Thomas Leo McCluskey

Planning domain definition using GIPO

In this paper an object-centric perspective on planning domain definition is presented along with an overview of GIPO (graphical interface for planning with objects), a supporting tools environment. It is argued that the object-centric view assists the domain developer in conceptualizing the domain’s structure, and we show how GIPO enables the developer to capture that conceptualization at an appropriate and matching conceptual level. GIPO is an experimental environment which provides a platform for exploring and demonstrating the range and scope of tools required to support the knowledge engineering aspects of creating and validating planning systems, both for classical pre-condition planning and hierarchical planning. GIPO embodies the object-centric view, leading to a range of benefits typically associated with object-oriented methods in other fields of software engineering such as highly visual development methods, code reuse and efficient, reliable development.

Acquiring planning domain models using LOCM

The problem of formulating knowledge bases containing action schema is a central concern in knowledge engineering for AI Planning. This paper describes LOCM, a system which carries out the automated generation of a planning domain model from example training plans. The novelty of LOCM is that it can induce action schema without being provided with any information about predicates or initial, goal or intermediate state descriptions for the example action sequences. Each plan is assumed to be a sound sequence of actions; each action in a plan is stated as a name and a list of objects that the action refers to. LOCM exploits assumptions about the kinds of domain model it has to generate, rather than handcrafted clues or planner-oriented knowledge. It assumes that actions change the state of objects, and require objects to be in a certain state before they can be executed. In this paper we describe the implemented LOCM algorithm, the assumptions that it is based on, and an evaluation using plans generated through goal directed solutions, through random walk, and through logging human generated plans for the game of Freecell. We analyse the performance of LOCM by its application to the induction of domain models from five domains.

A requirements capture method and its use in an air traffic control application

This paper describes our experience in capturing, using a formal specification language, a model of the knowledge‐intensive domain of oceanic air traffic control. This model is intended to form part of the requirements specification for a decision support system for air traffic controllers. We give an overview of the methods we used in analysing the scope of the domain, choosing an appropriate formalism, developing a domain model, and validating the model in various ways. Central to the method was the development of a formal requirements engineering environment which provided automated tools for model validation and maintenance.

On exploiting structures of classical planning problems: generalizing entanglements

Much progress has been made in the research and development of automated planning algorithms in recent years. Though incremental improvements in algorithm design are still desirable, complementary approaches such as problem reformulation are important in tackling the high computational complexity of planning. While machine learning and adaptive techniques have been usefully applied to automated planning, these advances are often tied to a particular planner or class of planners that are coded to exploit that learned knowledge. A promising research direction is in exploiting knowledge engineering techniques such as reformulating the planning domain and/or the planning problem to make the problem easier to solve for general, state-of-the-art planners. Learning (outer) entanglements is one such technique, where relations between planning operators and initial or goal atoms are learned, and used to reformulate a domain by removing unneeded operator instances. Here we generalize this approach significantly to cover relations between atoms and pairs of operators themselves, and develop a technique for producing inner entanglements. We present methods for detecting inner entanglements and for using them to do problem reformulation. We provide a theoretical treatment of the area, and an empirical evaluation of the methods using standard planning benchmarks and state-of-the-art planners.

Ontology-coupled active contours for dynamic video scene understanding

In this paper, we present an innovative approach coupling active contours with an ontological representation of knowledge, in order to understand scenes acquired by a moving camera and containing multiple non-rigid objects evolving over space and time. The developed active contours enable both segmentation and tracking of multiple targets in each captured scene over a video sequence with unknown camera calibration. Hence, this active contour technique provides information on the objects of interest as well as on parts of them (e.g. shape and position), and contains simultaneously low-level characteristics such as intensity or color features. The ontology we propose consists of concepts whose hierarchical levels map the granularity of the studied scene and of a set of inter- and intra-object spatial and temporal relations defined for this framework, object and sub-object characteristics e.g. shape, and visual concepts like color. The system obtained by coupling this ontology with active contours can study dynamic scenes at different levels of granularity, both numerically and semantically characterize each scene and its components i.e. objects of interest, and reason about spatiotemporal relations between them or parts of them. This resulting knowledge-based vision system was demonstrated on real-world video sequences containing multiple mobile highly-deformable objects.

Knowledge formulation for AI planning

In this paper we present an overview of the principle components of GIPO, an environment to support knowledge acquisition for AI Planning. GIPO assists in the knowledge formulation of planning domains, and in prototyping planning problems within these domains. GIPO features mixed-initiative components such as generic type composition, an operator induction facility, and various plan animation and validation tools. We outline the basis of the main tools, and show how an engineer might use them to formulate a domain model. Throughout the paper we illustrate the formulation process using the Hiking Domain.

Towards an Adaptive Information Retrieval System

Standard Information Retrieval Systems (IRS) can be used to retrieve information in response to specific requests, but they have no powers of adaption to particular users over repeated sessions. This paper describes a learning system which uses relevance feedback from a probabilistic IRS to incrementally evolve a context for a user, over a number of online sessions. We demonstrate the learning implementation with an example, and argue that it can help an IRS adapt to a users specific needs, by using this context to influence document display and selection.

THE APPLICATION OF MACHINE LEARNING TOOLS TO THE VALIDATION OF AN AIR TRAFFIC CONTROL DOMAIN THEORY

In this paper we describe a project (IMPRESS) in which machine learning (ML) tools were created and utilised for the validation of an Air Traffic Control domain theory written in first order logic. During the project, novel techniques were devised for the automated revision of general clause form theories using training examples. These techniques were combined in an algorithm which focused in on the parts of a theory which involve ordinal sorts, and applied geometrical revision operators to repair faulty component parts. While we illustrate the feasibility of applying ML to this area, we conclude that to be effective it must be focused to the application at hand, and used in mixed-initiative mode within a tools environment. The method is illustrated with experimental results obtained during the project.

Action Knowledge Acquisition with Opmaker2

AI planning engines require detailed specifications of dynamic knowledge of the domain in which they are to operate, before they can function. Further, they require domain-specific heuristics before they can function efficiently. The problem of formulating domain models containing dynamic knowledge regarding actions is a barrier to the widespread uptake of AI planning, because of the difficulty in acquiring and maintaining them. Here we postulate a method which inputs a partial domain model (one without knowledge of domain actions) and training solution sequences to planning tasks, and outputs the full domain model, including heuristics that can be used to make plan generation more efficient.

Knowledge Representation in Planning: A PDDL to OCLh Translation

Recent successful applications of AI planning technology have highlighted the knowledge engineering of planning domain models as an important research area. We describe an implemented translation algorithm between two languages used in planning representation: PDDL, a language used for communication of example domains between research groups, and OCLh, a language developed specifically for planning domain modelling. The algorithm is being used as part of OCLhs tool support to import models expressed in PDDL to OCLhs environment. Here we outline the translation algorithm, and discuss the issues that it uncovers. Although the tool performs reasonably well when its output is measured against hand-crafted OCLh, it results in only partially specified models. Analyis of the translation results shows that this is because many natural assumptions about domains are not captured in the PDDL encodings.