Reinforcement Learning has emerged as a useful framework for learning to perform a task optimally from experience in unknown systems. A major problem for such learning algorithms is how to balance optimally the exploration of the system, to gather knowledge, and the exploitation of current knowledge, to complete the task. Model-based Bayesian Reinforcement Learning (BRL) methods provide an optimal solution to this problem by formulating it as a planning problem under uncertainty. However, the complexity of these methods has so far limited their applicability to small and simple domains. To improve the applicability of model-based BRL, this thesis presents several extensions to more complex and realistic systems, such as partially observable and continuous domains. To improve learning efficiency in large systems, this thesis includes another extension to automatically learn and exploit the structure of the system. Approximate algorithms are proposed to efficiently solve the resulting inference and planning problems. / L'apprentissage par renforcement a émergé comme une technique utile pour apprendre à accomplir une tâche de façon optimale à partir d'expérience dans les systèmes inconnus. L'un des problèmes majeurs de ces algorithmes d'apprentissage est comment balancer de façon optimale l'exploration du système, pour acquérir des connaissances, et l'exploitation des connaissances actuelles, pour compléter la tâche. L'apprentissage par renforcement bayésien avec modèle permet de résoudre ce problème de façon optimale en le formulant comme un problème de planification dans l'incertain. La complexité de telles méthodes a toutefois limité leur applicabilité à de petits domaines simples. Afin d'améliorer l'applicabilité de l'apprentissage par renforcement bayésian avec modèle, cette thèse presente plusieurs extensions de ces méthodes à des systèmes beaucoup plus complexes et réalistes, où le domaine est partiellement observable et/ou continu. Afin d'améliorer l'efficacité de l'apprentissage dans les gros systèmes, cette thèse inclue une autre extension qui permet d'apprendre automatiquement et d'exploiter la structure du système. Des algorithmes approximatifs sont proposés pour résoudre efficacement les problèmes d'inference et de planification résultants.

An original way to define, analyze and design mechanical systems with inherently lifelike dynamic properties is presented. The construction of robotic manipulators which embody a complete set of technologically relevant biological principles is outlined. The ultimate objective is to develop a new class of mobile, autonomous, and interactive machines which dynamically emulate live musculoskeletal systems. / This study introduces the mathematical models and algorithms to transform and synthesize the results of research in musculoskeletal physiology into explicit engineering design specifications. The application of a new contractile muscle-like viscoelastic motor, as a servomechanical drive for articulated rigid link mechanisms as well as for a novel flexible trunk-like manipulator, is investigated. Key features of the neuromuscular force control by twitch summation are combined to formulate a pulse stream control method suitable for fluid powered mechanisms.

<p> Starting from Edward Edinger's portrayal of Jung's process of individuation as the creation of consciousness, this dissertation asks in what ways the creation of artificial intelligence (AI) can be seen as the recreation of consciousness, and specifically whether the AI's maturation from nonconsciousness to something equivalent to consciousness will have an analogous effect on humanity's development out of unconsciousness toward a greater state of cognitive freedom. Taking a functional perspective, this dissertation asks whether B. F. Skinner's metaphor of the human psyche as a black box, normally seen as expressing the belief that humans are mechanistic and determined, is in fact an attempt to insulate the most intimate of human experiences (the soul) from the intrusive gaze of the scientific mindset. Juxtaposing this black box metaphor with two other metaphors&mdash;that of the box that holds Schrodinger's cat and that of Pandora's box&mdash;this dissertation asks whether the presence of an entirely constructed entity that displays all the signs of soul will cause the artificially intelligent entity to act as a mirror, reflecting humanity's gaze past our inner defenses, to an inner absence where a metaphysical soul was once surmised to be. Although such a change in self-image would initially entail an apparent loss of meaning, this dissertation notes that such a lacuna of meaning is already growing in society and concludes that the loss of this concept would eventually result in a new concept of self that would represent an important milestone for the collective individuation of the species.</p>

Planning under uncertainty is an increasingly important research field, and it is clear that the design of robust and scalable algorithms which consider uncertainty is key to the development of effective autonomous and semi-autonomous systems. Partially Observable Markov Decision Processes (POMDPs) offer a powerful mathematical framework for making optimal action choices in noisy and/or uncertain environments. However, integration of the POMDP model with real world applications has been slow due to the high computation cost of exact approaches to POMDP planning. / In recent years, point-based POMDP solvers have emerged as efficient methods for providing approximate solutions by planning over a small subset of the belief space. This thesis first provides a survey on many of the proposed point-based POMDP solvers. We then conduct an empirical analysis on the key components of point-based methods, the belief collection and belief updating processes. This is an important contribution, as previous publications on point-based methods have only compared full algorithms, without comparing the underlying processes. As well, we verify the effect of a variety of parameters and optimizations that could be used within a point-based solver. Experiments are conducted on a variety of POMDP environments. / L'importance grandissante de la recherche dans le domaine de la planification sous incertitude est signe que l'élaboration d'algorithmes robustes et extensibles qui gèrent l'incertitude est un élément clé dans le développement de systèmes autonomes et semi-autonomes efficaces. Les processus de décision markoviens partiellement observables (POMDP) constituent une puissante fondation mathématique pour le choix d'actions optimales dans un environnement incertain. Il a cependant été difficile d'incorporer les POMDPs à des applications réelles, à cause de leur coût de calcul élevé lorsqu'une solution exacte est requise. / Récemment, les approches de résolution de POMDPs dites par points, qui planifient sur un petit sous-ensemble de l'état des croyance, se sont révélées être efficaces pour obtenir des solutions approximatives. Le présent m´emoire propose tout d'abord une revue de plusieurs approches par points. Par la suite, une analyse empirique des composantes primordiales des approches par points, de la collecte d'observations, ainsi que du processus de mise à jour de l'état des croyance, est proposée. De plus, les effets de différents paramètres et optimisations liés aux approches par points sont vérifiés. Des expériences sont conduites avec une variété d'environnements de type POMDP.

Thesis (M.F.A.)--Rochester Institute of Technology, 1988. / Includes bibliographical references (p. 31-32).

Thesis (Ph.D.)--Indiana University, Dept. of Mathematics, 2007. / Source: Dissertation Abstracts International, Volume: 68-07, Section: B, page: 4531. Adviser: Lawrence Moss. Title from dissertation home page (viewed Apr. 22, 2008).

Thesis (M.S.)--Ohio University, June, 1989. / Title from PDF t.p.

Thesis (M.S.)--Ohio University, March, 1989. / Title from PDF t.p.

Thesis (M.S.)--Ohio University, June, 1997. / Title from PDF t.p.

Thèse (M.Sc.A.)--Université de Sherbrooke (Canada), 2008. / Titre de l'écran-titre (visionné le 1 février 2007). In ProQuest dissertations and theses. Publié aussi en version papier.