Supporting deliberation within behaviour-based systems

The S* architecture [Tsotsos, 1995; 1996; 1997] was proposed as a solution to the problem of integrating deliberative functionality within a behaviour-based framework. This thesis examines S*, with some simplifying assumptions, and considers some theoretical and practical issues related to control using S*.

S* overcomes fundamental flaws in the concepts that belay behaviour-based robotics [Kirsh, 1991; Tsotsos, 1992], particularly the concept that explicit representation should be avoided. S* modifies the notion of the “world” to include explicit representations, and modifies the notion of the “behaviour” to be an SMPA process that reads and writes world representations. Consequently, S* does not resort to hybrid methodology to support deliberation, but rather does so in a uniform and conceptually convenient manner.

Using a formalism of S*, this thesis demonstrates that S* is Turing Equivalent, showing a reduction from a Two-Counter Turing Machine, using a finite set of “universal” behaviours and representations. As well, an object-oriented software framework is developed for implementing S* controllers (with simplifying assumptions), and two controllers are demonstrated.