Parsian - Small Size Robot League
Contents
1. By using Discrete Fuzzy Role Bases after getting every robot and the ball state and state differences we defined some membership functions and fuzzy role bases The output of these fuzzy sets is robots behaviors which can be defined by assuming the current robot states ball position and opponents movement method Here are some sample Fuzzy role bases which can define the behaviors global or local e Global behavior If the opponent is in attacking state and the ball is in its hand then our robots should be in defense state e Local behavior If the ball is in our robots hand and the closest opponent is far enough then the robot should dribble to the opponent goal e Local behavior If the ball is in our robots hand and the opponent robot is not far enough then our robot should pass the ball to the closest teammate e Andsoon By using these role bases and defining appropriate membership functions we are able to define the best robot behavior model at the time As these models have been defined as Fuzzy Behaviors The final approach is the best behavior for the robot This method is one the best methods comparing to other AI and Intelligent Search Methods with higher calculation and expanding ability By using this Multi Agent System we can teach the system the new behaviors we may experience Positions variables and their differences x y x y are used to calculate the angle and the velocity of the moving2. engine implemented inside to software to fully simulate a Small Size Robot Game This layer receives commands from the Task Manager Layer and changes the world model data in order to simulate the changes The simulation engine uses a graphical interface to sketch the virtual world Feature plans consist of implementing Open Dynamics Library http www ode org inside the simulation engine to improve its accuracy and flexibility Current simulation engine is capable to simulate all the team strategy with more than 70 of reliability factor As mentioned before this application is written in Microsoft Visual C We run this application under Microsoft Windows XP on an Intel Pentium IV PC 3 06 GHz and with one gigabyte of RAM Do oe we pt pea Oe ph eee Oe ain shg E Ske aw S tu Ory Jre Fig 8 Main Software Screenshot Simulation Mode 3 2 Game Strategy The team strategy is based on current world model which is provided by the low level layer of the main software One of the most important things in strategy implementation is the prediction Prediction reduces the effect of the system immanent delay when tracking and controlling fast moving robots Because of physical feedback loop delay in the controlling fast moving robots we use predictor to cope with the system latency The pose predictor orientation and position of the ball and robots can significantly improve the accuracy and precise of the strate
3. parts e Communication receiver e Motor and Spinner driver e Kicking System Driver Fig 5 Main Circuit As demonstrates in figure 5 circuit board is industrial standards compliant For increasing the reliability of the system each part of the circuit have separate power supplier Communication part includes AUREL XTR module and attached AVR microcontroller This circuit operates in 434 869 MHz frequency and manages data reception After receiving data from transmitter the microcontroller delivers decoded commands to the motor driver section through serial protocol In the driver module three motors are driven using two LM298s The microcontroller first processes the input signal of shaft encoders and then controls the speed of each motor through generating proper Pulse Width Modulated signals A simple infrared sensor in front of the robot senses the ball presence and triggers the microcontroller to start the spinner In the kicking part we use 5 DC to DC converters to increase the voltage from 18 to 180 volts This potential charges two 2200 micro farads capacitors When kick signal receives the microcontroller drives a relay which opens the discharge path to the solenoid The microcontroller adjusts the relay opening time in order to control the kick strength based on the received command 2 3 Wireless Communication The wireless communication system consists of 3 main parts Fig 6 Transmitter Module 2 3 1 Transceiv
4. the top plate and the motion is transmitted by a metal bar to the shooter head The roller is a cylinder mounted on two bearings and driven by a dc motor as shown in the picture Fig 1 Our aM generation robot left Mechanical Overview right 2 1 1 Wheels Our new omni wheel is shown in the figure 2 This kind of wheel became very popular in recent years These wheels improve robots friction with field and smooth the omni movement of the robot Fig 2 Our new omni directional wheels 2 1 2 Robots Geometrical Properties Diameter 178 mm Height 143 8 mm Ball coverage due to RoboCup rules 19 Weight 1 8 1 9 Kg 2 1 3 DC Motors We are using 25GA RETARDING GEAR MOTOR which are 12 volt DC motors with gear box 1 11 8 The motor speed is 500rpm that leads to linear speed of 1 5 m s We will use Faulhaber dc motor 2233V024S in the next generation of robots to increase speed and accuracy This will be discussed further in this document 2 1 4 Next Generation We are planning some major improvements which will be implemented in the next generation of mechanical design Weight Reduction In order to improve motor speed and lower inertia forces we are working hard on new materials to reduce total robot s weight Weight reduction also makes angular movement more accurate Removing extra parts would provide more space for better design New DC motors In order to reach higher speed and reducing energy cons
5. AVR microcontroller in the transmitter circuit of the wireless communication board which is used to read the original data from the Control unit computer and to create a packet of processed data for the XTR 434 869 to transmit The serial Rx port of the AVR microcontroller has an interrupt subroutine which is called anytime it receives one byte of data When we receive 30 bytes of data 5 bytes for each robot we apply the bit balancing algorithm to the 30 original data bytes giving out 60 bytes of processed data with equal number of 1s and Os in each byte The rf_transmit function is then called to create a packet containing lms of rectangular pulse signal plus one start byte plus two bytes containing the number of the bytes to be transmitted Then the bit balanced data followed by two ending bytes are added to the created packet The whole packet is sent to the XTR 434 869 TX port using the serial TX port of the AVR microcontroller for transmission Receiver there is an AVR microcontroller on the receiver circuit board of the wireless communication system on each robot which gets the received packet from the XTR 434 869 RX port It then decodes the received packet in order to reconstruct the original data The data is checked for any errors using 3 different procedures before being passed to the next unit 2 3 3 Serial Connection to PC The serial RS232 COM port of the Control unit computer is used to send data to and receive any needed data fr
6. Parsian Amirkabir Univ of Technology Robocup Small Size Team Team Description for Robocup 2007 Amir Amjad Amir Ghiasvand Mohammad Korjani Mani Milani Vali Allah Monajjemi Mohsen Roohani Hamed Gha ed Nia Electrical Engineering Department Amirkabir University of Technology Tehran Polytechnic 424 Hafez Ave Postal Code 158754413 Tehran Iran webmaster parsianrobotic com Abstract This document describes Parsian Robotic Small Size RoboCup team activities for preparing for Robocup 2007 Small Size League It s an overview of all activities done in past one and a half year It s the second time that Parsian is participating in SSL Our robots mechanical and electrical design wireless communication image processing AI algorithms and software are described in this paper 1 Introduction Parsian Robotic is a robotic research group working in Electrical Engineering department of Amirkabir University of technology Tehran Polytechnic After being qualified for RoboCup 2006 Germany This is the second year of Parsian s activity in small size robots field 2 Hardware Architecture 2 1 Mechanical Design Our current robot is a three wheel omni directional robot with angle of 120 deg between wheels The diameter is 60 5mm and the robot consist of two floor of aluminum plate and fastened together by use of three columns with a hole inside where the motor are placed The shooter consists of a regular solenoid that is attached at
7. cifications which is discussed later The more the shutter speed of the camera is the more blur less will be the image that it provides This helps2 the noise reduction algorithm to give a more definite understanding of the playing field It was important to pay attention to the frame rate of the output video of the camera though most of the studied cameras would meet our need Our vision system is now based on two Panasonic 3 Color CCD WV CT460 G Cameras These RGB cameras operate with speed of 30 frames per second The cameras outputs then get captured by use of a standard frame grabber DirectX9 0 SDK DirectShow Libraries then used as the main framework for the image processing software This software is also written in VC In order to identify each robot distinctively we used some different patterns of circles that we painted in one of those 12 colors Each robot is distinctively identified by these colored circles that are placed on top of each robot considering the arrangement of the circles In order to locate the circles of each color we thought of some different ways of processing the matrix of the image among which I will briefly explain two most inspiring ones e One way of locating the circles is to read all the pixels of the grabbed frame without bypassing any important or redundant pixels In this case a chosen pattern that best represents a circle will scan the whole image and find all the circles and finally d
8. e is no real connection between the software and the real world Instead the simulation engine simulates these relations 3 1 2 Manual Control Layer In development phase it s essential to have several tools to test every aspect of robots e g movement passing shooting and co operating These tools are available in this mode The software bypasses strategy layer in this mode and all robots in the field real or simulated can be fully controlled by the operator 3 1 3 Strategy amp Game Play Layer This layer is the most important layer for the act of playing soccer Briefly this layer decides who do what based on the world model and game state It will be fully described further in this document See section 3 2 3 1 4 Task Manager amp Output Commands Generator Layer Outputs of both manual control and strategy layer are Commands these commands first stay in a queue We named this queue Action Queue This layer fully manages this queue for each robot Several algorithms used to reduce the conflicts and increase the efficiency of every robot As every action in queue can have different priority some other algorithms are implemented to choose the best action After deciding the optimum action for each robot this layer finally generates the output commands package and sends the package through serial port to Wireless Communication module 3 1 5 Simulation Engine For making the development easier the simulation
9. er module We use AUREL XTR 434 and AUREL XTR 869 transceiver modules for data transmission and reception The XTR 434 uses the carrier frequency of 433 92 MHz and XTR 869 uses 869 MHz carrier frequency Both work with bandwidth of 150 KHz Both can work in data frequencies between 9600 100 000 bps but it is said not to be used for frequencies less than 57600 when not applying any bit balancing algorithms to the data though the data must be byte balanced otherwise As mentioned the only difference between these modules is the carrier frequency As it s vital for Robocup robots to be able to work in two different carrier frequencies two different modules are used they can be easily get replaced to change the team carrier frequency during matches The XTR 434 869 transceiver can be set to transmitter receiver and power down modes When switching to transmitter or receiver the device needs 2ms to switch to the new mode And for Ims after it should not be used for data transmission but a rectangular pulse signal must be transmitted in the meantime As we did not use the available manufactured PCBs of the XTR 434 869 module we came across some problems using our self made circuit board The correct way of placing the module on a circuit board the antenna connection the ground circuits and the antenna which is used as said in the user s manual is noticeable for better performance 2 3 2 Data Flow Control Transmitter there is an
10. gy In the strategy layer we need not only the current positions but also future game state In order to predict the future we apply gradient method based on fuzzy logic to process the position with the last ten frames Robots behavior changes based on the last position and action command fed to robots and future position of robots and ball The strategy has global and local behavior The global behavior consists of defense attack and stop mode Local behavior includes passing shooting dribbling robot placement and tracing the ball In each global behavior there are several models The strategy chooses the best model in each situation against opponents We use Multiple Hypothesis Testing method in order to overcome the problem For each model the hypothesis that it is the true model is raised and the probability that the hypothesis is correct is evaluated using the observation sequence Over time the most accurate model is assigned the largest probability and so dominates the behavior of estimator The output is a function of estimates made by the different model as well as the probability that each model is correct Following two assumptions are made e The true model is one of those proposed e The same model has been in action since t 0 Over time one model acquires a significantly greater probability than the other model and it corresponds to the model whose parameter values are most appropriate Combine Model 2 estimates
11. objects Robots and the ball By using difference of states positions this system has the Prediction ability So it can predict the ball and opponents robots positions and redefine every robot behaviors based on those parameters After defining every robot behavior by using another Fuzzy system we can convert every robot behavior into the target position it means we can define every robot target by using the calculated behaviors as an input for another fuzzy system Inputs of the Routing System are current robots positions and every robot target position This section would produce every robot path by using Half Field Neural Network in a fuzzy environment Optimizing the path is vital for reducing the robots energy For optimizing the path using half field neural network we use energy minimizing method based on Liapunov stability theorem Finally this layer fills an array with proper decisions for each robot These data then being used by Task Manager Layer 4 Vision System Studying some different documents and data sheets and parameters of different cameras we decided to use a CCD Analogue camera Some important parameters of the camera to use are discussed below The resolution of the camera output video must be at least 640x480 pixels so that the image processing algorithm can give a better and more delicate output The resolution of the grabbed frames sets a minimum limit to the capture device spe
12. om the AVR microcontroller A MAX 232 IC is used to convert the 12 and 12 volts of the COM port to the operating logic voltages of the circuit elements that is 0 and 5 3 Software 3 1 Main Software The main software application is written in Visual C This application performs three main tasks Robots control Strategy Layer Implementation Game Play and Simulation Here is the software general schema i Serial Data Tep IP Manual World Control Task i Model Layer Manager amp SerialData Generator Output amp Low Commands Level Generator Objects Strategy amp Layer Layer Game Play Layer Simulation Engine Fig 7 Software Schema 3 1 1 World Model Generator amp Low Level Objects Layer The main task of this layer is receiving data Image Processing amp Referee Box from real world and defining an accurate and reliable world model based on those data In each program cycle received data first gets verified Then this layer generates the world model This world model consists of robots and ball positions This layer also applies proper transformations to the received data to neutralize direction changes e g after half time This layer then sets unsets several flags for each robot and the game state based on received data Other layers have no access to any other information except what supplied by this layer When running software in simulation mode ther
13. rop the invalid circles found Note that in this procedure quite a lot of CPU usage and memory space will be used This will cause the program not to be able to produce as many new frame data as we need e Another way of locating the circles is to read not ALL the pixels of the image matrix In this case a different pattern will sweep the whole picture but dropping about 95 of the pixels if they don t contain important material This is the procedure we are using now To identify each robot and find the angles needed the algorithm goes through some routine mathematical calculations then Each camera connects to a separate PC Each PC runs the same software to process the image and sends its data through TCP IP to the merging server program This program merges collected data into one pre world model The merged data finally sends the information through TCP IP to the software PC It s obvious from this document that at least three PCs are needed beside our team of robots two for image processing and one for game play These three PCs must be connected togethet through high speed TCP IP network References 1 Siegwart Roland Introduction to Autonomous Mobile Robots 2004 MIT Press 2 Nils J Nilsson Introduction to Machine Learning Department of Computer Science Stanford University
14. s placed in the front the space for the motion is harder to find but the joint is easier to build figure 3 right This motion can be taken out from a four bar mechanism which the shooter freezes the coupler bar thus can lead to appropriate motion figure 4 Another way is that the shooter can be driven by a worm gear so it is auto locked as desired The advantage of using this method of shooting is that all teams predict the direction of shooting by drawing a line using center of robot and ball and the keeper and defenses location are adjusted according to that and this method can mislead the opponent robots especially in penalty and free kicks 2 1 5 Chip Kick As many teams are using chip kicks in small size competition we decided to develop that system for some are our robots as discussed before we should choose to use One method between rotational shooting and chip kick shooting because they can not come together in one robot so we decided to use the chip kicking for our defenses and rotational shooting for forwards Our chip kick mechanism is to lift the ball up just before the shoot it is a very common way for chip kicks The final decision about chip kick will be made after making and testing the rotational shoot foint at center le of motion Fig 3 New Method for passing a Fig 4 4 bar mechanism for moving the shooter 2 2 Electrical Design Electronic circuit consists of three
15. umption in our robots we decided to change our dc motors We will use Faulhaber Motors 2224U006SR for wheel drive and 2342SO06CR for spinner like many other teams these motors are reasonable choices because of high power and small size The main motor is a 9000rpm dc motor which is needed to be used with spur gears to reduce the speed By using different gears for keeper defenses and forwards we can achieve speeds needed for every position for example keeper needs higher speed comparing to forwards New pass and shoot method This year mechanical group decided to develop a new method of shooting and passing that has not been ever used as far as we know The main idea is to separate the angle of the robot and the angle of shooting system in limited range of variation We call this method rotational shooting The idea is to make a box consisting of the shooter and the spinner and then separating the shooter from this box by giving it one additional rotational degree of freedom This rotational joint can be placed in two different positions and thus can be built by two different strategies first to place the joint in the front of the robot and second to place the joint in the center of it By placing the joint in the center of robot the space occupied due to motion of shooter is at the front where it s easer to find that space but building this joint in the center of robot is much more difficult figure 3 left But if the joint i
Download Pdf Manuals
Related Search