Towards Answer Set Prolog Based Architectures for Intelligent Agents

An important research area in the field of AI is the design of intelligent agents acting in a changing environment. Research in this area has led to development of agent architectures that support various tasks such as planning, diagnosis, learning etc. One such architecture is based on the agent repeatedly executing the observe-think-act-loop (Baral and Gelfond 2000; Kowalski and Sadri 1999). This architecture is applicable if the world including the agent and its environment is viewed as a transition diagram whose states correspond to possible physical states of the world and whose arcs are labeled by actions. A transition, 〈σ, a, σ′〉, of a diagram denotes that action a in executable in state σ and that after the execution of a the system may move to state σ′. The diagram consists of all possible trajectories of the system. One of the approaches to describing these diagrams is a theory based on action languages formal models of parts of natural language used for reasoning about actions and their effects (Gelfond and Lifschitz 1998). A theory in an action language (often called an action description) describes a transition diagram that contains all possible trajectories of a given dynamic system. Currently there are several action languages that are used to study different features of dynamic domains. For instance action language AL (Baral and Gelfond 2000; Turner 1997) is simply an extension of action language A (Gelfond and Lifschitz 1993) by state constraints which express causal relations between fluents functions whose values depend on state and may change as a result of actions. The semantics of AL formalize McCarthy’s Principle of inertia which says that “Things tend to stay the same unless they are changed by actions” (McCarthy and Hayes 1969). Similarly AC (Baral and Gelfond 1997) expands A by allowing concurrent actions in the causal laws (which are prohibited in A). Another extension of AL is the action language H (Chintabathina, Gelfond, and Watson 2005) which is capable of modeling domains containing continuous processes properties of the domain whose values change continuously with time. Semantics of this language are based on a generalization of McCain-Turner equation (McCain and Turner 1995) which allows us to define processes using state constraints. (Chintabathina, Gelfond, and Watson 2007) introduces a new action language CARD which extends both