 Robotic Soccer is an exciting and challenging research domain which involves multiple agents that need to collaborate in an adversarial environment to achieve specific objectives. At Carnegie Mellon University we are continuing our development of CM United, a team of small robotic soccer agents that are used as a platform to investigate real-time perception, decision-making, execution, and learning. CM United 97 was the Robocup small-size robot champion after winning the five-team competition held in Japan. At Robocup 98 in Paris, CM United faced a significantly more advanced field of competition, involving 11 teams. CM United 98 won the competition to remain as the Robocup small-size champion. The CM United Robotic Soccer team is a fully autonomous system consisting of the physical robots, a global vision system, and multiple decision-making modules. The core of the team are the robotic agents, which we designed and built for CM United 98. The robots have a robust mechanical frame, limited onboard processing, and radio communication. The robot also is equipped with a kicking device that can propel the ball with some added velocity. Perception is achieved by a color-based vision algorithm which processes images of the complete field. CM United successfully tracks multiple moving objects. In addition, the system performs precise prediction of the ball's velocity, both direction and magnitude, even when sharp trajectory changes occur. Current and predicted ball and robot positions are the input to the reasoning modules, which select appropriate strategic actions for the robots. One of our main research goals for CM United is to develop collaborative team behaviors in our robots. In order to successfully achieve this goal, agents require robust individual skills. Robots can chase a ball, push the ball in a particular direction, and use the kicker to quickly propel the ball towards the target. The target can be either the goal or another teammate for a pass. The robots use the prediction of the ball's trajectory to intercept a moving ball. This allows robots to receive passes from other teammates. These skills are made more difficult by the presence of opponent robots, which are essentially moving obstacles. Obstacle avoidance is incorporated into our individual skills and successfully overcomes the often cluttered field. As the basis for collaboration, CM United 98 decomposes the team task by defining a set of roles or positions with associated behaviors. Our team consists of one goalie, one defender, and three attackers. The goalie robot has both special hardware and special behaviors. It stays parallel to and close to the goal, and moves the position itself to minimize the amount of open goal area. When the ball is shot on goal, it uses the ball's predicted trajectory to find the position where the ball will cross the goal line. The defender plays an important role in CM United 98. The goal of the defender is to prevent scoring opportunities by the opposing team. It can act as a second goal tender, annoy opponents by preventing them from getting to the ball, and clear the ball from its half of the field. The clearing behavior helped keep the ball on the opponent's side of the field and led to numerous goals. The attackers involve a challenging problem of collaboration. Although only a single robot can push the ball into the goal, a number of collaborating robots is needed in order to create such a scoring opportunity. We use a decision-theoretic analysis to select a robot to actively go to the ball to either pass or shoot. Meanwhile, the other attackers strategically position themselves to receive a pass or get a rebound. During the tournament, CM United won four of its five games by a combined score of 25-6 to become the RoboCup 98 Small Size Robot Champions.