Intelligent transportation systems involve a variety of information and control systems methodologies, from cooperative systems which aim at traffic flow optimization by means of vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communication, to information fusion from multiple traffic sensing modalities. This thesis aims to address three problems in intelligent transportation systems, one in optimal control of connected automated vehicles, one in discrete-event and hybrid traffic simulation model, and one in sensing and classifying roadway obstacles in smart cities. The first set of problems addressed relates to optimally controlling connected automated vehicles (CAVs) crossing an urban intersection without any explicit traffic signaling. A decentralized optimal control framework is established whereby, under proper coordination among CAVs, each CAV can jointly minimize its energy consumption and travel time subject to hard safety constraints. A closed-form analytical solution is derived while taking speed, control, and safety constraints into consideration. The analytical solution of each such problem, when it exists, yields the optimal CAV acceleration/deceleration. The framework is capable of accommodating for turns and ensures the absence of collisions. In the meantime, a measurement of passenger comfort is taken into account while the vehicles make turns. In addition to the first-in-first-out (FIFO) ordering structure, the concept of dynamic resequencing is introduced which aims at further increasing the traffic throughput. This thesis also studies the impact of CAVs and shows the benefit that can be achieved by incorporating CAVs to conventional traffic. To validate the effectiveness of the proposed solution, a discrete-event and hybrid simulation framework based on SimEvents is proposed, which facilitates safety and performance evaluation of an intelligent transportation system. The traffic simulation model enables traffic study at the microscopic level, including new control algorithms for CAVs under different traffic scenarios, the event-driven aspects of transportation systems, and the effects of communication delays. The framework spans multiple toolboxes including MATLAB, Simulink, and SimEvents. In another direction, an unsupervised anomaly detection system is developed based on data collected through the Street Bump smartphone application. The system, which is built based on signal processing techniques and the concept of information entropy, is capable of generating a prioritized list of roadway obstacles, such that the higher-ranked entries are most likely to be actionable bumps (e.g., potholes) requiring immediate attention, while those lower-ranked are most likely to be nonactionable bumps(e.g., flat castings, cobblestone streets, speed bumps) for which no immediate action is needed. This system enables the City to efficiently prioritize repairs. Results on an actual data set provided by the City of Boston illustrate the feasibility and effectiveness of the system in practice.

<div>An Autonomous vehicle depends on the combination of latest technology or the ADAS safety features such as Adaptive cruise control (ACC), Autonomous Emergency Braking (AEB), Automatic Parking, Blind Spot Monitor, Forward Collision Warning or Avoidance (FCW or FCA), Lane Departure Warning. The current trend follows incorporation of these technologies using the Artificial neural network or Deep neural network, as an imitation of the traditionally used algorithms. Recent research in the field of deep learning and development of competent processors for autonomous or self driving car have shown amplitude of prospect, but there are many complexities for hardware deployment because of limited resources such as memory, computational power, and energy. Deployment of several mentioned ADAS safety feature using multiple sensors and individual processors, increases the integration complexity and also results in the distribution of the system, which is very pivotal for autonomous vehicles.</div><div><br></div><div>This thesis attempts to tackle two important adas safety feature: Forward collision Warning, and Object Detection using the machine learning and Deep Neural Networks and there deployment in the autonomous embedded platform.</div><div><br></div><div><div>This thesis proposes the following: </div><div>1. A machine learning based approach for the forward collision warning system in an autonomous vehicle.<br></div><div>2.3-D object detection using Lidar and Camera which is primarily based on Lidar Point Clouds. </div><div><br></div><div>The proposed forward collision warning model is based on the forward facing automotive radar providing the sensed input values such as acceleration, velocity and separation distance to a classifier algorithm which on the basis of supervised learning model, alerts the driver of possible collision. Decision Tress, Linear Regression, Support Vector Machine, Stochastic Gradient Descent, and a Fully Connected Neural Network is used for the prediction purpose.</div><div><br></div><div>The second proposed methods uses object detection architecture, which combines the 2D object detectors and a contemporary 3D deep learning techniques. For this approach, the 2D object detectors is used first, which proposes a 2D bounding box on the images or video frames. Additionally a 3D object detection technique is used where the point clouds are instance segmented and based on raw point clouds density a 3D bounding box is predicted across the previously segmented objects.</div></div>

<div>Self driving cars becoming more popular nowadays, which transport with it's own intelligence and take appropriate actions at adequate time. Safety is the key factor in driving environment. A simple fail of action can cause many fatalities. Computer Vision has major part in achieving this, it help the autonomous vehicle to perceive the surroundings. Detection is a very popular technique in helping to capture the surrounding for an autonomous car. At the same time tracking also has important role in this by providing dynamic of detected objects. Autonomous cars combine a variety of sensors such as RADAR, LiDAR, sonar, GPS, odometry and inertial measurement units to perceive their surroundings. Driver-assistive technologies like Adaptive Cruise Control, Forward Collision Warning system (FCW) and Collision Mitigation by Breaking (CMbB) ensure safety while driving.</div><div>Perceiving the information from environment include setting up sensors on the car. These sensors will collect the data it sees and this will be further processed for taking actions. The sensor system can be a single sensor or multiple sensor. Different sensors have different strengths and weaknesses which makes the combination of them important for technologies like Autonomous Driving. Each sensor will have a limit of accuracy on it's readings, so multi sensor system can help to overcome this defects. This thesis is an attempt to develop a multi sensor multi object tracking method to perceive the surrounding of the ego vehicle. When the Object detection gives information about the presence of objects in a frame, Object Tracking goes beyond simple observation to more useful action of monitoring objects. The experimental results conducted on KITTI dataset indicate that our proposed state estimation system for Multi Object Tracking works well in various challenging environments.</div>

Autonomous vehicles is a rapidly expanding field, and promise to play an important role in society. In more isolated environments, vehicle automation can bring significant efficiency and production benefits and it eliminates repetitive jobs that can lead to inattention and accidents. The thesis addresses the problem of lateral and longitudinal dynamics control of autonomous ground vehicles with the purpose of accurate and smooth path following. Clothoids are used in the design of optimal predictive controllers aimed at minimizing the lateral forces and jerks in the vehicle. First, a clothoid-based path sparsification algorithm is proposed to efficiently describe the reference path. This approach relies on a sparseness regularization technique such that a minimal number of clothoids is used to describe the reference path. Second, a clothoid-based model predictive controller (MPCC) is proposed. This controller aims at producing a smooth driving by taking advantage of the clothoid properties.  Third, we formulate the problem as an economic model predictive controller (EMPC). In EMPC the objective function contains an economic cost (here represented by comfort or smoothness), which is described by the second and first derivatives of the curvature.  Fourth, the generation of feasible speed profiles, and the longitudinal vehicle control for following these, is studied. The speed profile generation is formulated as an optimization problem with two contradictory objectives: to drive as fast as possible while accelerating as little as possible. The longitudinal controller is formulated in a similar way, but in a receding horizon fashion. The experimental evaluation with the EMPC demonstrates its good performance, since the deviation from the path never exceeds 30 cm and in average is 6 cm. In simulation, the EMPC and the MPCC are compared with a pure-pursuit controller (PPC) and a standard MPC. The EMPC clearly outperforms the PPC in terms of path accuracy and the standard MPC in terms of driving smoothness. / <p>QC 20160503</p> / iQMatic

Autonomous vehicles perform semantic segmentation to orient themselves, but training neural networks for semantic segmentation requires large amounts of labeled data. A hand-labeled real-life dataset requires considerable effort to create, so we instead turn to virtual simulators where the segmented labels are known to generate large datasets virtually for free. This work investigates how effective synthetic datasets are in driving scenarios by collecting a dataset from a simulator and testing it against a real-life hand-labeled dataset. We show that we can get a model up and running faster by mixing synthetic and real-life data than traditional dataset collection methods and achieve close to baseline performance.

<div>This thesis presents work and simulations containing the use of Artificial Intelligence for Unmanned Aerial Vehicles in search and rescue and/or surveillance operations. The goal is to create a vision system that leverages Artificial Intelligence, mainly Deep Learning techniques to build a pipeline that enables fast and accurate classification of the environment of the robot. Deep Neural Networks are trained and tested on ’emergency situational data. Further, the power of this vision system is leveraged to extend the problem onto a multiagent system to handle fault tolerance. The multi-agent system is also made resilient to Byzantine malicious attacks to help improve the reliability of the system.</div><div><br></div><div>This thesis also shows the use of Artificial Intelligence for effective surveillance for defense related purposes. Tracking the GPS coordinates of a boat using only the video of the boat captured by a camera and the GPS coordinates of the camera itself is demonstrated. The solution was tested by the Department of Defense - Department of the Navy, Naval Information Warfare Center Pacific.</div>

<p>Due to the rapid development of technologies for unmanned aircraft systems (UAS's), the supply and demand market for UAS's is expanding globally. With the great number of UAS's ready to fly in civilian airspace, an UAS aircraft traffic management system that can guarantee the safe, resilient and efficient operation of UAS's is absent. The vast majority of existing literature on UAS traffic lacks of the attention to the fundamental characteristics of UAS operation, which leads to models and methods that are difficult to implement or lacks scalability. Motivated by these challenges, this research aims at achieving three objectives: 1) the proper frameworks that scale well with high-frequency, high-density UAS operations, 2) the models that captures the fundamental characteristics of UAS operations, 3) the methods that can be implemented in practice with guarantees of efficiency, safety, and resilience. In particular, the objectives are studied at low-level UAS traffic congestion control, agent-level UAS configuration control and unknown agent prediction. The proposed frameworks and obtained results offer comprehensive and practical guidelines of real world UAS operations at different levels.</p>

Automobility refers to the continued, self-perpetuating dominance of privately-owned, gasoline-powered vehicles used primarily by single occupants—a system which clearly has broad environmental and societal impacts. Despite increasing societal interest in transitions to more sustainable transportation technologies, there has been little consideration of how such innovations might challenge, maintain or support different aspects of automobility, and what that means for technology deployment, transport policy, and user practices. To bring attention to the complexity and apparent durability of the automobility system, in this paper we develop a conceptual framework that explores automobility through a categorization of frames, or shared cultural meanings. This framework moves beyond the typical focus on private, functional considerations of user choice, financial costs and time use to also consider symbolic and societal frames of automobility that exist among users, non-users, industry, policymakers and other relevant social groups. We illustrate this framework with eight particular frames of automobility that fall into four broad categories: private-functional frames such as (1) cocooning and fortressing and (2) mobile digital offices; private-symbolic frames such as (3) gender identity and (4) social status; societal-functional frames such as (5) environmental stewardship and (6) suburbanization; and societal-symbolic frames such as (7) self-sufficiency and (8) innovativeness. Finally, we start the process of discussing several transportation innovations in light of these automobility frames, namely electrified, autonomous and shared mobility—examining early evidence for which frames would be challenged or supported by such transitions. We believe that appreciation of the complex and varied frames of automobility can enrich discussion of transitions and policy relating to sustainable transportation.

abstract: Autonomous vehicle technology has been evolving for years since the Automated Highway System Project. However, this technology has been under increased scrutiny ever since an autonomous vehicle killed Elaine Herzberg, who was crossing the street in Tempe, Arizona in March 2018. Recent tests of autonomous vehicles on public roads have faced opposition from nearby residents. Before these vehicles are widely deployed, it is imperative that the general public trusts them. For this, the vehicles must be able to identify objects in their surroundings and demonstrate the ability to follow traffic rules while making decisions with human-like moral integrity when confronted with an ethical dilemma, such as an unavoidable crash that will injure either a pedestrian or the passenger. Testing autonomous vehicles in real-world scenarios would pose a threat to people and property alike. A safe alternative is to simulate these scenarios and test to ensure that the resulting programs can work in real-world scenarios. Moreover, in order to detect a moral dilemma situation quickly, the vehicle should be able to identify objects in real-time while driving. Toward this end, this thesis investigates the use of cross-platform training for neural networks that perform visual identification of common objects in driving scenarios. Here, the object detection algorithm Faster R-CNN is used. The hypothesis is that it is possible to train a neural network model to detect objects from two different domains, simulated or physical, using transfer learning. As a proof of concept, an object detection model is trained on image datasets extracted from CARLA, a virtual driving environment, via transfer learning. After bringing the total loss factor to 0.4, the model is evaluated with an IoU metric. It is determined that the model has a precision of 100% and 75% for vehicles and traffic lights respectively. The recall is found to be 84.62% and 75% for the same. It is also shown that this model can detect the same classes of objects from other virtual environments and real-world images. Further modifications to the algorithm that may be required to improve performance are discussed as future work. / Dissertation/Thesis / Masters Thesis Mechanical Engineering 2019

Accurate modeling of acoustic propagation in the ocean waveguide is important to SONAR-performance prediction, and requires, particularly in shallow water environments, characterizing the bottom reflection loss with a precision that databank-based modeling cannot achieve. Recent advances in the technology of autonomous underwater vehicles (AUV) make it possible to envision a survey system for seabed characterization composed of a short array mounted on a small AUV. The bottom power reflection coefficient (and the related reflection loss) can be estimated passively by beamforming the naturally occurring marine ambient-noise acoustic field recorded by a vertical line array of hydrophones. However, the reduced array lengths required by small AUV deployment can hinder the process, due to the inherently poor angular resolution. In this dissertation, original data-processing techniques are presented which, by introducing into the processing chain knowledge derived from physics, can improve the performance of short arrays in this particular task. Particularly, the analysis of a model of the ambient-noise spatial coherence function leads to the development of a new proof of the result at the basis of the bottom reflection-loss estimation technique. The proof highlights some shortcomings inherent in the beamforming operation so far used in this technique. A different algorithm is then proposed, which removes the problem achieving improved performance. Furthermore, another technique is presented that uses data from higher frequencies to estimate the noise spatial coherence function at a lower frequency, for sensor spacing values beyond the physical length of the array. By "synthesizing" a longer array, the angular resolution of the bottom-loss estimate can be improved, often making use of data at frequencies above the array design frequency, otherwise not utilized for beamforming. The proposed algorithms are demonstrated both in simulation and on real data acquired during several experimental campaigns.