Large-scale spatial data such as air quality of a city, biomass content in a lake, Wi-Fi Received Signal Strengths (RSS, also referred as fingerprints) in indoor spaces often play vital roles to applications like indoor localization. However, it is extremely labor-intensive and time-consuming to collect those data manually. In this thesis, the main goal is to develop efficient means for large-scale spatial data collection. Robotic technologies nowadays offer an opportunity on mobile sensing, where data are collected by a robot traveling in target areas. However, since robots usually have a limited travel budget depending on battery capacity, one important problem is to schedule a data collection path to best utilize the budget. Inspired by existing literature, we consider to collect data along informative paths. The process to search the most informative path given a limited budget is known as the informative path planning (IPP) problem, which is NP-hard. Thus, we propose two heuristic approaches, namely a greedy algorithm and a genetic algorithm. Experiments on Wi-Fi RSS based localization show that data collected along informative paths tend to achieve lower errors than that are opportunistically collected. In practice, the budget of a mobile robot can vary due to insufficient charging or battery degradation. Although it is possible to apply the same path planning algorithm repetitively whenever the budget changes, it is more efficient and desirable to avoid solving the problem from scratch. This can be possible since informative paths for the same area share common characteristics. Based on this intuition, we propose and design a reinforcement learning based IPP solution, which is able to predict informative paths given any budget. In addition, it is common to have multiple robots to conduct sensing tasks cooperatively. Therefore, we also investigate the multi-robot IPP problem and present two solutions based on multi-agent reinforcement learning. Mobile crowdsourcing (MCS) offers another opportunity to lowering the cost of data collection. In MCS, data are collected by individual contributors, which is able to accumulate a large amount of data when there are sufficient participants. As an example, we consider the collection of a specific type of spatial data, namely Wi-Fi RSS, for indoor localization purpose. The process to collect RSS is also known as site survey in the localization community. Though MCS based site survey has been suggested a decade ago~\cite{park2010growing}, so far, there has not been any published large-scale fingerprint MCS campaign. The main issue is that it depends on user's participation, and users may be reluctant to make a contribution. To investigate user behavior in a real-world site survey, we design an indoor fingerprint MCS system and organize a data collection campaign in the McMaster University campus for five months. Although we focus on Wi-Fi fingerprints, the design choices and campaign experience are beneficial to the MCS of other types of spatial data as well. The contribution of this thesis is two-fold. For applications where robots are available for large-scale spatial sensing, efficient path planning solutions are investigated so as to maximize data utility. Meanwhile, for MCS based data acquisition, our real-world campaign experience and user behavior study reveal essential design factors that need to be considered and aspects for further improvements. / Thesis / Doctor of Philosophy (PhD) / A variety of applications such as environmental monitoring require to collect large-scale spatial data like air quality, temperature and humidity. However, it usually incurs dramatic costs like time to obtain those data, which is impeding the deployment of those applications. To reduce the data collection efforts, we consider two mobile sensing schemes, i.e, mobile robotic sensing and mobile crowdsourcing. For the former scheme, we investigate how to plan paths for mobile robots given limited travel budgets. For the latter scheme, we design a crowdsourcing platform and study user behavior through a real word data collection campaign. The proposed solutions in this thesis can benefit large-scale spatial data collection tasks.

Reinforcement learning (RL) has been at the core of recent advances in fulfilling the AI promise towards general intelligence. Unlike other machine learning (ML) paradigms, such as supervised learning (SL) that learn to mimic how humans act, RL tries to mimic how humans learn, and in many tasks, managed to discover new strategies and achieved super-human performance. This is possible mainly because RL algorithms are allowed to interact with the world to collect the data they need for training by themselves. This is not possible in SL, where the ML model is limited to a dataset collected by humans which can be biased towards sub-optimal solutions. The downside of RL is its high cost when trained on real systems. This high cost stems from the fact that the actions taken by an RL model during the initial phase of training are merely random. To overcome this issue, it is common to train RL models using simulators before deploying them in production. However, designing a realistic simulator that faithfully resembles the real environment is not easy at all. Furthermore, simulator-based approaches don’t utilize the sheer amount of field-data available at their disposal. This work investigates new ways to bridge the gap between SL and RL through an offline pre-training phase. The idea is to utilize the field-data to pre-train RL models in an offline setting (similar to SL), and then allow them to safely explore and improve their performance beyond human-level. The proposed training pipeline includes: (i) a process to convert static datasets into RL-environment, (ii) an MDP-aware data augmentation process of offline-dataset, and (iii) a pre-training step that improves RL exploration phase. We show how to apply this approach to design an action recommendation engine (ARE) that automates network operation centers (NOC); a task that is still tackled by teams of network professionals using hand-crafted rules. Our RL algorithm learns to maximize the Quality of Experience (QoE) of NOC users and minimize the operational costs (OPEX) compared to traditional algorithms. Furthermore, our algorithm is scalable, and can be used to control large-scale networks of arbitrary size.

Deep reinforcement learning (RL) represents a data-driven framework for sequential decision making that has demonstrated the ability to solve challenging control tasks. This data-driven, learning-based approach offers the potential to improve operations in complex systems, but only if it can be trusted to produce reliable performance both during training and upon deployment. These requirements have hindered the adoption of deep RL in many real-world applications. In order to overcome the limitations of existing methods, this dissertation introduces reliable deep RL algorithms that deliver (i) stable training from limited data and (ii) robust, safe deployment in the presence of uncertainty. The first part of the dissertation addresses the interactive nature of deep RL, where learning requires data collection from the environment. This interactive process can be expensive, time-consuming, and dangerous in many real-world settings, which motivates the need for reliable and efficient learning. We develop deep RL algorithms that guarantee stable performance throughout training, while also directly considering data efficiency in their design. These algorithms are supported by novel policy improvement lower bounds that account for finite-sample estimation error and sample reuse. The second part of the dissertation focuses on the uncertainty present in real-world applications, which can impact the performance and safety of learned control policies. In order to reliably deploy deep RL in the presence of uncertainty, we introduce frameworks that incorporate safety constraints and provide robustness to general disturbances in the environment. Importantly, these frameworks make limited assumptions on the training process, and can be implemented in settings that require real-world interaction for training. This motivates deep RL algorithms that deliver robust, safe performance at deployment time, while only using standard data collection from a single training environment. Overall, this dissertation contributes new techniques to overcome key limitations of deep RL for real-world decision making and control. Experiments across a variety of continuous control tasks demonstrate the effectiveness of our algorithms.

No description available.

"This thesis focuses on two key problems in reinforcement learning: How to design reward functions to obtain intended behaviors in autonomous systems using the learning-based control? Given complex mission specification, how to shape the reward function to achieve fast convergence and reduce sample complexity while learning the optimal policy? To answer these questions, the first part of this thesis investigates inverse reinforcement learning (IRL) method with a purpose of learning a reward function from expert demonstrations. However, existing algorithms often assume that the expert demonstrations are generated by the same reward function. Such an assumption may be invalid as one may need to aggregate data from multiple experts to obtain a sufficient set of demonstrations. In the first and the major part of the thesis, we develop a novel method, called Non-parametric Behavior Clustering IRL. This algorithm allows one to simultaneously cluster behaviors while learning their reward functions from demonstrations that are generated from more than one expert/behavior. Our approach is built upon the expectation-maximization formulation and non-parametric clustering in the IRL setting. We apply the algorithm to learn, from driving demonstrations, multiple driver behaviors (e.g., aggressive vs. evasive driving behaviors). In the second task, we study whether reinforcement learning can be used to generate complex behaviors specified in formal logic — Linear Temporal Logic (LTL). Such LTL tasks may specify temporally extended goals, safety, surveillance, and reactive behaviors in a dynamic environment. We introduce reward shaping under LTL constraints to improve the rate of convergence in learning the optimal and probably correct policies. Our approach exploits the relation between reward shaping and actor-critic methods for speeding up the convergence and, as a consequence, reducing training samples. We integrate compositional reasoning in formal methods with actor-critic reinforcement learning algorithms to initialize a heuristic value function for reward shaping. This initialization can direct the agent towards efficient planning subject to more complex behavior specifications in LTL. The investigation takes the initial step to integrate machine learning with formal methods and contributes to building highly autonomous and self-adaptive robots under complex missions."

The need to cope with the rapid transformation of the conventional electrical grid into the future smart grid, with multiple connected microgrids, has led to the investigation of optimal smart grid architectures. The main components of the future smart grids such as generators, substations, controllers, smart meters and collector nodes are evolving; however, truly effective integration of these elements into the microgrid context to guarantee intelligent and dynamic functionality across the whole smart grid remains an open issue. Energy trading is a significant part of this integration. In microgrids, energy trading refers to the use of surplus energy in one microgrid to satisfy the demand of another microgrid or a group of microgrids that form a microgrid community. Different techniques are employed to manage the energy trading process such as optimization-based and conventional game-theoretical methods, which bring about several challenges including complexity, scalability and ability to learn dynamic environments. A common challenge among all of these methods is adapting to changing circumstances. Optimization methods, for example, show promising performance in static scenarios where the optimal solution is achieved for a specific snapshot of the system. However, to use such a technique in a dynamic environment, finding the optimal solutions for all the time slots is needed, which imposes a significant complexity. Challenges such as this can be best addressed using game theory techniques empowered with machine learning methods across grid infrastructure and microgrid communities. In this thesis, novel Bayesian coalitional game theory-based and Bayesian reinforcement learning-based coalition formation algorithms are proposed, which allow the microgrids to exchange energy with their coalition members while minimizing the associated cost and power loss. In addition, a deep reinforcement learning scheme is developed to address the problem of large convergence time resulting from the sizeable state-action space of the methods mentioned above. The proposed algorithms can ideally overcome the uncertainty in the system. The advantages of the proposed methods are highlighted by comparing them with the conventional coalitional game theory-based techniques, Q-learning-based technique, random coalition formation, as well as with the case with no coalitions. The results show the superiority of the proposed methods in terms of power loss and cost minimization in dynamic environments.

With recent research advances, the dream of bringing domestic robots into our everyday lives has become more plausible than ever. Domestic robotics has grown dramatically in the past decade, with applications ranging from house cleaning to food service to health care. To date, the majority of the planning and control machinery for these systems are carefully designed by human engineers. A large portion of this effort goes into selecting the appropriate models and control techniques for each application, and these skills take years to master. Relieving the burden on human experts is therefore a central challenge for bringing robot technology to the masses. This work addresses this challenge by introducing a physics engine as a model space for an autonomous robot, and defining procedures for enabling robots to decide when and how to learn these models. We also present an appropriate space of motor controllers for these models, and introduce ways to intelligently select when to use each controller based on the estimated model parameters. We integrate these components into a framework called Physics-Based Reinforcement Learning, which features a stochastic physics engine as the core model structure. Together these methods enable a robot to adapt to unfamiliar environments without human intervention. The central focus of this thesis is on fast online model learning for objects with under-specified dynamics. We develop our approach across a diverse range of domestic tasks, starting with a simple table-top manipulation task, followed by a mobile manipulation task involving a single utility cart, and finally an open-ended navigation task with multiple obstacles impeding robot progress. We also present simulation results illustrating the efficiency of our method compared to existing approaches in the learning literature.

Thesis (Ph.D.) -- University of Texas at Arlington, 2008.

No description available.

Thesis (Ph.D.)--Brigham Young University. Dept. of Computer Science, 2005. / Includes bibliographical references (p. 163-168).