Mark Simon

Robust Real Time Color Tracking

This paper describes the vision system that was developed for the RoboCup F180 team FU-Fighters. The system analyzes the video stream captured from a camera mounted above the field. It localizes the robots and the ball predicting their positions in the next video frame and processing only small windows around the predicted positions. Several mechanisms were implemented to make this tracking robust. First, the size of the search windows is adjusted dynamically. Next, the quality of the detected objects is evaluated, and further analysis is carried out until it is satisfying. The system not only tracks the position of the objects, but also adapts their colors and sizes. If tracking fails, e.g. due to occlusions, we start a global search module that localizes the lost objects again. The pixel coordinates of the objects found are mapped to a Cartesian coordinate system using a non-linear transformation that takes into account the distortions of the camera. To make tracking more robust against inhomogeneous lighting, we modeled the appearance of colors in dependence of the location using color grids. Finally, we added a module for automatic identification of our robots. The system analyzes 30 frames per second on a standard PC, causing only light computational load in almost all situations.

Predicting Away Robot Control Latency

This paper describes a method to reduce the effects of the system immanent control delay for the RoboCup small size league. It explains how we solved the task by predicting the movement of our robots using a neural network. Recently sensed robot positions and orientations as well as the most recent motion commands sent to the robot are used as input for the prediction. The neural network is trained with data recorded from real robots. We have successfully field-tested the system at several RoboCup compe- titions with our FU-Fighters team. The predictions improve speed and accuracy of play.

Reinforcing the Driving Quality of Soccer Playing Robots by Anticipation Verbesserung der Fahreigenschaften von fußballspielenden Robotern durch Antizipation

Summary This paper shows how an omnidirectional robot can learn to correct inaccuracies when driving, or even learn to use corrective motor commands when a motor fails, whether partially or completely. Driving inaccuracies are unavoidable, since not all wheels have the same grip on the surface, or not all motors can provide exactly the same power. When a robot starts driving, the real system response differs from the ideal behavior assumed by the control software. Also, malfunctioning motors are a fact of life that we have to take into account. Our approach is to let the control software learn how the robot reacts to instructions sent from the control computer. We use a neural network, or a linear model for learning the robots response to the commands. The model can be used to predict deviations from the desired path, and take corrective action in advance, thus increasing the driving accuracy of the robot. The model can also be used to monitor the robot and assess if it is performing according to its learned response function. If it is not, the new response function of the malfunctioning robot can be learned and updated. We show, that even if a robot loses power from a motor, the system can re-learn to drive the robot in a straight path, even if the robot is a black-box and we are not aware of how the commands are applied internally.

Parabolic Flight Reconstruction from Multiple Images from a Single Camera in General Position

This paper shows that it is possible to retrieve all parameters of the parabolic flight trajectory of an object from a time stamped sequence of images captured by a single camera looking at the scene. Surprisingly, it is not necessary to use two cameras (stereo vision) in order to determine the coordinates of the moving object with respect to the floor. The technique described in this paper can thus be used to determine the three-dimensional trajectory of a ball kicked by a robot. The whole calculation can be done, at the limit, with just three measurements of the ball position captured in three consecutive frames. Therefore, this technique can be used to forecast the future motion of the ball a few milliseconds after the kick has taken place. The computation is fast and allows a robot goalie to move to the correct blocking position. Interestingly, this technique can also be used to self-calibrate stereo cameras.

The Soul of A New Machine: The Soccer Robot Team of the FU Berlin

This paper describes the hardware and software of the robotic soccer team built at the Freie Universitat Berlin which took part in the 2000 RoboCup Championship in Melbourne, Australia. Our team, the FU Fighters, consists of five robots of less than 18 cm horizontal cross-section. Four of the robots have the same mechanical design, while the goalie is slightly different. All the hardware was designed and assembled at the FU Berlin. The paper describes the hierarchical control architecture used to generate the behavior of individual agents and the whole team. Our reactive approach is based on the dual dynamics framework proposed by Jager, but extended with a third module of sensor readings. Fast changing sensors are aggregated in time to form slowly changing percepts in a temporal resolution hierarchy. We describe the main blocks of the software and their interactions.