Cognitive Agent Programming : a semantic approach

In this thesis we are concerned with the design and investigation of dedicated programming languages for programming agents. We focus in particular on programming languages for rational agents, i.e., flexibly behaving computing entities that are able to make "good" decisions about what to do. An important line of research in this area is based on Bratman’s so-called Belief Desire Intention (BDI) philosophy. The idea of BDI philosophy is that the behavior of rational agents can be predicted by ascribing beliefs, desires, and intentions to the agent, and by assuming that the agent will tend to act in pursuit of its desires, taking into account its beliefs about the world. The idea was then coined that it might not only be possible to explain and describe rational agents in terms of the BDI notions, but that it might also be possible to program rational agents, using these notions as first class citizens in a programming language. The research that is done along these lines not only uses the notions of beliefs, desires, and intentions, but also related notions such as goals and plans. We refer to these notions as "cognitive" notions, and to programming languages for agents based on these notions as "cognitive agent programming languages". Our work proposes new constructs for representing these cognitive notions in a programming language, and investigates existing constructs. We take a semantic approach, in that we define formal semantics for the proposed constructs, and investigate the constructs by performing a semantic analysis. We investigate in particular ways for representing goals, and we study a construct called "plan revision rule" of the cognitive agent programming language 3APL, which can be used for revising an agent’s plan if the circumstances call for this. Regarding the representation of goals, we investigate the representation of subgoals in the plans of agents. We show how declarative subgoals, i.e., subgoals representing a desired state, can be programmed in 3APL, even though the semantics of subgoals of 3APL defines them to behave in procedurally. Further, we propose a semantics for the representation of conflicting goals that is based on default logic, and investigate properties of this semantics. Also, we provide an analysis of ways in which goals have been represented in cognitive agent programming languages. Regarding plan revision rules, we analyze the semantic issues that arise with the introduction of these rules. That is, the semantics of plan execution becomes non-compositional with the introduction of these rules. This is problematic when reasoning about the execution of plans. We propose a dynamic logic that is tailored to handle plan revision by circumventing the non-compositionality issue in a certain way, and we show how these rules can be restricted such that the semantics becomes compositional again. Finally, we propose a way to introduce support for modularization in cognitive agent programming languages that is based on the goals of the agent, and we show that the Maude term rewriting language is well suited for implementing logic-based cognitive agent programming languages.

• View at Worldcat

Free Full Text ( Author Manuscript , 1mb )