Gat integrates some of the reactive aspects of the subsumption architecture of Brooks with some of the deliberative aspects of other higher-level models in order to produce a robot capable of collision-free navigation through dynamic environments in pursuit of explicitly represented high-level goals.
The architecture consists of three layers each managing progressively more time consuming and time-committing levels of activity. The subsumption-like control layer, the lowest level of the system, is responsible for the reactivity of the agent. The sequencing layer is the "operating system" layer of the agent which manages conflicts introduced by the control and deliberative layers using sets of semaphores. The deliberative layer is an asynchronously connected layer responsible for computing time-consuming tasks such as planning and maintaining world models.