Model Theory for PRS-Like Agents: Modelling Belief Update and Action Attempts

Abstract

Abstract. In this paper, we extend our earlier work on modelling the mental states of PRS-like agents by considering the dynamics of belief and modelling of action attempts. The major constraint on our theory is that belief update is modelled within the theory of action as part of the logic ADL, a logic of belief, desire and intention that incorporates propositional dynamic logic. Some logical properties of belief update and attempts are given. The account provides a more complete modelling of both the statics and dynamics of agent programs based on the PRSlike architecture, and thus is a suitable foundation for developing model checking algorithms for this class of agents.

An Analysis of Three Puzzles in the Logic of Intention

“…In their future-directed aspect, actions of the agent are modelled as attempts with no guarantee of success; viewed from the aspect of the (ongoing) present, the world is modelled as a sequence of states that actually occur, and intentions are incorporated into such models. The approaches of Cohen and Levesque [4] and Wooldridge [21] are of the latter type, taking as a starting point models of actions inspired by computational logic; those of Rao and Georgeff [14,15] and Wobcke [17,19] are more oriented towards the future-directed aspects (so notions of the agent's possible actions and the failure of an agent's attempts are more pertinent in these approaches). However, the main difficulty is to capture both present-directed and future-directed aspects of intention in one formalism.…”

“…In earlier work, Wobcke [17,19], motivated by these problems, we developed a logic called Agent Dynamic Logic (ADL), in which possible worlds corresponded to branching-time execution structures of an agent program, and in which intentions referred to actions successfully performed by the agent. In Wobcke [18], we showed how to automatically derive various logical properties of intention.…”

“…We have developed an operational semantics aiming to succinctly abstract the main properties of this family of architectures [16], models of the "mental states" of agents based on this architecture [17], a formal derivation of some "rationality postulates" of those agents [18], extended the account to incorporate belief update and attempts [19], and examined model generation algorithms for use in model checking for agents of this class [20]. The theory is based on a logic, Agent Dynamic Logic (ADL), that combines elements from Computation Tree Logic [6], Propositional Dynamic Logic [13], and BDI Logic [14].…”

Model generation for PRS-like agents

“…In this paper, we present an approach to representing the execution structures, including the agent's beliefs and intentions, for any PRS-type agent program, Georgeff and Lansky [3], in a reachability graph, as in Spin, Holzmann [4]. The construction is based on our logical approach to Agent Dynamic Logic, an extension of PDL that includes modal operators for belief, goal and intention, Wobcke [7,8]. The notion of intention is a more "rational" version of the operational notion used in PRS-type systems, though does not capture all the complexities of Bratman's theory [2].…”

Model Checking for PRS-Like Agents

“…This project includes development of a precise operational semantics that synthesizes a number of properties common to various PRS-like architectures, Wobcke [16], and the development of formal approaches to the semantics of the mental notions employed in this type of architecture, Wobcke [17], extended to belief update and the modelling of success and failure through action attempts, Wobcke [18]. To make this paper self-contained, many of the formal details must be omitted: the reader is referred to this earlier work for such details.…”