Motivated by the problem of automatically composing network accessible services, such as those on the World Wide Web, this paper proposes an approach to building agent technology based on the the notion of generic procedures and customizing user constraint. We argue that an augmented version of the logic programming language Golog provides a natural formalism for programming Web services. To this end, we adapt and extend the Golog language to enable programs that are generic, customizable and usable in the context of the Web. We realize these extensions in an augmented ConGolog interpreter that combines online execution of information-providing Web services with offline simulation of world-altering Web services, to determine a sequence of Web Services for subsequent execution. Our implemented system is currently interacting with services on the Web. 1

This article gives an overview of AI (Artificial Intelligence) plan-ning techniques and discusses their application to the Web service composition problem.

in this paper, we develop an account of the kind of deliberation that an agent that is doing planning or executing high-level programs under incomplete information must be able to perform.

In this paper we consider three different kinds of domain dependent control knowledge (temporal, procedural and HTN-based) that are useful in planning. Our approach is declarative and relies on the language of logic programming with answer set semantics (LPASS). We show that the addition of these three kinds of control knowledge only involves adding a few more rules to a planner written in LPASS that can plan without any control knowledge. Thus domain dependent control knowledge can be modularly added to (or removed from) a planning problem without the need of modifying the planner. We formally prove the correctness of our planner, both in the absence and presence of the control knowledge. Finally, we do some initial experimentation that shows the reduction in planning time when procedural domain knowledge is used and the plan length is big.

We investigate the relationship between two accounts of knowledge and action in the situation calculus: the Scherl and Levesque (SL) approach that models knowledge with possible worlds, and the Demolombe and Pozos Parra (DP) approach that models knowledge by a set of “knowledge fluents. ” We construct combined action theories: basic action theories that encode a correspondence between an SL and a DP theory. We prove, subject to certain restrictions, that knowledge of fluent literals are provably the same after a sequence of actions. Moreover, this knowledge equivalence extends to a rich class of formulae. These results allow us to translate certain SL theories into equivalent DP theories that avoid the computational drawbacks of possible world reasoning. They also enable us to prove the correctness of the DP treatment of knowledge and action in terms of a possible world specification. 1

An effective method is presented for deriving state knowledge in the presence of sensing actions. It is shown how conditional plans can be inferred with the help of a generalized concept of plan skeletons as search heuristics, which allow the planner to introduce conditional branching points by need.

In a recent paper, we presented a new logic called ES for reasoning about the knowledge, action, and perception of an agent. Although formulated using modal operators, we argued that the language was in fact a dialect of the situation calculus but with the situation terms suppressed. This allowed us to develop a clean and workable semantics for the language without piggybacking on the generic Tarski semantics for first-order logic. In this paper, we reconsider the relation between ES and the situation calculus and show how to map sentences of ES into the situation calculus. We argue that the fragment of the situation calculus represented by ES is rich enough to handle the basic action theories defined by Reiter as well as Golog. Finally, we show that in the full second-order version of ES, almost all of the situation calculus can be accommodated.

The Fluent Calculus is presented as a comprehensive specification and programming language for endowing robots with the ability of task planning in complex environments. Based on a solution to the classical Frame Problem in pure first-order logic, our calculus allows to solve planning problems where the robot has incomplete state knowledge and which involve the use of sensors, actions with uncertain effects, actions with ramications (i.e., indirect effects), and the concurrent execution of actions. Our new theory of sensing is distinguished by its simple inference scheme for calculating the effects of actions on state knowledge and by its comparatively simple account of non-knowledge. A realization of the Fluent Calculus by means of constraint logic programming is presented. Outstanding novel features of the system are to solve the inferential Frame Problem under incomplete state information, to allow for solving planning problems with knowledge goals, and to combine nondeterminism, concurrency, and ramification.

High-level controllers that operate robots in dynamic, uncertain domains are concerned with at least two reasoning tasks dealing with the effects of noisy sensors and effectors: They have a) to project the effects of a candidate plan and b) to update their beliefs during on-line execution of a plan. In this paper, we show how the pGOLOG framework, which in its original form only accounted for the projection of high-level plans, can be extended to reason about the way the robot's beliefs evolve during the on-line execution of a plan. pGOLOG, an extension of the high-level programming language GOLOG, allows the specification of probabilistic beliefs about the state of the world and the representation of sensors and effectors which have uncertain, probabilistic outcomes. As an application of belief update, we introduce belief-based programs, GOLOG-style programs whose tests appeal to the agent's beliefs at execution time.

The situation calculus, as proposed by McCarthy and Hayes, and developed over the last decade by Reiter and co-workers, is reconsidered. A new logical variant is proposed that captures much of the expressive power of the original, but where certain technical results are much more easily proven. This is illustrated using two existing non-trivial results: the regression theorem and the determinacy of knowledge theorem of Reiter. We also obtain a regression theorem for knowledge, and show how to reduce reasoning about knowledge and action to non-epistemic non-dynamic reasoning about the initial situation.