A Hierarchy of Reactive Behaviors Handles Complexity

This paper discusses the hierarchical control architecture used to generate the behavior of individual agents and a team of robots for the RoboCup Small Size competition. Our reactive approach is based on control layers organized in a temporal hierarchy. Fast and simple behaviors reside on the bottom of the hierarchy, while an increasing number of slower and more complex behaviors are implemented in the higher levels. In our architecture deliberation is not implemented explicitly, but to an external viewer it seems to be present. Each layer is composed of three modules. First, the sensor module, where the perceptual dynamics aggregates the readings of fast changing sensors in time to form complex, slow changing percepts. Next, the activation module computes the activation dynamics that determines whether or not a behavior is allowed to influence actuators, and finally the actuator module, where the active behaviors influence the actuators to match a target dynamics. We illustrate our approach by describing the bottom-up design of behaviors for the RoboCup domain.