abstract: Current work in planning assumes that user preferences and/or domain dynamics are completely specified in advance, and aims to search for a single solution plan to satisfy these. In many real world scenarios, however, providing a complete specification of user preferences and domain dynamics becomes a time-consuming and error-prone task. More often than not, a user may provide no knowledge or at best partial knowledge of her preferences with respect to a desired plan. Similarly, a domain writer may only be able to determine certain parts, not all, of the model of some actions in a domain. Such modeling issues requires new concepts on what a solution should be, and novel techniques in solving the problem. When user preferences are incomplete, rather than presenting a single plan, the planner must instead provide a set of plans containing one or more plans that are similar to the one that the user prefers. This research first proposes the usage of different measures to capture the quality of such plan sets. These are domain-independent distance measures based on plan elements if no knowledge of the user preferences is given, or the Integrated Preference Function measure in case incomplete knowledge of such preferences is provided. It then investigates various heuristic approaches to generate plan sets in accordance with these measures, and presents empirical results demonstrating the promise of the methods. The second part of this research addresses planning problems with incomplete domain models, specifically those annotated with possible preconditions and effects of actions. It formalizes the notion of plan robustness capturing the probability of success for plans during execution. A method of assessing plan robustness based on the weighted model counting approach is proposed. Two approaches for synthesizing robust plans are introduced. The first one compiles the robust plan synthesis problems to the conformant probabilistic planning problems. The second approximates the robustness measure with lower and upper bounds, incorporating them into a stochastic local search for estimating distance heuristic to a goal state. The resulting planner outperforms a state-of-the-art planner that can handle incomplete domain models in both plan quality and planning time. / Dissertation/Thesis / Ph.D. Computer Science 2014

No description available.

Conversational agents can benefit healthcare across different application domains. However, the automated generation of reliable agents is still challenging and lags behind traditional conversational domains. This research exploited the interplay of information management and automated planning to efficiently model the expected behavior of goal-oriented health dialogues. The proposed approach supports the dynamic generation of predictable policies that are used for the management of the health dialogue as well as the identification of the dialogue state. This work advances the state of the art in health dialogue management by automating the generation (and update) of efficient dialogue managers with a reduced cost since they do not require handcrafting of the dialogue policy or large conversational datasets.

This thesis considers computer-assisted troubleshooting of heavy vehicles such as trucks and buses. In this setting, the person that is troubleshooting a vehicle problem is assisted by a computer that is capable of listing possible faults that can explain the problem and gives recommendations of which actions to take in order to solve the problem such that the expected cost of restoring the vehicle is low. To achieve this, such a system must be capable of solving two problems: the diagnosis problem of finding which the possible faults are and the decision problem of deciding which action should be taken. The diagnosis problem has been approached using Bayesian network models. Frameworks have been developed for the case when the vehicle is in the workshop only and for remote diagnosis when the vehicle is monitored during longer periods of time. The decision problem has been solved by creating planners that select actions such that the expected cost of repairing the vehicle is minimized. New methods, algorithms, and models have been developed for improving the performance of the planner. The theory developed has been evaluated on models of an auxiliary braking system, a fuel injection system, and an engine temperature control and monitoring system.

Multiagent systems offer a design paradigm used to conceptualise and implement systems composed of autonomous agents. Autonomy facilitates proactive independent behaviour yet in practice agents are constrained in order to ensure the system satisfies a desired social objective. Explicit constraints on agent behaviour, in the form of social norms, encourage this desirable system behaviour, yet research has largely focused on norm representation languages and protocols for norm proposal and adoption. The fundamental problem of how to automate the process of norm synthesis has largely been overlooked with norms assumed provided by the designer. Previous work has shown that automating the design of social norms is intractable in the worst case. Existing approaches, relying on state space enumerations, are effective for small systems but impractical for larger ones. Furthermore, they do not produce a set of succinct, general norms but rather a large number of state-specific restrictions. This work presents conflict-rooted synthesis, an automated norm synthesis approach that utilises a planning-based action schemata to overcome these limitations. These action schemata facilitate localised searches around specifications of undesirable states, using representations of sets of system states to avoid a full state enumeration. The proposed technique produces concise, generalised social norms that are applicable in multiple system states while also providing guarantees that agents are still able to achieve their original goals in the constrained system. To improve efficiency a set of theoretically sound, domain-independent optimisations are presented that reduce the state space searched without compromising the quality of the norms synthesised. A comparison with an alternative model checking based technique illustrates the advantages and disadvantages of our approach, while an empirical evaluation highlights the improved efficiency and quality of norms it produces at the cost of a less expressive specification of undesirable states. We empirically investigate the effectiveness of each of the proposed optimisations using a set of benchmark domains, quantifying how successful each of them is at reducing search complexity in practice. The results show that, with all optimisations enabled, conflict-rooted synthesis produces more generally applicable and succinct norms and consumes fewer system resources. Additionally, we show that this approach synthesises norms in systems where the competing approach is intractable. We provide a discussion of our approach, highlighting the impact our abstract search approach has on the fields of multiagent systems and automated planning, and discuss the limitations and assumptions we have made. We conclude with a presentation of future work.

Le résumé en français n'a pas été communiqué par l'auteur. / Le résumé en anglais n'a pas été communiqué par l'auteur.

Esta tese discute o papel da Análise Forense de Imagens como reguladora de mídia digital na sociedade. Isto inclui um estudo com mais de 400 indivíduos para determinar suas capacidades de detectar edições em imagens. Os resultados desse experimento indicam que humanos são facilmente enganados por imagens digitais, tendo dificuldades em diferenciar entre imagens pristinas e editadas. A tese então analisa a efetividade do arsenal de análise forense de imagens contra o estado-da-arte de composição de imagens. Através da análise de padrões fundamentais de imagens, as técnicas forenses são capazes de detectar a presença da maioria das operações de composição testadas. A tese então apresenta uma abordagem alternativa para análise forense de imagens, baseada na geração automática de planos. Ao tratar o processo de inspeção de uma imagem como um plano composto de múltiplos passos, propusemos uma arquitetura que é capaz de indicar os passos necessários para analisar uma imagem. Os planos são baseados em uma modelagem formal do conhecimento e técnicas forenses, de modo que possam ser traduzidos em passos a serem executados. A tese então demonstra que os limites de tal abordagem dependem da dificuldade de validar tal solução. Isso é uma consequência da natureza dos problemas de análise forense de imagens: essencialmente, são problemas de confiança distribuída entre indivíduos com acesso limitado à informação. Essa configuração é analisada de diferentes perspectivas em busca dos limites práticos para a análise forense de imagens digitais. Os resultados dessa análise sugerem que a área falha em produzir soluções acessíveis para a sociedade não por limitações técnicas, mas pela falta de um engajamento multi-disciplinar. A tese então discute como paradoxos filosóficos surgem naturalmente em cenários de análise forense de imagens. A análise forense de imagens digitais lida, essencialmente, com comunicação humana e, como tal, está sujeita a todas suas complexidades. Finalmente, é argumentado que o caminho para construir soluções úteis para a sociedade requer um esforço coletivo de diferentes disciplinas do conhecimento. É responsabilidade da comunidade forense desenvolver uma teoria epistemológica comum e acessível para este projeto coletivo. / This thesis discusses the role of Digital Image Forensics as a regulator of digital media in society. This includes a perceptual study with over 400 subjects to assess their ability to notice editing in images. The results of such experiment indicate that humans are easily fooled by digital images, not being able to tell apart edited and pristine images. The thesis then analyzes the effectiveness of the available arsenal of digital image forensics technology to detect image editing performed by state-of-the-art image-compositing techniques. By analyzing fundamental image patterns, forensics techniques can effectively detect the occurrence of most types of image compositing operations. In response to these two studies, the thesis presents an alternative approach to digital image forensics, based on automated plan generation. By treating the image inspection process as a plan comprised of different steps, it proposes an architecture that is able to guide an analyst choosing the next best step for inspecting an image. The generated plans are flexible, adapting on the fly to the observed results. The plans are based on a formal modelling of current forensics knowledge and techniques, so that they can be translated in steps to be executed. The thesis then shows that the limits of such an approach lie in the difficulty to validate results, which is a consequence of the setup of forensics problems: they are problems of distributed trust among parties with limited information. This scenario is analyzed from different perspectives in search for the practical limits of Digital Image Forensics as a whole. The results of such an analysis suggest that the field is lacking in providing practical and accessible solutions to society due to limited engagement in multidisciplinary research rather than due to limited technical proficiency. The thesis then discusses how paradoxes from philosophy, mathematics, and epistemology arise naturally in both real forensics scenarios, and in the theoretical foundations of the field. Digital Image Forensics ultimately deals with human communication and, as such, it is subject to all its complexities. Finally, it is argued that the path for providing useful solutions for society requires a collective engagement from different disciplines. It is the responsibility of the forensics community to develop a common, accessible epistemological framework for this collective enterprise.

abstract: A critical challenge in the design of AI systems that operate with humans in the loop is to be able to model the intentions and capabilities of the humans, as well as their beliefs and expectations of the AI system itself. This allows the AI system to be "human- aware" -- i.e. the human task model enables it to envisage desired roles of the human in joint action, while the human mental model allows it to anticipate how its own actions are perceived from the point of view of the human. In my research, I explore how these concepts of human-awareness manifest themselves in the scope of planning or sequential decision making with humans in the loop. To this end, I will show (1) how the AI agent can leverage the human task model to generate symbiotic behavior; and (2) how the introduction of the human mental model in the deliberative process of the AI agent allows it to generate explanations for a plan or resort to explicable plans when explanations are not desired. The latter is in addition to traditional notions of human-aware planning which typically use the human task model alone and thus enables a new suite of capabilities of a human-aware AI agent. Finally, I will explore how the AI agent can leverage emerging mixed-reality interfaces to realize effective channels of communication with the human in the loop. / Dissertation/Thesis / Doctoral Dissertation Computer Science 2018

abstract: Allocating tasks for a day's or week's schedule is known to be a challenging and difficult problem. The problem intensifies by many folds in multi-agent settings. A planner or group of planners who decide such kind of task association schedule must have a comprehensive perspective on (1) the entire array of tasks to be scheduled (2) idea on constraints like importance cum order of tasks and (3) the individual abilities of the operators. One example of such kind of scheduling is the crew scheduling done for astronauts who will spend time at International Space Station (ISS). The schedule for the crew of ISS is decided before the mission starts. Human planners take part in the decision-making process to determine the timing of activities for multiple days for multiple crew members at ISS. Given the unpredictability of individual assignments and limitations identified with the various operators, deciding upon a satisfactory timetable is a challenging task. The objective of the current work is to develop an automated decision assistant that would assist human planners in coming up with an acceptable task schedule for the crew. At the same time, the decision assistant will also ensure that human planners are always in the driver's seat throughout this process of decision-making. The decision assistant will make use of automated planning technology to assist human planners. The guidelines of Naturalistic Decision Making (NDM) and the Human-In-The -Loop decision making were followed to make sure that the human is always in the driver's seat. The use cases considered are standard situations which come up during decision-making in crew-scheduling. The effectiveness of automated decision assistance was evaluated by setting it up for domain experts on a comparable domain of scheduling courses for master students. The results of the user study evaluating the effectiveness of automated decision support were subsequently published. / Dissertation/Thesis / Masters Thesis Computer Science 2019

As our modern-built structures are becoming increasingly complex, carrying out basic tasks such as identifying points or objects of interest in our surroundings can consume considerable time and cognitive resources. In this thesis, we present a computational approach to converting contextual information about a person's physical environment into natural language, with the aim of helping this person identify given task-related entities in their environment. Using efficient methods from automated planning - the field of artificial intelligence concerned with finding courses of action that can achieve a goal -, we generate discourse that interactively guides a hearer through completing their task. Our approach addresses the challenges of controlling, adapting to, and monitoring the situated context. To this end, we develop a natural language generation system that plans how to manipulate the non-linguistic context of a scene in order to make it more favorable for references to task-related objects. This strategy distributes a hearer's cognitive load of interpreting a reference over multiple utterances rather than one long referring expression. Further, to optimize the system's linguistic choices in a given context, we learn how to distinguish speaker behavior according to its helpfulness to hearers in a certain situation, and we model the behavior of human speakers that has been proven helpful. The resulting system combines symbolic with statistical reasoning, and tackles the problem of making non-trivial referential choices in rich context. Finally, we complement our approach with a mechanism for preventing potential misunderstandings after a reference has been generated. Employing remote eye-tracking technology, we monitor the hearer's gaze and find that it provides a reliable index of online referential understanding, even in dynamically changing scenes. We thus present a system that exploits hearer gaze to generate rapid feedback on a per-utterance basis, further enhancing its effectiveness. Though we evaluate our approach in virtual environments, the efficiency of our planning-based model suggests that this work could be a step towards effective conversational human-computer interaction situated in the real world. / Die zunehmende Komplexität moderner Gebäude und Infrastrukturen führt dazu, dass alltägliche Aktivitäten, wie z.B. die Identifizierung von gesuchten Objekten in unserer Umgebung und das Auffinden von Orten, beträchtliche Zeit und kognitive Ressourcen in Anspruch nehmen können. In dieser Dissertation werden computerbasierte Verfahren präsentiert, welche eine Person dabei unterstützen, Zielobjekte in Ihrem Umfeld zu identifizieren. Dabei werden Informationen über die Situation und das physische Umfeld der Person - der sog. situierte Kontext - in natürliche Sprache umgewandelt. So wird Diskurs generiert, der einen Hörer interaktiv zum Erreichen eines Zieles bzw. zum Abschließen einer Aufgabe führt. Hierbei kommen Methoden aus der Planung zum Einsatz, einem Gebiet der künstlichen Intelligenz, welches sich mit der Berechnung von zielgerichteten Handlungsabfolgen beschäftigt. Die in dieser Arbeit vorgestellten Verfahren widmen sich den Herausforderungen der Kontrolle des situierten Kontexts, der Anpassung an den situierten Kontext sowie der Überwachung des situierten Kontexts. Zu diesem Zweck wird zunächst ein Sprachgenerierungssystem entwickelt, das plant, wie der nicht-linguistische Kontext einer Szene manipuliert werden kann, damit die Referenz auf relevante Objekte erleichtert wird. Dadurch ist es möglich, die kognitive Beanspruchung eines Hörers bei der Interpretation einer Referenz über mehrere sprachliche Äußerungen zu verteilen. Damit die linguistischen Entscheidungen des Systems in einem vorgegebenen Kontext optimiert werden können, wird weiterhin gelernt, die Äußerungen von Sprechern danach zu differenzieren, wie hilfreich sie in bestimmten Situationen für die Hörer waren. Dabei wird das Verhalten von menschlichen Sprechern, welches sich als hilfreich erwiesen hat, modelliert. Das daraus entstehende System kombiniert symbolisches und statistisches Schließen und stellt somit einen Lösungsansatz für das Problem dar, wie nicht-triviale referentielle Entscheidungen in reichem Kontext getroffen werden können. Zum Schluss wird ein komplementärer Mechanismus vorgestellt, der potentielle Missverständnisse bzgl. generierter Referenzen verhindern kann. Zu diesem Zweck kommt Blickerfassungstechnologie zum Einsatz. Auf Basis der Überwachung und Auswertung des Blicks des Hörers können Rückschlüsse über die Interpretation gegebener Referenzen gemacht werden; dieser Mechanismus funktioniert auch in sich dynamisch verändernden Szenen zuverlässig. Somit wird ein System präsentiert, welches den Blick des Hörers nutzt, um rasch Feedback zu generieren. Dieses Vorgehen verbessert die Effektivität des Diskurses zusätzlich. Die vorgestellten Verfahren werden in virtuellen Umwelten evaluiert. Die Effizienz des planungsbasierten Modells ist allerdings ein Indiz dafür, dass die in dieser Arbeit gemachten Vorschläge dazu dienen können, effektive Mensch-Computer-Interaktion auf Basis von Sprache auch in der realen Welt umzusetzen.