A framework for the development of hybrid AI control systems

It is widely acknowledged that Behavior-based control schemes provide impressive performance for robots operating in the real-world. However, a large number of computational substrates and techniques have been presented for such systems, each with their own advantages. Additionally, researchers have advocated the hybridisation of such systems with traditional symbolic-based control elements. This paper presents guidelines for designing and developing hybrid control systems which may incorporate many of these diverse elements. Underlying the approach is the adoption of the Subsumption Architecture as a design metaphor to provide a framework for the unification of computationally distinct techniques into hybrid control systems. A controller framework in the form of a set of library routines is described which allows for the rapid development of such systems. This also facilitates the development of distributed systems through the ability to handle communication between separate processor units in a fully transparent manner.