Although an automated reader for the blind first appeared nearly two-hundred years ago, computers can currently "read" document text about as well as a seven-year-old. Scene text recognition brings many new challenges. A central limitation of current approaches is a feed-forward, bottom-up, pipelined architecture that isolates the many tasks and information involved in reading. The result is a system that commits errors from which it cannot recover and has components that lack access to relevant information. We propose a system for scene text reading that in its design, training, and operation is more integrated. First, we present a simple contextual model for text detection that is ignorant of any recognition. Through the use of special features and data context, this model performs well on the detection task, but limitations remain due to the lack of interpretation. We then introduce a recognition model that integrates several information sources, including font consistency and a lexicon, and compare it to approaches using pipelined architectures with similar information. Next we examine a more unified detection and recognition framework where features are selected based on the joint task of detection and recognition, rather than each task individually. This approach yields better results with fewer features. Finally, we demonstrate a model that incorporates segmentation and recognition at both the character and word levels. Text with difficult layouts and low resolution are more accurately recognized by this integrated approach. By more tightly coupling several aspects of detection and recognition, we hope to establish a new unified way of approaching the problem that will lead to improved performance. We would like computers to become accomplished grammar-school level readers.

The broad problem I address in this dissertation is design of autonomous agents that can efficiently learn how to achieve desired behaviors in large, complex environments. I focus on one essential design component: the ability to form new behavioral units, or skills, from existing ones. I propose a characterization of a useful class of skills in terms of general properties of an agent's interaction with its environment—in contrast to specific properties of a particular environment—and I introduce methods that can be used to identify and acquire such skills autonomously.

The decentralized Markov decision process (Dec-POMDP) is a powerful formal model for studying multiagent problems where cooperative, coordinated action is optimal, but each agent acts based on local data alone. Unfortunately, it is known that Dec-POMDPs are fundamentally intractable: they are NEXP-complete in the worst case, and have been empirically observed to be beyond feasible optimal solution. To get around these obstacles, researchers have focused on special classes of the general Dec-POMDP problem, restricting the degree to which agent actions can interact with one another. In some cases, it has been proven that these sorts of structured forms of interaction can in fact reduce worst-case complexity. Where formal proofs have been lacking, empirical observations suggest that this may also be true for other cases, although less is known precisely. This thesis unifies a range of this existing work, extending analysis to establish novel complexity results for some popular restricted-interaction models. We also establish some new results concerning cases for which reduced complexity has been proven, showing correspondences between basic structural features and the potential for dimensionality reduction when employing mathematical programming techniques. As our new complexity results establish that worst-case intractability is more widespread than previously known, we look to new ways of analyzing the potential average-case difficulty of Dec-POMDP instances. As this would be extremely difficult using the tools of traditional complexity theory, we take a more empirical approach. In so doing, we identify new analytical measures that apply to all Dec-POMDPs, whatever their structure. These measures allow us to identify problems that are potentially easier to solve on average, and validate this claim empirically. As we show, the performance of well-known optimal dynamic programming methods correlates with our new measure of difficulty. Finally, we explore the approximate case, showing that our measure works well as a predictor of difficulty there, too, and provides a means of setting algorithm parameters to achieve far more efficient performance.

The resource-constrained scheduling problem (RCSP) involves the assignment of a limited set of resources to a collection of tasks, with the intent of satisfying some particular qualitative objective, under a variety of technological and temporal constraints. Real-world environments, however, introduce a variety of complications to the standard RCSP. The dynamic resource-constrained scheduling problem describes a class of real-world RCSPs that exist within the context of dynamic and unpredictable environments, where the details of the problem are often incomplete, and subject to change over time, without notice. Previous approaches to solving resource-constrained scheduling problems failed to focus on the dynamic nature of real-world environments. The scheduling process occurs away from the environment in which the resulting schedule is executed. Complete prior knowledge of the order set is assumed, and reaction to changes in the environment, if at all, is limited. We have developed a generic, multi-faceted, knowledge-based approach to solving dynamic resource-constrained scheduling problems, which focuses on issues of flexibility during the solution process to enable effective reaction to dynamic environments. Our approach is characterized by a highly opportunistic control scheme that provides the ability to adapt quickly to changes in the environment, a least-commitment scheduling procedure that preserves maneuverability by explicitly incorporating slack time into the developing schedule, and the systematic consultation of a range of relevant scheduling perspectives at key decision-making points that provides an informed view of the current state of problem-solving at all times. The Dynamic Scheduling System (DSS) is a working implementation of our scheduling approach, capable of representing a wide range of dynamic RCSPs, and producing quality schedules under a variety of real-world conditions. It handles a number of additional domain complexities, such as inter-order tasks and mobile resources with significant travel requirements. We discuss our scheduling approach and its application to two different RCSP domains, and evaluate its effectiveness in each, using special application systems built with DSS.

Autonomous agents may not have access to complete information about the state of the environment. For example, a robot soccer player may only be able to estimate the locations of other players not in the scope of its sensors. However, even though all the information needed for ideal decision making cannot be sensed, all that is sensed is usually not needed. The noise and motion of spectators, for example, can be ignored in order to focus on the game field. Standard formulations do not consider this situation, assuming that all the can be sensed must be included in any useful abstraction. This dissertation extends the Markov Decision Process Homomorphism framework (Ravindran, 2004) to partially observable domains, focusing specically on reducing Partially Observable Markov Decision Processes (POMDPs) when the model is known. This involves ignoring aspects of the observation function which are irrelevant to a particular task. Abstraction is particularly important in partially observable domains, as it enables the formation of a smaller domain model and thus more efficient use of the observed features.

Stereo Image Super-Resolution (stereoSR) has attracted significant attention in recent years due to the extensive deployment of dual cameras in mobile phones, autonomous vehicles and robots. In this work, we propose a new StereoSR method, named SwinFSR, based on an extension of SwinIR, originally designed for single image restoration, and the frequency domain knowledge obtained by the Fast Fourier Convolution (FFC). Specifically, to effectively gather global information, we modify the Residual Swin Transformer blocks (RSTBs) in SwinIR by explicitly incorporating the frequency domain knowledge using the FFC and employing the resulting residual Swin Fourier Transformer blocks (RSFTBlocks) for feature extraction. Besides, for the efficient and accurate fusion of stereo views, we propose a new cross-attention module referred to as RCAM, which achieves highly competitive performance while requiring less computational cost than the state-of-the-art cross-attention modules. Extensive experimental results and ablation studies demonstrate the effectiveness and efficiency of our proposed SwinFSR. iv / Thesis / Master of Applied Science (MASc)

State-of-the-art domestic robot assistants are essentially autonomous mobile manipulators capable of exerting human-scale precision grasps. To maximize utility and economy, non-technical end-users would need to be nearly as efficient as trained roboticists in control and collaboration of manipulation task behaviors. However, it remains a significant challenge given that many WIMP-style tools require superficial proficiency in robotics, 3D graphics, and computer science for rapid task modeling and recovery. But research on robot-centric collaboration has garnered momentum in recent years; robots are now planning in partially observable environments that maintain geometries and semantic maps, presenting opportunities for non-experts to cooperatively control task behavior with autonomous-planning agents exploiting the knowledge. However, as autonomous systems are not immune to errors under perceptual difficulty, a human-in-the-loop is needed to bias autonomous-planning towards recovery conditions that resume the task and avoid similar errors. In this work, we explore interactive techniques allowing non-technical users to model task behaviors and perceive cooperatively with a service robot under robot-centric collaboration. We evaluate stylus and touch modalities that users can intuitively and effectively convey natural abstractions of high-level tasks, semantic revisions, and geometries about the world. Experiments are conducted with 'pick-and-place' tasks in an ideal 'Blocks World' environment using a Kinova JACO six degree-of-freedom manipulator. Possibilities for the architecture and interface are demonstrated with the following features; (1) Semantic 'Object' and 'Location' grounding that describe function and ambiguous geometries (2) Task specification with an unordered list of goal predicates, and (3) Guiding task recovery with implied scene geometries and trajectory via symmetry cues and configuration space abstraction. Empirical results from four user studies show our interface was much preferred than the control condition, demonstrating high learnability and ease-of-use that enable our non-technical participants to model complex tasks, provide effective recovery assistance, and teleoperative control.

In anticipation of the complex, dynamic battlefields of the future, military operations are increasingly demanding robots with increased autonomous capabilities to support soldiers. Effective communication is necessary to establish a common ground on which human-robot teamwork can be established across the continuum of military operations. However, the types and format of communication for mixed-initiative collaboration is still not fully understood. This study explores two approaches to communication in human-robot interaction, transparency and communication pattern, and examines how manipulating these elements with a robot teammate affects its human counterpart in a collaborative exercise. Participants were coupled with a computer-simulated robot to perform a cordon-and-search-like task. A human-robot interface provided different transparency types - about the robot's decision making process alone, or about the robot's decision making process and its prediction of the human teammate's decision making process - and different communication patterns - either conveying information to the participant or both conveying information to and soliciting information from the participant. This experiment revealed that participants found robots that both conveyed and solicited information to be more animate, likeable, and intelligent than their less interactive counterparts, but working with those robots led to more misses in a target classification task. Furthermore, the act of responding to the robot led to a reduction in the number of correct identifications made, but only when the robot was solely providing information about its own decision making process. Findings from this effort inform the design of next-generation visual displays supporting human-robot teaming.

This dissertation is about building learning control architectures for agents embedded in finite, stationary, and Markovian environments. Such architectures give embedded agents the ability to improve autonomously the efficiency with which they can achieve goals. Machine learning researchers have developed reinforcement learning (RL) algorithms based on dynamic programming (DP) that use the agent's experience in its environment to improve its decision policy incrementally. This is achieved by adapting all evaluation function in such a way that the decision policy that is "greedy" with respect to it improves with experience. This dissertation focuses on finite, stationary and Markovian environments for two reasons: it allows the development and use of a strong theory of RL, and there are many challenging real-world RL tasks that fall into this category. This dissertation establishes a novel connection between stochastic approximation theory and RL that provides a uniform framework for understanding all the different RL algorithms that have been proposed to date. It also highlights a dimension that clearly separates all RL research from prior work on DP. Two other theoretical results showing how approximations affect performance in RL provide partial justification for the use of compact function approximators in RL. In addition, a new family of "soft" DP algorithms is presented. These algorithms converge to solutions that are more robust than the solutions found by classical DP algorithms. Despite all of the theoretical progress, conventional RL architectures scale poorly enough to make them impractical for many real-world problems. This dissertation studies two aspects of the scaling issue: the need to accelerate RL, and the need to build RL architectures that can learn to solve multiple tasks. It presents three RL architectures, CQ-L, H-DYNA, and BB-RL, that accelerate learning by facilitating transfer of training from simple to complex tasks. Each architecture uses a different method to achieve transfer of training; CQ-L uses the evaluation functions for simple tasks as building blocks to construct the evaluation function for complex tasks, H-DYNA uses the evaluation functions for simple tasks to build an abstract environment model, and BB-RL uses the decision policies found for the simple tasks as the primitive actions for the complex tasks. A mixture of theoretical and empirical results are presented to support the new RL architectures developed in this dissertation.

No description available.