Robocup Electrical Team 2006 Description Paper Name: Strive2006 (Shanghai University, P.R.China) Address: Box.3#,No.149,Yanchang load,shanghai, 200072 Email: wanmic@163.com Homepage: robot.ccshu.org Abstract: This paper describes the Robocup Electrical Team 2006. We are working hard and preparing for the Small-Size (F180) League Robocup competition 2006 in Bremen. This paper gives an over view about our current team, including both the hardware design of the robot vehicle and the software system. 1) Introduction: The Robocup Electrical Team of Shanghai University established in 2003. We have already developed 4-wheel robot vehicles ourselves and participated in many local games. We obtained good results and acquired very precious experiences as well. We have prepared a long time for this competition and developed a new version of our robot vehicles. The objective of our Robocup electrical team 2006 is to become a competitive Robocup team and to participate in the Robocup World championship 2006 in Bremen. 2) System Overview: 2.1) Hardware Design: Our system is based on DSP,FPGA,PIC chip. The communication is controlled by PIC which deciding the configuration and status of the nrf2401 chip. The flow of data is sent to the FPGA. We establish dual RAM with the FPGA, it supplys a public buffer so as to the communication between the DSP and PIC. This structure reduces the overall burden of system efficaciously. The motor is driven by 4AM11 and the driver logical control is set up by CMOS combinational logic circuit. The flow of control is to be the dual-closed loop configuration which constitutes rotate speed loop and electric current loop. The module of the examining rotate speed is used by the motor s incremental shaft encoder of itself. We use the voltage convert frequency module to examining current. The kick-ball module is that the robot kicks off from the energy of the high voltage. The high voltage is gained by the conventional boost converter. The whole controlled module is separated from the driver module and high voltage module by the opyocoupler. 2.2) Vision System: The vision system is composed of two cameras, two grabbers, a personal computer and an image information-processing program. Each camera is equipped with a wide-angle lens in order to get and output the whole field information. The grabbers are set to capture the image at 50 fields per second. The library of the grabber has many useful functions which can be used to get information about pictures in different formats. When the vision system starts working, the program will set a timer whose frequency 1
is 50Hz. The color format of the input image is YUV422, for the grayscale image information can be gotten directly. Two threads are run to search for robots and ball respectively. In the first thread, robot edge detection will be used to convert these grayscale images to binary images. Then the mask that has already been created will be placed at every position of the field. Ten positions that match the mask best are determined the positions of robots. According to the color of these positions center, the robots will be isolated by team. The blobs around the team label can be found after sweeping around the center of our robots. Based on these, the ID and orientation of our robots can be identified. In the other thread, the ball is searched outwardly with the spiral way of rectangle from where it was identified last time until it is found. After getting all the information which is need of the eleven objects, it will be transferred to the decision system. Image distortion is ineluctable because of the wide-angle lens and the deflection of the cameras. So barrel distortion correction and pitch correction are adopted. Moreover, since the robot is much taller than the ball, their positions in the images exists relative error. The approach to deal with the problem is height correction to decrease errors. 2.3) Mechanical system design: Construction diagrams from different directions: Fig1 vehicle The 2006 mechanical design team was established for the purpose of researching, designing and building a team of autonomous robots for the RoboCup Small-Sized League 2006 competition. This year we also use omni-direction vehicles. Of course we have improved a little in our chassis design. For example, we protected our wheel from impacting with other robots and we also decrease the thickness of our aluminum board. Our driven motor is Faulhaber DC 2224U micro-motor. It ensures our vehicle move at the speed of 3m/s and have an acceleration of 8m/s2. This year, we have done a lot of experiment in our ball-control mechanism which include dribbling and ball-kicking. We choose different dribbler material, different roller shape, and change the velocity of the rolling bar. Finally, we get the perfect dribbling mechanism. In ball-kicking experiment, we use two solenoids, one is for kicking the other is for chipping. When this mechanism hits 2
the ball above straightly, the ball can reach a height of 120cm which is quite enough. The result of chipping experiment is also much better than we previously estimated. We can chip the ball over other robots although they are very close. Now we are trying to adjust the distance and height of the ball when we chip the ball. This time the EE group decide to use a huge capacitance which occupied a great deal of space. The battery also need much room, so it is not easy to assemble all the parts in regular volume. Luckily, we made it with the help of Pro/Engineer. 2.4) Strategy: This section describes the main strategy flow of the game. Our strategy bases on a dynamic role allotment system. We can get all the information we need from our vision system: the robots and ball s positions, both the values and directions of their moving speeds. As for the robots, their rotations must be captured as well. Through the information above, we will decide which play modes we should take, attack or defend? And then our strategy system will assign a basic role(such as chief-attacker, offensive assister, blocker, defender, goalie etc.. ) from a role database that includes more than 10 different roles to individual robot. In our strategy, these roles of the robots are not fixed, they are probably to be converted to each other. Furthermore, we use a role-taking competing system to raise the efficiency of the tactics being carried out. This method avoids a completely tactics failure due to a mission failure of a single robot(in consideration of the differences between different robots and interferences from opponent robots). In role-taking competing system, we put forward a concept of chief-role candidate. For example, when in attack mode, except some necessary roles for defend, other robots will not be assigned roles at the beginning. They are all chief-role candidates. At first, they all run as quickly as they can to reach the most important aim position the strategy system has calculated. If one of these candidates has obvious advantage of taking that role, it will be assigned the role, other robots roles will be assigned according to the strategy then. However, when these candidates are about to reach the position and none of them has the obvious advantage of taking the current role, our computational system will calculate the trajectories of them from their current positions to the same aim position. By the comprehensive comparison of both the lengths and complications of their trajectories, we finally assign each robot a basic role. In addition to some basic actions, different roles have their special moves. Usually, the computational system executes the chief-attacker go to chase the ball. If the chief-attacker has already got the ball in its control, the system will ask the robot execute the following commands: pass, dribble, or shoot action. And as to the offensive assister, whether it should go to help blocking the opponent robots or go straight to some advantaged position to render assistance to the chief-attacker. The strategy loop runs on every 25ms which is determined by the vision hardware. So we have a sub system help to interfere the dynamic role allotment system to repress the roles to be converted too multifarious. And this can ensure the coherent of a robot s movement and the stability of the whole system. All of the above actions will keep repeating throughout the game play. 3
2.5) Improvement for APF: Whether the path planning of the robot vehicle is good or not has a far-reaching influence on the whole system performance. Our study has been focus on the disadvantages of the path planning based on traditional artificial potential field(apf), including the local minima, oscillations in the presence of obstacles. We put forward a method to improve it. Before calculating the resultant force that is put on an obstacle in the potential field, we programme all the obstacles in order to optimize the path to be planned. In the traditional APF, every obstacle is programmed individually. So sometimes the path planned is not satisfied. To a certain robot, we divide its obstacles into two groups: visible or invisible. The location of the robot and obstacle A, B, C and D are shown in figure 1. A is very close to B. (the maximum distance between two obstacles which could be linked is set) Connect robot and A, robot and B, then prolong the line along the direction to the obstacle. We could find that C is located in the shadow which is surrounded by line AB and the two lines mentioned above. We regard C as invisible to the current robot s location, thus obstacle A, B and D is visible. A C ROBOT B D Fig2 visible and invisible We should pay more attention to visible obstacles. Our method is to build a link between two close visible obstacles. Linked obstacles are regarded as a whole. Simulations: 1) Fig3-1 sim1 Fig3-2 sim1 The path planned in Fig3-1 is trap in local minima between obstacle 7 and 3. We build a link between obstacle 7 and 3, obstacle 3 and 1. These three obstacles are regarded as a whole. The robot vehicle reaches its goal point successfully which shows in Fig3-2. 4
2) Fig4-1 sim2 Fig4-2 sim2 In Fig4-1, the path oscillations in the presence of obstacles. In Fig4-2, after linking, obstacle 2, 4 and 6 are regarded as a whole. Obstacle 5 is invisible. The robot vehicle reaches its goal point fluently. 3) Fig5 sim3 Fig6 sim4 Fig5 shows the robot vehicle skillfully avoid a trap formed by obstacle 1,2 and 7. Fig6 shows a very complicated link-net which is buildup by five obstacles. While carrying out some tactics, such path planning is very effective. Artificial potential field method is usually used for obstacle avoidance for mobile robots. Comparing with the previous methods, its algorithm is simple, direct, and its computing amount is small, can satisfy the highly practical request in robot football system. This method resolve the problem of local minima and improve the quality of the path to be planned. 3) Summary: This new version of our robot vehicle has been developed over one year. Every member of our team has given great enthusiasm and made much effort on his or her work. We are very confident with our robot and wish to show it in the coming competition. 5