is a partner or enemy robot. In tota 5,400 bytes are needed for storing the assumptions about a the eements in the fied, except the ba and the sef position, those are carried separatey. Since the CMUCam 3 incorporates a SD connector, currenty is possibe to decide where to store the information. By now the map is stored in a 1 GB SD secondary memory card.

4 2.8 Movement Panner The map representation aows cacuating the robot movements in a straightforward way. As impemented in the service robot, here the robot uses Potentia Fieds Theory [8]. Under this idea, the robot is considered as a partice under the infuence of an artificia potentia fied U whose oca variations refects the free space structure and it depends on the obstaces and the goa point that the robot needs to reach. The potentia fied function is defined as the sum of an attraction fied that push the robot to the goa (the ba in this case) and a repusive fied that take it away from the obstaces (other robots). The probabiistic occupancy grid representation gives directy a good approximation to the force fied: Fig. 3. Exampe of Potentia Fied The movement panning is done by iterations, in which and artificia attracting force is induced by the ba and repusive forces are induced by the enemy robots. The sum of those forces wi be the fina direction the robot wi foow, in the form of steps and turns. 2.8 Knowedge Representation Unike the ViRbot service robots, CLIPS Expert System [9] cannot be onger used to represent rues and to fire actions in these imited robots, and another way of managing actions must be used. Commony, variabe ranges are used to decide the robot actions, in the form of: if var1 > ower_va1 and var1 < upper_va1 and var2 > ower_va2 and var2 < upper_va2 and varn > ower_van and varn < upper_van then robot.action(); ese if Exampe of a Computer Program for performing actions based on data ranges but when having many variabes becomes difficut to estabish such ranges in order to avoid etting some ranges out of decisions or, in the opposite case, giving some ranges more than one action. Aso the priority management becomes difficut to modify. This is known as the Subsumption Mode [10] and it s used in ViRbot as the Arbiter.

5 In this case, we are impementing a Fuzzy Rue Decision System [11]. A fuzzy system consist of a number of Fuzzy Rues in the form of ( ) : IF x1 is F1 and and xn is Fn THEN y1 is G1 R L, L, y is G (1) This is known as pure fuzzy form, where F 1, L, F represent fuzzy vaues of each n variabe and the output function G is aso a fuzzy variabe. In this case we have a MIMO (Mutipe Input Mutipe Output) fuzzy rue system. We use this kind of system to activate the different behaviors ike kick or step. For instance one fuzzy rue can be: R: If ba_distance is near and kicking_direction is good then kick is good When fuzzifying one variabe, it s possibe to give a natura meaning to a vaue (or range of vaues) of a given variabe. By natura we mean inguistic understandabe meaning. For exampe we can define the terms near and far for the ba_distance variabe as foows: μ μ near Variabe far Variabe m m cm cm Fig. 4. Exampe of Fuzzy Variabes 2.9 Behavior Seection When evauating a fuzzy rue, the fuzzy vaue μ is evauated for every fuzzy variabe and the rue s fina μ is cacuated using fuzzy inference (in our case a simpe product). Once a fuzzy rues have been evauated, a the resuts are combined using a defuzzifier for obtaining the fina output (in our case we use the maximum μ to fire the behavior μ). As seen above, this way of computing outputs is very simiar to defining ranges, but outputs are basicay combined and then the action with the biggest output vaue is seected. This method aows easy error detection and bad activation vaues. Aso, fuzzy rues can be stored on secondary memory in order to be transferred to other robots ony adjusting their beonging factors (μ functions), and more new rues can be added with a we estabished method for combining them Learning Modue Currenty no earning is impemented in our robots.

6 2.11 Future Works We are intending to deveop a goa-area detection by form to agree with the Robocup reguations for future competitions and sef-caibrated vision and contro agorithms are currenty being deveoped by our new members. 3 Concusions and Discussion This paper proposes some basic modifications to the ViRbot architecture, aowing biped robots to pay soccer. Aso proposes a way of representing and managing knowedge and easiy computing behavior responses with visua sensors. In the past edition of Robocup in Suzhou China, we reached the quarter-finas of the event, being the best team in America in our first participation. We detected some probems with the ocaization and specificay with the anke movement when turning. We have re-designed our robots based on the Robonova-I design but adding 6 DOF more to be abe to rotate the whoe eg around the hip articuation, and a 360 degrees arm rotation by the shouders to be abe to do the throw-in chaenge. By now the biggest chaenge wi be communicating the robots with wi-fi technoogy to become more coaborative and achieve more team coordination. Aso the vision caibration wi be fundamenta for getting a better pace in the competition. We have to demonstrate why we did arrive to finas ast year. References 1. Jesús Savage, Adaberto LLarena, Gerardo Carrera, Sergio Cuear, David Esparza, Yukihiro Minami, Uises Peñueas; ViRbot: A System for the Operation of Mobie Robots. RoboCup 2007 Symposium, Atanta, EU, Robocup@Home, Thrun S., Montemero M.., Koer D., Wegbreit B., FastSLAM: An Efficient Soution to the Simutaneous Locaization And Mapping Probem with Unknown Data Association, Proceedings of the IEEE Int. Conference on Robotics and Automation (ICRA), Latombe, J.C. Robot Motion Panning. Massachusetts, USA: Kuwer Academic Pubishers, CLIPS Reference Manua Version 6.0. Technica Report Number JSC Software Technoogy Branch, Lyndon B. Johnson Space Center, Houston, TX, Muer J.P. and M.Pischei, and M. Thie. Modeing reactive behavior in verticay ayered agent architectures Inteigent Agents - Theories, Architectures, and Languages, voume 890 of Lecture Notes in AI. Springer, January Conner, D. Fuzzy Logic Contro System. EDN, pp , Takagi, T. and Sugeno, M., Fuzzy identification of systems and its appication to modeing and contro, IEEE Trans. Systems Man Cybernet. 15 (1985) Haykin, S. Neura Networks. A Comprehensive Foundation, 2nd Edition, Prentice Ha 1999