Cornell Big Red: Small-Size-League Winner

systems engineering concepts and practices to students to prepare them for designing, integrating, and maintaining highly complex systems. Another objective of the project is to explore the interplay between AI, dynamics, and control theory. This article describes the Cornell RoboCup team, which won the RoboCup-99 small-league championship in Stockholm, Sweden. T he Cornell RoboCup Project was created to teach systems engineering (Blanchard and Fabrycky 1997) concepts and practices to students to prepare them for designing, integrating, and maintaining highly complex systems. Another objective of the project is to explore the interplay between AI, dynamics, and control theory. This article describes the Cornell RoboCup team, which won the RoboCup-99 small-league championship in Stockholm, Sweden. The article is organized as follows: We first describe the overall system, followed by the AI architecture used in the competition. We then discuss the high-fidelity simulation environment used throughout development, followed by our conclusions. The Cornell University team consists of two mechanical designs, one for the field players and one for the goalie (figure 1). All robots have a unidirectional kicking mechanism powered by a solenoid (two for the goalie). The kicking mechanism can impart a maximum velocity of two meters a second relative to the robot. The field players have a design mass of 1.5 kilograms, a maximum linear acceleration of 5.1 m/s 2 , and a maximum linear velocity of 2.5 meters a second. The goalkeeper has a different design from the field players; it is equipped with a holding and kicking mechanism that can catch a front shot on goal, hold it for an indefinite amount of time to find a clear pass to a teammate, and execute the pass. The main function of the on-board electronics is to receive left and right wheel velocity signals by wireless communication and implement local feedback control to ensure that the wheel velocities are regulated about the desired values. The on-board electronics also control the robot kicking mechanism. The global vision system runs at a speed of 35 hertz with a resolution of 320 ؋ 240. A dedicated vision workstation identifies the ball and robot locations as well as the orientation of the robots. The basic algorithm used is blob analysis (Gonzalez and Woods 1992). To determine the identity of each robot and its orientation , the robots have color patches on top as well as the team color marker (blue or yellow Ping-Pong ball). The …