Skip to main content

Research Repository

Advanced Search

Chris Harper's Qualifications (1)

Engineering
Doctor of Philosopy

Status Complete
Part Time Yes
Years 1994 - 2004
Project Title A Rational Methodology for Designing Behaviour Based Systems for Safety Related Applications
Project Description This thesis proposes a design methodology for intelligent autonomous systems in safety critical applications, which are required to undergo a certification process to gain assurance of their safety properties. The thesis uses the behaviour-based model of artificial intelligence. Most of the contemporary approaches to development of behaviour-based systems rely on inductive techniques, but inductive inference has two problems that make it unsuitable for safety related applications. First, the empirical nature of the inductive development process means that the only plausible assurance method for these systems is statistical validation by testing. However the degree of validation required for most safety critical applications is so high that such an approach is impractical. Second, inductive inference is not logically conclusive, which limits the ability of designers to make rational statements about the properties of the system being developed, which compromises the quality of the certification exercise.
This thesis takes a different approach by proposing a rational design theory for Behaviour-based AI. The design methodology draws on the basic theory as the basis for constructing systems. The thesis argues that this is the only way for an objective safety certification to be achieved. The thesis has developed an underlying theory of design for Colony-style Subsumption Architecture, a well-known type of behaviour based system architecture. The thesis extends Lyapunov stability theory to a second order, observing that there is a subsumption property of the second order theorem that models subsumption architecture. The thesis proposes a design methodology based on this idea, which contains a set of procedures for construction of systems in accordance with the basic theory. The thesis describes the application of the methodology in some simulation experiments, including an Inverted Pendulum and an Aircraft Flight Control problem.