We designed robust and fast robots with reliable radio communication and high speed vision. To generate actions, we implemented a reactive control architecture with interacting behaviors on different time scales. These control loops are designed in a bottom-up fashion. Lower level behaviors are configured by an increasing number of higher level behaviors that can use a longer history to determine their actions.
This framework could be used in the future to implement mechanisms, like adaptation and learning using Neural Networks [6]. We will participate in the RoboCup'99 F180 league competition to benchmark our approach. Until then, a richer set of behaviors will be available.