Recent advances in Artificial Intelligence (AI) have brought AI closer to laypeople than ever before. This leads to a pervasive problem: how would a user ascertain whether an AI system will be safe, reliable, or useful in a given situation? This problem becomes particularly challenging when it is considered that most autonomous systems are not designed by their users; the internal software of these systems may be unavailable or difficult to understand; and the functionality of these systems may even change from initial specifications as a result of learning. To overcome these challenges, this dissertation proposes a paradigm for third-party autonomous assessment of black-box taskable AI systems. The four main desiderata of such assessment systems are: (i) interpretability: generating a description of the AI system's functionality in a language that the target user can understand; (ii) correctness: ensuring that the description of AI system's working is accurate; (iii) generalizability creating a solution approach that works well for different types of AI systems; and (iv) minimal requirements: creating an assessment system that does not place complex requirements on AI systems to support the third-party assessment, otherwise the manufacturers of AI system's might not support such an assessment. To satisfy these properties, this dissertation presents algorithms and requirements that would enable user-aligned autonomous assessment that helps the user understand the limits of a black-box AI system's safe operability. This dissertation proposes a personalized AI assessment module that discovers the high-level ``capabilities'' of an AI system with arbitrary internal planning algorithms/policies and learns an accurate symbolic description of these capabilities in terms of concepts that a user understands. Furthermore, the dissertation includes the associated theoretical results and the empirical evaluations. The results show that (i) a primitive query-response interface can enable the development of autonomous assessment modules that can derive a causally accurate user-interpretable model of the system's capabilities efficiently, and (ii) such descriptions are easier to understand and reason with for the users than the agent's primitive actions.

In recent years, the development of Control Barrier Functions (CBF) has allowed safety guarantees to be placed on nonlinear control affine systems. While powerful as a mathematical tool, CBF implementations on systems with high relative degree constraints can become too computationally intensive for real-time control. Such deployments typically rely on the analysis of a system's symbolic equations of motion, leading to large, platform-specific control programs that do not generalize well. To address this, a more generalized framework is needed. This thesis provides a formulation for second-order CBFs for rigid open kinematic chains. An algorithm for numerically computing the safe control input of a CBF is then introduced based on this formulation. It is shown that this algorithm can be used on a broad category of systems, with specific examples shown for convoy platooning, drone obstacle avoidance, and robotic arms with large degrees of freedom. These examples show up to three-times performance improvements in computation time as well as 2-3 orders of magnitude in the reduction in program size.

The notion of the safety of a system when placed in an environment with humans and other machines has been one of the primary concerns of practitioners while deploying any cyber-physical system (CPS). Such systems, also called safety-critical systems, need to be exhaustively tested for erroneous behavior. This generates the need for coming up with algorithms that can help ascertain the behavior and safety of the system by generating tests for the system where they are likely to falsify. In this work, three algorithms have been presented that aim at finding falsifying behaviors in cyber-physical Systems. PART-X intelligently partitions while sampling the input space to provide probabilistic point and region estimates of falsification. PYSOAR-C and LS-EMIBO aims at finding falsifying behaviors in gray-box systems when some information about the system is available. Specifically, PYSOAR-C aims to find falsification while maximizing coverage using a two-phase optimization process, while LS-EMIBO aims at exploiting the structure of a requirement to find falsifications with lower computational cost compared to the state-of-the-art. This work also shows the efficacy of the algorithms on a wide range of complex cyber-physical systems. The algorithms presented in this thesis are available as python toolboxes.

Automated driving systems (ADS) have come a long way since their inception. It is clear that these systems rely heavily on stochastic deep learning techniques for perception, planning, and prediction, as it is impossible to construct every possible driving scenario to generate driving policies. Moreover, these systems need to be trained and validated extensively on typical and abnormal driving situations before they can be trusted with human life. However, most publicly available driving datasets only consist of typical driving behaviors. On the other hand, there is a plethora of videos available on the internet that capture abnormal driving scenarios, but they are unusable for ADS training or testing as they lack important information such as camera calibration parameters, and annotated vehicle trajectories. This thesis proposes a new toolbox, DeepCrashTest-V2, that is capable of reconstructing high-quality simulations from monocular dashcam videos found on the internet. The toolbox not only estimates the crucial parameters such as camera calibration, ego-motion, and surrounding road user trajectories but also creates a virtual world in Car Learning to Act (CARLA) using data from OpenStreetMaps to simulate the estimated trajectories. The toolbox is open-source and is made available in the form of a python package on GitHub at https://github.com/C-Aniruddh/deepcrashtest_v2.

A storage system requiring file redundancy and on-line repairability can be represented as a Steiner system, a combinatorial design with the property that every $t$-subset of its points occurs in exactly one of its blocks. Under this representation, files are the points and storage units are the blocks of the Steiner system, or vice-versa. Often, the popularities of the files of such storage systems run the gamut, with some files receiving hardly any attention, and others receiving most of it. For such systems, minimizing the difference in the collective popularity between any two storage units is nontrivial; this is the access balancing problem. With regard to the representative Steiner system, the access balancing problem in its simplest form amounts to constructing either a point or block labelling: an assignment of a set of integer labels (popularity ranks) to the Steiner system's point set or block set, respectively, requiring of the former assignment that the sums of the labelled points of any two blocks differ as little as possible and of the latter that the sums of the labels assigned to the containing blocks of any two distinct points differ as little as possible. The central aim of this dissertation is to supply point and block labellings for Steiner systems of block size greater than three, for which up to this point no attempt has been made. Four major results are given in this connection. First, motivated by the close connection between the size of the independent sets of a Steiner system and the quality of its labellings, a Steiner triple system of any admissible order is constructed with a pair of disjoint independent sets of maximum cardinality. Second, the spectrum of resolvable Bose triple systems is determined in order to label some Steiner 2-designs with block size four. Third, several kinds of independent sets are used to point-label Steiner 2-designs with block size four. Finally, optimal and close to optimal block labellings are given for an infinite class of 1-rotational resolvable Steiner 2-designs with arbitrarily large block size by exploiting their underlying group-theoretic properties.

Many real-world engineering problems require simulations to evaluate the design objectives and constraints. Often, due to the complexity of the system model, simulations can be prohibitive in terms of computation time. One approach to overcome this issue is to construct a surrogate model, which approximates the original model. The focus of this work is on the data-driven surrogate models, in which empirical approximations of the output are performed given the input parameters. Recently neural networks (NN) have re-emerged as a popular method for constructing data-driven surrogate models. Although, NNs have achieved excellent accuracy and are widely used, they pose their own challenges. This work addresses two common challenges, the need for: (1) hardware acceleration and (2) uncertainty quantification (UQ) in the presence of input variability. The high demand in the inference phase of deep NNs in cloud servers/edge devices calls for the design of low power custom hardware accelerators. The first part of this work describes the design of an energy-efficient long short-term memory (LSTM) accelerator. The overarching goal is to aggressively reduce the power consumption and area of the LSTM components using approximate computing, and then use architectural level techniques to boost the performance. The proposed design is synthesized and placed and routed as an application-specific integrated circuit (ASIC). The results demonstrate that this accelerator is 1.2X and 3.6X more energy-efficient and area-efficient than the baseline LSTM. In the second part of this work, a robust framework is developed based on an alternate data-driven surrogate model referred to as polynomial chaos expansion (PCE) for addressing UQ. In contrast to many existing approaches, no assumptions are made on the elements of the function space and UQ is a function of the expansion coefficients. Moreover, the sensitivity of the output with respect to any subset of the input variables can be computed analytically by post-processing the PCE coefficients. This provides a systematic and incremental method to pruning or changing the order of the model. This framework is evaluated on several real-world applications from different domains and is extended for classification tasks as well.

Imitation learning is a promising methodology for teaching robots how to physically interact and collaborate with human partners. However, successful interaction requires complex coordination in time and space, i.e., knowing what to do as well as when to do it. This dissertation introduces Bayesian Interaction Primitives, a probabilistic imitation learning framework which establishes a conceptual and theoretical relationship between human-robot interaction (HRI) and simultaneous localization and mapping. In particular, it is established that HRI can be viewed through the lens of recursive filtering in time and space. In turn, this relationship allows one to leverage techniques from an existing, mature field and develop a powerful new formulation which enables multimodal spatiotemporal inference in collaborative settings involving two or more agents. Through the development of exact and approximate variations of this method, it is shown in this work that it is possible to learn complex real-world interactions in a wide variety of settings, including tasks such as handshaking, cooperative manipulation, catching, hugging, and more.

Autonomous Vehicles (AV) are inevitable entities in future mobility systems thatdemand safety and adaptability as two critical factors in replacing/assisting human drivers. Safety arises in defining, standardizing, quantifying, and monitoring requirements for all autonomous components. Adaptability, on the other hand, involves efficient handling of uncertainty and inconsistencies in models and data. First, I address safety by presenting a search-based test-case generation framework that can be used in training and testing deep-learning components of AV. Next, to address adaptability, I propose a framework based on multi-valued linear temporal logic syntax and semantics that allows autonomous agents to perform model-checking on systems with uncertainties. The search-based test-case generation framework provides safety assurance guarantees through formalizing and monitoring Responsibility Sensitive Safety (RSS) rules. I use the RSS rules in signal temporal logic as qualification specifications for monitoring and screening the quality of generated test-drive scenarios. Furthermore, to extend the existing temporal-based formal languages’ expressivity, I propose a new spatio-temporal perception logic that enables formalizing qualification specifications for perception systems. All-in-one, my test-generation framework can be used for reasoning about the quality of perception, prediction, and decision-making components in AV. Finally, my efforts resulted in publicly available software. One is an offline monitoring algorithm based on the proposed logic to reason about the quality of perception systems. The other is an optimal planner (model checker) that accepts mission specifications and model descriptions in the form of multi-valued logic and multi-valued sets, respectively. My monitoring framework is distributed with the publicly available S-TaLiRo and Sim-ATAV tools.

Complex systems appear when interaction among system components creates emergent behavior that is difficult to be predicted from component properties. The growth of Internet of Things (IoT) and embedded technology has increased complexity across several sectors (e.g., automotive, aerospace, agriculture, city infrastructures, home technologies, healthcare) where the paradigm of cyber-physical systems (CPSs) has become a standard. While CPS enables unprecedented capabilities, it raises new challenges in system design, certification, control, and verification. When optimizing system performance computationally expensive simulation tools are often required, and search algorithms that sequentially interrogate a simulator to learn promising solutions are in great demand. This class of algorithms are black-box optimization techniques. However, the generality that makes black-box optimization desirable also causes computational efficiency difficulties when applied real problems. This thesis focuses on Bayesian optimization, a prominent black-box optimization family, and proposes new principles, translated in implementable algorithms, to scale Bayesian optimization to highly expensive, large scale problems. Four problem contexts are studied and approaches are proposed for practically applying Bayesian optimization concepts, namely: (1) increasing sample efficiency of a highly expensive simulator in the presence of other sources of information, where multi-fidelity optimization is used to leverage complementary information sources; (2) accelerating global optimization in the presence of local searches by avoiding over-exploitation with adaptive restart behavior; (3) scaling optimization to high dimensional input spaces by integrating Game theoretic mechanisms with traditional techniques; (4) accelerating optimization by embedding function structure when the reward function is a minimum of several functions. In the first context this thesis produces two multi-fidelity algorithms, a sample driven and model driven approach, and is implemented to optimize a serial production line; in the second context the Stochastic Optimization with Adaptive Restart (SOAR) framework is produced and analyzed with multiple applications to CPS falsification problems; in the third context the Bayesian optimization with sample fictitious play (BOFiP) algorithm is developed with an implementation in high-dimensional neural network training; in the last problem context the minimum surrogate optimization (MSO) framework is produced and combined with both Bayesian optimization and the SOAR framework with applications in simultaneous falsification of multiple CPS requirements.

Autonomous Vehicles (AVs) have the potential to significantly evolve transportation. AVs are expected to make transportation safer by avoiding accidents that happen due to human errors. When AVs become connected, they can exchange information with the infrastructure or other Connected Autonomous Vehicles (CAVs) to efficiently plan their future motion and therefore, increase the road throughput and reduce energy consumption. Cooperative algorithms for CAVs will not be deployed in real life unless they are proved to be safe, robust, and resilient to different failure models. Since intersections are crucial areas where most accidents happen, this dissertation first focuses on making existing intersection management algorithms safe and resilient against network and computation time, bounded model mismatches and external disturbances, and the existence of a rogue vehicle. Then, a generic algorithm for conflict resolution and cooperation of CAVs is proposed that ensures the safety of vehicles even when other vehicles suddenly change their plan. The proposed approach can also detect deadlock situations among CAVs and resolve them through a negotiation process. A testbed consisting of 1/10th scale model CAVs is built to evaluate the proposed algorithms. In addition, a simulator is developed to perform tests at a large scale. Results from the conducted experiments indicate the robustness and resilience of proposed approaches.