The problem of designing control schemes for teams of robots to satisfy complex high-level tasks is a challenging problem which becomes more difficult when adding constraints on relative locations of robots. This paper presents a method for automatically creating hybrid controllers that ensure a team of heterogeneous robots satisfy some user specified high-level task while guaranteeing collision avoidance and predicting and reducing deadlock. The generated hybrid controller composes atomic controllers based on information the robots gather during runtime; thus these atomic controllers can be reused in different scenarios for multiple tasks. As a demonstration of this general approach we examine a task in which a group of robots sort different items to be recycled.