Skuba 2007 Team Description

Transcription

1 Skuba 2007 Team Description Jirat Srisabye 1,1, Napat Parkpien 1,1, Poom Kongniratsiakul 1,1, Phachachon Hoonsuwan 1,2, Saran Bowarnkitiwong 1,1, Marut Archawananthakul 1,1, Ratchai Dumnernkittikul 1,1, Santi Chongkaonar 1,1, Anuchit Ratanaparadorn 1,1, Chayaporn Keawpromman 1,1, Varut Limnirunkul 1,1, and Yodyium Tipsuwan 1,1 1 Department of Computer Engineering, 2 Department of Electrical Engineering, Faculty of Engineering, Kasetsart University, 50 Phaholyothin Rd, Ladyao Jatujak, Bangkok 10900, jirat_o@hotmail.com, Abstract. In this paper we describe the Skuba Small-Size League team, which are designed to meet the rules for participation in the RoboCup World Championship 2007 in Atlanta, USA. The overview describes both the robot hardware and the overall software architecture of our team. Keywords: Small-size, Robocup, Vision, Robot Control, Artificial Intelligence. 1 Introduction The Skuba team has started its work at Intelligent Mechatronic Laboratory in Kasetsart University since Our team had participated at the small size league of Robocup World Championship 2006 in Bremen, Germany. In the last participation, we gained valuable experience and could evaluate our team s strength and weakness points. RoboCup is an international joint project to promote AI, robotics and computer vision. In the Small Size League, two teams of five robots, which are 18 cm in diameter, play soccer on a 4 by 5.4 m carpeted soccer field. We have a team in the Small Size League (SSL) with four main components: the vision system, the AI system, five robots and the referee box. The vision system process two video signals from the cameras mounted on top of the field. It computes the positions and the orientations of the ball and robots on the field then transmit the information back to the AI system.

2 The AI system receives the information and makes strategic decisions. The decisions are converted to commands that are sent back to the robots via a wireless link. The robots execute these commands and set actions as ordered by the AI system. In this paper we analyze these some main topics. In the following section we give an overview of our robot base focusing on the mechanical and electrical. In the third section we describe the vision system. And in the fourth section we explain our AI architecture. 2 Robots In this section, we describe the overview of both our robots mechanical and electrical design. 2.1 Mechanical design The Skuba's robots have a mass of 2.8 kilograms (include battery). The robots which we used at RoboCup 2006 in Bremen are shown in figure 1. Currently, we have done some modifications to the mechanical structure of robots for better ball handling and stronger kicking system. Fig.1. Skuba s 2006 Robot Kicker design We combine two kicking mechanisms in a small area as seen in Figure 2. One kicking mechanism kicks the ball flat across the field. The second kicking mechanism shifts the ball over obstacles. Both two mechanisms are driven by push type solenoid. This year we have modified the surface of the kick face to increase the kicking precision and we add power of kick by increase capacitor size to 9900UF 250VDC.

3 Fig.2. Skuba s new kicker design Omnidirectional wheels Omnidirectional wheels allow robot to drive on a straight path without pre rotation. Our omnidirectional are shown in figure 3. Fig.3. our omnidirectional wheel design (left) and four wheeled robot (right). We use four omnidirectional wheels in each robot. Wheel positions are located at 33,147,225 and 315 degree of each robot, respectively. To drive a robot, we use four DC motors (Faulhaber 6V 2224SR [1]) with external gear head reduction ratio of 13.5:1 along with four quadrature encoders with 512 pulses per revolution Dribbling design The Skuba's dribbling device is a rotating silicone cylinder. A 6 Volt 2224 Faulhaber drives the shaft onto which the silicone is wrapped. A 4.3:1 internal gearbox is used between the motor and the shaft.

4 2.2 Electrical design Last year, the major problem was slipping wheels. We determined that we wanted to solve the problem of robot control by using local sensing of the robot. We use a rate gyroscope to measure rotational velocity and one dual axis accelerometer to measure two degrees of translational acceleration. Our local control loop diagram is shown in figure 4. Fig.4. Local control loop diagram Rate Gyroscope We use the rate gyro, Analog Devices [2] ADXRS300 to measure the angular velocity of the robot. We designed the Butterworth low-pass filter [3] with 3rd filter order and connect the signal to main microcontroller A/D input Accelerometer We use an Analog Devices ADXL202E accelerometer, which have two axes for x and y directions. The accelerometer output is two sets of pulse that represent acceleration in each direction. The time associated with the pulses that indicate acceleration which are read by two input capture pins on the main microcontroller Microcontroller board The microcontroller board uses five Microchip's dspic30f2010 [4] 16 bit microcontroller. We use four microcontrollers to execute the low level motor control loop and use one to executes the local sensing control loop. The dspic30f2010 MCU has several peripheral such as output comparator, 12-bit ADC, UART, SPI, and QEI (Quadrature Encoder Interface). The frequency used in our system is MHz but we multiply it by phase lock-loop parameter that set to 16 (118 MHz).

5 2.2.4 Wireless communication Wireless communication is controlled by two Radiometrix BIM and BIM [5] transceivers with radio frequency at either 914MHz or 433MHz. The transceiver module is a self-contained plug-in radio incorporating a 64kbit/s packet controller with a serial port interface. Furthermore, our main MCU board can use a TRF-2.4GHz [6], which can support up to 125 different communication channels Batteries The Skuba robots are powered by Sanyo NIMH [7] (Nickel Metal Hydride) 5000 mah 12.0V batteries. The batteries are split into 2 sets of ten cells each. The robots are currently able to last 30 minutes on one set of batteries. Fig.5. CAD model of Skuba s 2007 robot

6 2 Vision system The main problem with the Skuba s vision system 2006 was the latency that is the time it takes from when an image is first captured until object information from that image is sent to the AI computer. Testing revealed that the Skuba vision 2006 system had a latency of 3-5 frames (about 150 milliseconds). Currently, we solve the latency problem by filtering and prediction of vision data. Our vision structure diagram is shown in figure 6. Fig.6. Skuba s vision structure 2.1 Vision Client Capture Device. The Skuba vision apply the global vision and use the output signal of a top-mounted PAL camera as the input signal of a low cost capture card. We employ 3CCD camcorder which is capable of grabbing 640 x 480 images at 50 Hz (even and odd line). Preprocessing. The preprocessing is used to improve the quality of the image by image processing filter.

7 Transform Color Space. The image that is captured by capture device is RGB space, which is a common format for image display and manipulation, and is provided directly by most video capture hardware. It is the main problem lies in the intensity value of light and shadow being spread across all three parameters. This makes it difficult to threshold. We transform color model to the HSV space, which consists of a hue, a saturate and a value. The HSV space is more stable than RGB space in different light properties. Color Segmentation. The color segmentation assigns each image pixel into color classes. Our approach is single-pixel classification into discrete classes [8]. Object localization. After color segmentation, we receive all the color regions. The filtering process discards incorrect regions. Then, object localization computes the position and orientation of objects in the field from the final regions. Local object tracker. We acquire all vision objects from object localization. Then, all objects are tagged with coordinates and other properties to simplify process after that. Our approach is working by comparing the latest data to the current data, and uses the nearest data by Euclidean distance. Fig.7. Skuba s new vision client

8 2.2 Vision Server Matching vision data. The matching vision data use for merge and match vision data from two vision clients. Filter vision data. Vision data which is receive from vision client has a lot of noise, so we need to filter it. Our approach is working by Kalman s filter [9] to decrease noise. Global object tracker. The Global object tracker has a same function as the local object tracker, but it tracks final data from two vision clients. Predict vision data. Currently, our approach is a linear predictor [10] that is a simple method. In addition, we plan to use the neural network prediction [10] instead because our robot model is non-linear. Transmit to AI. This component consists of network link used for communication between the vision system and the AI system that process on separate PC. 2.3 Camera Calibration Camera calibration is a part in Object localization. We compute the internal and external parameters of the cameras using the Tsai [11] algorithm. These parameters are used to correct the distortion produced by the camera lenses. Fig.8. Camera calibration toolkits

9 3 AI Structure Skuba has a hierarchical model in our AI structure. The game receives vision data and referee command then selects a play which relate with referee command. Whenever a play is executed it calls role that is the action functions for all positions present. The role functions then run skills for the related robots. Fig.9. Execute hierarchy 3.1 Path planning We use a modified potential field method [18] for robot navigation. The potential field method of avoiding obstacles consists of evaluating a repulsive force for each obstacle. The attractive force, that tends to drive the robot to its target, accelerates the robot towards its target while the repulsive forces accelerate in the opposite direction of the obstacles.

10 3.2 Simulation A simulator is developed in order to develop hardware and software simultaneously. The simulator receives a sequence of packets that is identical to packets a sequence of packets that is sent to robots. The simulator then calculates some simple physics and returns the coordinate of objects in the field to the software as same as the vision system does. Fig.10. Visualization of the simulator 4 Conclusion The new hardware and software design has improved the speed, precision, and flexibility of our robots. With some filters, we could acquire precisely coordinates of all players. The simulator can be used efficiently to help developing both hardware and software. Skuba team is currently participating in Robocup Thailand Championship 2007, which is arranged in January. We hope our team would be qualified so we can play and share experience with other teams around the world.

11 References 1. Faulhaber Product Analog Devices Product Low pass Filter Design Microchip Product Radiometrix Product Electronic Source Product Sanyo s Battery Product Bruce, J., and Veloso, M.: Fast and accurate vision-based pattern detection and identification. In: Proceedings of the IEEE International Conference on Robotics and Automation, Taiwan (2003) 9. C. H. Messom, G. Sen Gupta, S. Demidenko and Lim Yuen Siong : Improving Predictive Control of a Mobile Robot: Application of Image Processing and Kalman Filtering. In: Proceedings of IMTC Instrumentation and Measurement Technology Conference Vail, CO, USA, May Sven Behnke, Anna Egorova, Alexander Gloye, Raúl Rojas, and Mark Simon.: Predicting away robot control latency. In: Proceedings of 7th RoboCup International Symposium, Padua, Italy, Tsai, R.Y.: A versatile camera calibration technique for high accuracy 3D machine vision using off-the-shell TV cameras and lenses. IEEE Journal of robotics and Automation, Simon, M., Behnke, S., and Rojas, R.:"Robust Real Time Color Tracking", P. Stone, T. Balch, G. Kraetzschmar (Eds.): RoboCup 2000: Robot Soccer. World Cup IV Spinger Verlag 2001, pp Zickler, S., Licitra, M.: RoboCup SSL 2005 Team Description: Wingers (University at Buffalo) Bruce, J., Zickler, S., Licitra, M. and Veloso M.: CMRoboDragons 2006 Team Description RoboCup 2006 symposium CDROM, team descriptions, (2006.6) 15. Bruce, J., Balch, T. and M. Veloso: Fast color image segmentation for interactive robots. In: Proceedings of the IEEE Conference on Intelligent Robots and Systems (IROS), Japan (2000) 16. Johannson, S.J., Saffiotti, A.: Using the Electric Field Approach in the RoboCup Domain. In Birk, A., Coradeschi, S., Tadokoro, S., eds.: RoboCup 2001: Robot Soccer World Cup V. Volume 2377 of Lecture Notes in Artificial Intelligence., Springer (2002) 17. Khatib, O.: Real-time Obstacle Avoidance for Manipulators and Mobile Robots. The International Journal of Robotics Research 5 (1986) 90 98) 18. Y. Koren and J. Borenstein: Potential Field Methods and Their Inherent Limitations for Mobile Robot Navigation. Proceedings of the IEEE Conference on Robotics and Automation, Sacramento, California, April 7-12, 1991, pp Tucker Balch and Ronald Arkin: Avoiding the Past: A Simple but Effective Strategy for Reactive Navigation. International Conference on Robotics and Automation, Tsai, R.Y., A versatile camera calibration technique for high-accuracy 3D machine vision metrology using off-the-shelf TV cameras and lenses. IEEE Journal of Robotics and Automation, : p Bruce, J.: CMVision realtime color vision system. (The CORAL Group s Color Machine Vision Project) jbruce/cmvision/.

12 22. Y.P. Leow, K.H. Low and W.K. Loh Kinematic Modeling and Analysis of Mobile Robot with Omni Directional Wheels Proceedings of the Seventh International Conference On Automation, Robotics, Control And Vision, Singapore (2002) 23. Termtanasombat, N., Wongsaisuwan1, M., Chongstitvatana, P., Wannasuphoprasit, W.: Plazma-Z 2006 Team Description RoboCup 2006 symposium CDROM, team descriptions, (2006.6) 24. Velázquez1, Monserrat Muñoz and Alfredo Weitzenfeld.: EagleKnights 2006 Team Description RoboCup 2006 symposium CDROM, team descriptions, (2006.6) 25. E. Brookner, Tracking and Kalman Filtering Made Easy. A Wiley-Interscience Publication, Kalm`ar-Nagy, T., Ganguly, P., D Andrea, R.: Near-Optimal Dynamic Trajectory Generation and Control of an Omnidirectional Vehicle, submitted to International Journal of Robotics Research 27. D Andrea, R., Kalm`ar-Nagy, T., Ganguly, P., Babish, M.: The Cornell RoboCup Team, Stone P., Balch T., Kraetzschmar (Eds), Robocup 2000: Robot SoccerWorld Cup IV, Springer Verlag, Berlin, Muioz, V., Ollero, A., Prado, M., Simon, A.,Mobile robot trajectory planning with dynamics and kinematics constraints, in Proceedings of the IEEE lntemational Conference on Robotics and Automation, pp , 1994.