At Carnegie Mellon, we have developed several small-size robot teams that have helped us to investigate a variety of aspects of the small-size RoboCup competition. The CM-Dragons’01 is our new team complete with new hardware and sensing and behavior-processing algorithms. Although still in the development phase, a number of modules have been developed that we feel can contribute to the RoboCup community. In this paper, we briefly describe our vision and tracking modules, the new robot hardware and our new communications modules. Our primary interest is on presenting our advances in modeling and prediction using an Extended Kalman-Bucy Filter (EKBF) that tracks the ten robots and the ball through vision. We identify that Kalman-Bucy filters are susceptible to white noise caused by misidentifications. Within CM-Dragons’01, we developed a new approach, Improbability Filtering, that addresses this problem in a computationally efficient yet principled manner.