We consider the problem of computing optimal generalised policies for relational Markov decision processes. We describe an approach combining some of the benefits of purely inductive techniques with those of symbolic dynamic programming methods. The latter reason about the optimal value function using first-order decisiontheoretic regression and formula rewriting, while the former, when provided with a suitable hypotheses language, are capable of generalising value functions or policies for small instances. Our idea is to use reasoning and in particular classical first-order regression to automatically generate a hypotheses language dedicated to the domain at hand, which is then used as input by an inductive solver. This approach avoids the more complex reasoning of symbolic dynamic programming while focusing the inductive solver’s attention on concepts that are specifically relevant to the optimal value function for the domain considered. 1

Abstract not found

We study data-driven Web services provided by Web sites interacting with users or applications. The Web site can access an underlying database, as well as state information updated as the interaction progresses, and receives user input. The structure and contents of Web pages, as well as the actions to be taken, are determined dynamically by querying the underlying database as well as the state and inputs. The properties to be verified concern the sequences of events (inputs, states, and actions) resulting from the interaction, and are expressed in linear or branchingtime temporal logics. The results establish under what conditions automatic verification of such properties is possible and provide the complexity of verification. This brings into play a mix of techniques from logic and automatic verification.

In this paper we address the problem of planning with complex actions. We are motivated by the problem of automated Web service composition, in which planning must be performed using predefined complex actions or services as the building blocks of a plan. Planning with complex actions is also compelling in primitive action planning domains because it enables the exploitation of reusable subplans, potentially improving the efficiency of planning. This paper provides a formal, semantically-justified account of how to plan with complex actions using operator-based planning techniques. A key contribution of this work is the definition, characterization, and computation of preconditions and conditional effects for complex actions. While we use the situation calculus and Golog to formalize the task and our solution, the results in this paper are directly applicable to most action theories and planning systems.

We examine the approach of encoding planning problems as CSPs more closely. First we present a simple CSP encoding for planning problems and then a set of transformations that can be used to eliminate variables and add new constraints to the encoding. We show that our transformations uncover additional structure in the planning problem, structure that subsumes the structure uncovered by GRAPHPLAN planning graphs. We solve the CSP encoded planning problem by using standard CSP algorithms. Empirical evidence is presented to validate the effectiveness of this approach to solving planning problems, and to show that even a prototype implementation is more effective than standard GRAPHPLAN. Our prototype is even competitive with far more optimized planning graph based implementations. We also demonstrate that this approach can be more easily lifted to more complex types of planning than can planning graphs. In particular, we show that the approach can be easily extended to planning with resources.

We introduce a new approximate solution technique for first-order Markov decision processes (FOMDPs). Representing the value function linearly w.r.t. a set of first-order basis functions, we compute suitable weights by casting the corresponding optimization as a first-order linear program and show how off-the-shelf theorem prover and LP software can be effectively used. This technique allows one to solve FOMDPs independent of a specific domain instantiation; furthermore, it allows one to determine bounds on approximation error that apply equally to all domain instantiations. We apply this solution technique to the task of elevator scheduling with a rich feature space and multi-criteria additive reward, and demonstrate that it outperforms a number of intuitive, heuristicallyguided policies. 1

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.

This paper contributes an analysis and formalisation of Damasio’s theory on core consciousness. Three important concepts in this theory are ‘emotion’, ‘feeling’, and ‘feeling a feeling ’ (or core consciousness). In particular, a simulation model is described of the dynamics of basic mechanisms leading via emotion and feeling to core consciousness, and dynamic properties are formally specified that hold for these dynamics at a more global level. These properties have been automatically checked for the simulation model. Moreover, a formal analysis is made of relevant notions of representation used by Damasio. As part of this analysis, specifications of representation relations have been verified and confirmed against the simulation model. 1

We study the notion of update of an ontology expressed as a Description Logic knowledge base. Such a knowledge base is constituted by two components, called TBox and ABox. The former expresses general knowledge about the concepts and their relationships, whereas the latter describes the state of affairs regarding the instances of concepts. We investigate the case where the update affects only the instance level of the ontology, i.e., the ABox. Building on classical approaches on knowledge base update, our first contribution is to provide a general semantics for instance level update in Description Logics. We then focus on DL-Lite, a specific Description Logic where the basic reasoning tasks are computationally tractable. We show that DL-Lite is closed with respect to instance level update, in the sense that the result of an update is always expressible as a new DL-Lite ABox. Finally we provide an algorithm that computes the result of an update in DL-Lite, and we show that it runs in polynomial time with respect to the size of both the original knowledge base and the update formula.

In (Petrick & Bacchus 2002), a “knowledge-level ” approach to planning under incomplete knowledge and sensing was presented. In comparision with alternate approaches based on representing sets of possible worlds, this higher level representation is richer, but the inferences it supports are weaker. Nevertheless, because of its richer representation, it is able to solve problems that cannot be solved by alternate approaches. In this paper we examine a collection of new techniques for increasing both the representational and inferential power of the knowledge-level approach. These techniques have been fully implemented in the PKS (Planning with Knowledge and Sensing) planning system. Taken together they allow us to solve a range of new types of planning problems under incomplete knowledge and sensing.