In this paper we present DAML-S, a DAML+OIL ontology for describing the properties and capabilities of Web Services. Web Services - Web-accessible programs and devices - are garnering a great deal of interest from industry, and standards are emerging for low-level descriptions of Web Services. DAML-S complements this effort by providing Web Service descriptions at the application layer, describing what a service can do, and not just how it does it. In this paper we describe three aspects of our ontology: the service profile, the process model, and the service grounding. The paper focuses on the grounding, which connects our ontology with low-level XML-based descriptions of Web Services.

One vision of the "Semantic Web" of the future is that software agents will interact with each other using formal metadata that reveal their interfaces. We examine one plausible paradigm, where agents provide service descriptions that tell how they can be used to accomplish other agents' goals. From the point of view of these other agents, the problem of deciphering a service description is quite similar to the standard AI planning problem, with some interesting twists. Two such twists are the possibility of having to reconcile contradictory ontologies -- or conceptual frameworks -- used by the agent, and having to rearrange the data structures of a message-sending agent so they match the expectations of the recipient. We argue that the former problem requires human intervention and maintenance, but that the latter can be fully automated.

Classical planning is the problem of finding a sequence of actions to achieve a goal given an exact characterization of a domain. An algorithm to solve this problem is presented, which searches a space of plan prefixes, trying to extend one of them to a complete sequence of actions. It is guided by a heuristic estimator based on regression-match graphs, which attempt to characterize the entire subgoal structure of the remaining part of the problem. These graphs simplify the structure by neglecting goal interactions and by assuming that variables in goal conjunctions should be bound in such a way as to make as many conjuncts as possible true without further work. In some domains, these approximations work very well, and experiments show that many classical planning problems can be solved with very little search.

The problem of place recognition is central to robot map learning. A robot needs to be able to recognize when it has returned to a previously visited place, or at least to be able to estimate the likelihood that it has been at a place before. Our approach is to compare images taken at two places, using a stochastic model of changes due to shift, zoom, and occlusion to predict the probability that one of them could be a perturbation of the other. We have performed experiments to gather the valve of a /spl chi//sup 2/ statistic applied to image matching from a variety of indoor locations. Image pairs gathered from nearby locations generate low /spl chi//sup 2/ values, and images gathered from different locations generate high values. The rate of false positive and false negative matches is low.

Planning is designing the behavior of some entity that acts, either an individual, a group, or an organization. The output is some kind of blueprint for behavior, which we call a plan. People make a lot of plans, sometimes for themselves, sometimes for other people, sometimes for machines. The question that defines the topic of this issue is: How can we automate planning? The question is interesting for the usual reasons. First, automating planning might shed light on how people and other animals design their behavior. Second, complex planning problems might be solved better with the aid of computers. There are a wide variety of planning problems, differentiated by the types of their inputs and outputs. Typically, planning problems get more and more difficult as more flexible inputs are allowed and fewer constraints on the output are required. Typically, these problems get very difficult very quickly, as some of the papers in this issue will attest. The classical approach to planning problems is to start from specifications of the effects of actions, then try to infer a string of actions that bring about a particular state of affairs. In recent years many variations on this theme have been explored. One important special case of planning is scheduling, for which the input to the behavior designer is a set of actions that must be carried out, and the output is an order in which to carry them out. Some orderings are better than others, because each of the given tasks will require a set of resources with a finite capacity. For example, in a transportation scheduling problem, we might be given a set of objects to move to various places, and a set of trucks to carry them in, and the desired output might be a schedule of movements that allows every truck to be as full as possible and idle as little as possible.

Intelligent agency requires some ability to predict the future. An agent must ask itself what is presently its best course of action given what it now knows about what the world will be like when it intends to act. This paper presents a system that uses a probabilistic model to reason about the effects of an agent's proposed actions on a dynamic and uncertain world, computing the probability that relevant propositions will hold at a specified point in time. The model allows for incomplete information about the world, the occurrence of exogenous (unplanned) events, unreliable sensors, and the possibility of an imperfect causal theory. The system provides an application program with answers to questions of the form “is the probability that ϕ will hold in the world at time t greater than τ?” It is unique among algorithms for probabilistic temporal reasoning in that it tries to limit its inference according to the proposition, time, and probability threshold provided by the application. The system will also notify the application if subsequent evidence invalidates its answer to a query.

There are two strands of research into automated planning: the study of nonlinear, progressive, heuristic planners, and the study of linear, regressive, rigorous planners. We focus on the latter, in particular the formulation of E. Pednault. Regression planning can be viewed as a nondeterministic algorithm in which a goal is reduced by preserving it from the initial situation, by insertion of a step to accomplish it, or by adaptation of an existing step. This algorithm can be proven complete, in the sense that it finds any straight-line plan with no redundant steps. An extension of the algorithm to make use of "formal objects" is also complete. A practical implementation of the algorithm can solve several nontrivial problems. A powerful predicate-calculus simplifier is an important component. It is still unclear whether these elegant results make contact with practicality.

There are several different kinds of reason-maintenance system in existence, which provide rather different functionalities. I present a general structure that subsumes most such systems, and that allows some new behaviors to emerge. The general framework is based on logic-style clauses (disjunctions of literals) instead of justifications. Literals are tagged with labels that say what assumption sets make them true and false. Nonmonotonicity is implemented by allowing clauses to contain disjuncts of the form Lp, which supports propagation through a clause whenever p is not known to be true. The resulting system supports two popular styles of dependency-directed backtracking, using nogoods and assumption retraction. Assumption retraction does not require a separate contradiction-elimination phase, but occurs automatically during label propagation. Label propagation can be achieved by the usual variants of Boolean constraint propagation, provided there are no “odd loops” through the clauses, and it can be shown that the system itself never creates odd loops.

Ever since its introduction by McCarthy and Hayes in 1969, the so-called frame problem has been the object of much fascination and debate. Although it was defined in the narrow context of the situation calculus, a specific temporal formalism, it was clear from the start that it is in fact a manifestation of some fundamental problem in temporal reasoning.

Nonmonotonic formal systems have been proposed as an extension to classical first-order logic that will capture the process of human “default reasoning” or “plausible inference” through their inference mechanisms, just as modus ponens provides a model for deductive reasoning. But although the technical properties of these logics have been studied in detail and many examples of human default reasoning have been identified, for the most part these logics have not actually been applied to practical problems to see whether they produce the expected results.

Planning routes and executing them requires both topological and metric information. A natural implementation of a ‘cognitive map’ might therefore consist of an assertional data base for topological information and a ‘fuzzy map’ for the metric information. A fuzzy map captures facts about objects by recording their relative positions, orientations, and scales in convenient frames of reference. It is fuzzy in the sense that coordinates are specified to lie in a range rather than having fixed values. The fuzzy map allows easy retrieval of information. The same information is also represented in a discrimination tree, which allows an object to be retrieved given its location and other attributes. The problem of constructing a fuzzy map is more difficult; we present a partial solution, an algorithm that assimilates a fact first by imposing constraints on the fuzzy coordinates of the objects involved, then by rearranging or growing the tree of frames of reference. Route planning is modelled as a process of finding the overall direction and topology of the path, then filling in the details by deciding how to go around barriers. It uses the retrieval algorithms. Our program SPAM carries out all these processes.

Much previous work in artificial intelligence has neglected representing time in all its complexity. In particular, it has neglected continuous change and the indeterminacy of the future. To rectify this, I have developed a first‐order temporal logic, in which it is possible to name and prove things about facts, events, plans, and world histories. In particular, the logic provides analyses of causality, continuous change in quantities, the persistence of facts (the frame problem), and the relationship between tasks and actions. It may be possible to implement a temporal‐inference machine based on this logic, which keeps track of several “maps” of a time line, one per possible history.

‘Non-monotonic’ logical systems are logics in which the introduction of new axioms can invalidate old theorems. Such logics are very important in modeling the beliefs of active processes which, acting in the presence of incomplete information, must make and subsequently revise assumptions in light of new observations. We present the motivation and history of such logics. We develop model and proof theories, a proof procedure, and applications for one non-monotonic logic. In particular, we prove the completeness of the non-monotonic predicate calculus and the decidability of the non-monotonic sentential calculus. We also discuss characteristic properties of this logic and its relationship to stronger logics, logics of incomplete information, and truth maintenance systems.