Using Improved Object Detection in Robocup Soccer for Collision Avoidance and Recovery

RoboCup legged soccer is an international soccer competition between autonomous, four-legged robots. In past competitions, collisions between robots were common. My objective, therefore, was to augment the robots’ sensor systems to allow them to avoid and recover from such collisions; this would both speed up the game and enable the robots to dribble the ball around other players. I have increased the accuracy and stability of the ball’s position estimation and have added a calculation of the ball’s relative velocity. Using this velocity, I have developed an algorithm to determine if a robot has collided while trying to get to the ball; if the velocity remains near zero for too long, the robot will try another path to the ball. I have also greatly increased the accuracy of the detection of other robots and have added the ability to store the positions of all other players on the field. A dribbling robot will then use these position estimates for collision avoidance, using the method of artificial potential fields. I describe a number of modifications that were necessary for this method to work well. Finally, for both the ball and other players, I have incorporated readings from the robots’ infrared distance sensors, and have added an update of the relative position of an object based on the observer’s own movements.