Файл: Embedded Robotics (Thomas Braunl, 2 ed, 2006).pdf

Football, or soccer as it is called in some countries, is often referred to as “the world game”. No other sport is played and followed by as many nations around the world. So it did not take long to establish the idea of robots playing soccer against each other. As has been described earlier on the Micro Mouse Contest, robot competitions are a great opportunity to share new

ideas and actually see good concepts at work.

Robot soccer is more than one robot generation beyond simpler competitions like solving a maze. In soccer, not only do we have a lack of environment structure (less walls), but we now have teams of robots playing an opposing team, involving moving targets (ball and other players), requiring planning, tactics, and strategy – all in real time. So, obviously, this opens up a whole new dimension of problem categories. Robot soccer will remain a great challenge for years to come.

18.1 RoboCup and FIRA Competitions

263263

18 Robot Soccer

264

RoboCup and FIRA Competitions

will change the environment (for example kick the ball) and thereby render one’s own action plans useless if too slow.

One of the frequent disappointments of robot competitions is that enormous research efforts are reduced to “show performance” in a particular event and cannot be appreciated adequately. Adapting from the home lab environment to the competition environment turns out to be quite tricky, and many programs are not as robust as their authors had hoped. On the other hand, the actual competitions are only one part of the event. Most competitions are part of conferences and encourage participants to present the research behind their competition entries, giving them the right forum to discuss related ideas.

Mobile robot competitions brought progress to the field by inspiring people and by continuously pushing the limits of what is possible. Through robot competitions, progress has been achieved in mechanics, electronics, and algorithms [Bräunl 1999].

CIIPS Glory with local vision on each robot

Note the colored patches on top of the Lilliputs players. They need them to determine each robot’s position and orientation with global vision.

Figure 18.1: CIIPS Glory line-up and in play vs. Lilliputs (1998)

265

18 Robot Soccer

18.2 Team Structure

The CIIPS Glory robot soccer team (Figure 18.1) consists of four field players and one goal keeper robot [Bräunl, Graf 1999], [Bräunl, Graf 2000]. A local intelligence approach has been implemented, where no global sensing or control system is used. Each field player is equipped with the same control software, only the goal keeper – due to its individual design and task – runs a different program.

Different roles (left/right defender, left/right attacker) are assigned to the four field players. Since the robots have a rather limited field of view with their local cameras, it is important that they are always spread around the whole field. Therefore, each player’s role is linked to a specific area of the field. When the ball is detected in a certain position, only the robot responsible for this area is meant to drive toward and play the ball. The robot which has detected the ball communicates the position of the ball to its team mates which try to find favorable positions on the field to be prepared to take over and play the ball as soon as it enters their area.

Situations might occur when no robot sees the ball. In that case, all robots patrol along specific paths in their assigned area of the field, trying to detect the ball. The goal keeper usually stays in the middle of the goal and only moves once it has detected the ball in a reasonably close position (Figure 18.2).

y

x

Figure 18.2: Robot patrolling motion

This approach appears to be quite efficient, especially since each robot acts individually and does not depend on any global sensing or communication system. For example, the communication system can be switched off without any major effects; the players are still able to continue playing individually.

266

Mechanics and Actuators

18.3 Mechanics and Actuators

According to the RoboCup Small-Size League and FIRA RoboSot regulations the size of the SoccerBots has been restricted to 10cm by 15cm. The height is also limited, therefore the EyeCon controller is mounted on a mobile platform at an angle. To catch the ball, the robot has a curved front. The size of the curved area has been calculated from the rule that at least two-thirds of the ball’s projected area must be outside the convex hull around the robot. With the ball having a diameter of approximately 4.5cm, the depth of the curved front must be no more than 1.5cm.

The robots are equipped with two motorized wheels plus two casters at the front and back of the vehicle. Each wheel is controlled separately, which enables the robots to drive forward, backward, as well as drive in curves or spin on the spot. This ability for quick movement changes is necessary to navigate successfully in rapidly changing environments such as during robot soccer competitions.

Two additional servo motors are used to activate a kicking device at the front of the robot and the movement of the on-board camera.

In addition to the four field players of the team, one slightly differing goal keeper robot has been constructed. To enable it to defend the goal successfully it must be able to drive sideways in front of the goal, but look and kick forward. For this purpose, the top plate of the robot is mounted at a 90° angle to the bottom plate. For optimal performance at the competition, the kicking device has been enlarged to the maximum allowed size of 18cm.

18.4 Sensing

Sensing a robot’s environment is the most important part for most mobile robot applications, including robot soccer. We make use of the following sensors:

•Shaft encoders

•Infrared distance measurement sensors

•Compass module

•Digital camera

In addition, we use communication between the robots, which is another source of information input for each robot. Figure 18.3 shows the main sensors of a wheeled SoccerBot in detail.

Shaft encoders The most basic feedback is generated by the motors’ encapsulated shaft encoders. This data is used for three purposes:

•PI controller for individual wheel to maintain constant wheel speed.

•PI controller to maintain desired path curvature (i.e. straight line).

•Dead reckoning to update vehicle position and orientation.

267

18 Robot Soccer

The controller’s dedicated timing processor unit (TPU) is used to deal with the shaft encoder feedback as a background process.

Figure 18.3: Sensors: shaft encoder, infrared sensors, digital camera

•Avoid collision with an obstacle.

•Navigate and map an unknown environment.

•Update internal position in a known environment.

268

Image Processing

18.5 Image Processing

Vision is the most important ability of a human soccer player. In a similar way, vision is the centerpiece of a robot soccer program. We continuously analyze the visual input from the on-board digital color camera in order to detect objects on the soccer field. We use color-based object detection since it is computationally much easier than shape-based object detection and the robot soccer rules define distinct colors for the ball and goals. These color hues are taught to the robot before the game is started.

The lines of the input image are continuously searched for areas with a mean color value within a specified range of the previously trained hue value and of the desired size. This is to try to distinguish the object (ball) from an area similar in color but different in shape (yellow goal). In Figure 18.4 a simplified line of pixels is shown; object pixels of matching color are displayed in gray, others in white. The algorithm initially searches for matching pixels at either end of a line (see region (a): first = 0, last = 18), then the mean color value is calculated. If it is within a threshold of the specified color hue, the object has been found. Otherwise the region will be narrowed down, attempting to find a better match (see region (b): first = 4, last = 18). The algorithm stops as soon as the size of the analyzed region becomes smaller than the desired size of the object. In the line displayed in Figure 18.4, an object with a size of 15 pixels is found after two iterations.

(a)

(b)

269