Dans le cadre de cette thèse, nous nous intéressons à la coordination et l'optimisation du trafic aux intersections des réseaux routiers, avec la particularité de considérer des véhicules autonomes intelligents. Cette thèse est organisée en deux grandes parties. La première se concentre sur le problème du partage d'un espace de voie par deux files de véhicules évoluant en sens opposés. L'état de l'art montre le peu de travaux abordant cette question. Nous explorons deux approches par coordination réactive, en relation avec un critère de minimisation des retards. Les performances de ces approches ont été mesurées statistiquement en simulation. La deuxième partie de la thèse s'attaque au problème générique de la gestion du trafic au sein d'un réseau routier. Nous développons une approche originale à deux égards: d'une part elle explore un principe de passage en alternance des flux permettant de ne pas arrêter les véhicules aux intersections, et d'autre part, elle propose des algorithmes d'optimisationdistribuée de ce passage alterné au niveau de chaque intersection et au niveau du réseau global. La thèse présente successivement les choix de modélisation, les algorithmes et l'étude en simulation de leurs performances comparées à desapproches existantes / In this thesis, we focus on traffic coordination and optimization in road intersections, while accounting for intelligent autonomous vehicles. This thesis is organized in two parts. The first part focuses on the problem of sharing a one-lane road between two opposite flows of vehicles. The state of the art shows few studies addressing this issue. We propose two reactive coordination approaches that minimize vehicle delays and measure their performances statistically through simulations. The second part of the thesis addresses the problem of generic traffic management in a traffic network. We develop a stop-free approach that explores a principle alternating vehicles between flows at intersections, and it provides distributed algorithms optimizing this alternation at each intersection and in the overall network. We present the modeling choices, the algorithms and the simulation study of our approach and we compare its performances with existing approaches

Intelligence, surveillance, and reconnaissance unmanned aircraft systems (UAS) are the most common variety of UAS in use today and provide invaluable capabilities to both the military and civil services. Keeping the sensors centered on a point of interest for an extended period of time is a demanding task requiring the full attention and cooperation of the UAS pilot and sensor operator. There is great interest in developing technologies which allow an operator to designate a target and allow the aircraft to automatically maneuver and track the designated target without operator intervention. Presently, the barriers to entry for developing these technologies are high: expertise in aircraft dynamics and control as well as in real- time motion video analysis is required and the cost of the systems required to flight test these technologies is prohibitive. However, if the research intent is purely to develop a vehicle maneuvering controller then it is possible to obviate the video analysis problem entirely. This research presents a solution to the target tracking problem which reliably provides automatic target detection and tracking with low expense and computational overhead by making use of the infrared sensor from a Nintendo Wii Remote Controller.

Approved for public release; distribution is unlimited / Unmanned vehicle technology has matured significantly over the last two decades. This is evidenced by its widespread use in industrial and military applications ranging from deep-ocean exploration to anti-submarine warefare. Indeed, the feasiblity of short-range, special-purpose vehicles (whether aunonomous or remotely operated) is no longer in question. The research efforts have now begun to shift their focus on development of reliable, longer-range, high-endurance and fully autonomous systems. One of the major underlying technologies required to realize this goal is Artificial Intelligence (AI). The latter offers great potential to endow vehicles with the intelligence needed for full autonomy and extended range capability; this involves the increased application of AI technologies to support mission planning and execution, navigation and contingency planning. This thesis addresses two issues associated with the above goal for Autonomous Underwater Vehicles (AUV's). Firstly, a new approach is proposed for path planning in underwater environments that is capable of dealing with uncharted obstacles and which requires significantly less planning time and computer memory. Secondly, it explores the use of expert system technology in the planning of AUV missions.

Autonomous vehicles have made milestone strides within the past decade. Advances up the autonomy ladder have come lock-step with the advances in machine learning, namely deep-learning algorithms and huge, open training sets. And while advances in CPUs have slowed, GPUs have edged into the previous decade's TOP 500 supercomputer territory. This new class of GPUs include novel deep-learning hardware that has essentially side-stepped Moore's law, outpacing the doubling observation by a factor of ten. While GPUs have make record progress, networks do not follow Moore's law and are restricted by several bottlenecks, from protocol-based latency lower bounds to the very laws of physics. In a way, the bottlenecks that plague modern networks gave rise to Edge computing, a key component of the Connected Autonomous Vehicle system, as the need for low-latency in some domains eclipsed the need for massive processing farms. The Connected Autonomous Vehicle ecosystem is one of the most complicated environments in all of computing. Not only is the hardware scaled all the way from 16 and 32-bit microcontrollers, to multi-CPU Edge nodes, and multi-GPU Cloud servers, but the networking also encompasses the gamut of modern communication transports. I propose a framework for negotiating, encapsulating and transferring data between vehicles ensuring efficient bandwidth utilization and respecting real-time privacy levels.

abstract: Recently, automation, shared use, and electrification are proposed and viewed as the “three revolutions” in the future transportation sector to significantly relieve traffic congestion, reduce pollutant emissions, and increase transportation system sustainability. Motivated by the three revolutions, this research targets on the passenger-focused scheduled transportation systems, where (1) the public transit systems provide high-quality ridesharing schedules/services and (2) the upcoming optimal activity planning systems offer the best vehicle routing and assignment for household daily scheduled activities. The high quality of system observability is the fundamental guarantee for accurately predicting and controlling the system. The rich information from the emerging heterogeneous data sources is making it possible. This research proposes a modeling framework to systemically account for the multi-source sensor information in urban transit systems to quantify the estimated state uncertainty. A system of linear equations and inequalities is proposed to generate the information space. Also, the observation errors are further considered by a least square model. Then, a number of projection functions are introduced to match the relation between the unique information space and different system states, and its corresponding state estimate uncertainties are further quantified by calculating its maximum state range. In addition to optimizing daily operations, the continuing advances in information technology provide precious individual travel behavior data and trip information for operational planning in transit systems. This research also proposes a new alternative modeling framework to systemically account for boundedly rational decision rules of travelers in a dynamic transit service network with tight capacity constraints. An agent-based single-level integer linear formulation is proposed and can be effectively by the Lagrangian decomposition. The recently emerging trend of self-driving vehicles and information sharing technologies starts creating a revolutionary paradigm shift for traveler mobility applications. By considering a deterministic traveler decision making framework, this research addresses the challenges of how to optimally schedule household members’ daily scheduled activities under the complex household-level activity constraints by proposing a set of integer linear programming models. Meanwhile, in the microscopic car-following level, the trajectory optimization of autonomous vehicles is also studied by proposing a binary integer programming model. / Dissertation/Thesis / Doctoral Dissertation Civil, Environmental and Sustainable Engineering 2018

No description available.

Le véhicule autonome offre plusieurs avantages : le confort, la réduction du stress, et la réduction de la mortalité routière. Néanmoins, les accidents mortels impliquant sa responsabilité, ont mis en exergue ses limitations et ses imperfections. Ces accidents soulèvent des questions sur la fiabilité et des voix ont fait part d'une forte préoccupation pour la sécurité des usagers du véhicule autonome. En outre, les tâches de perception et de localisation des véhicules autonomes peuvent avoir des incohérences amenant à des erreurs qui nuiraient à la stabilité du véhicule. Les sources de ces incohérences peuvent être de natures différentes et agir à la fois sur le capteur lui-même (Hardware), ou bien sur le post-traitement de l'information (Software). Dans ce contexte, plusieurs difficultés doivent être surmontées pour arriver à sécuriser l'interaction des systèmes automatisés de conduite avec les conducteurs humains face à ces problèmes, l'adoption d'une stratégie de commande tolérante aux défauts est primordiale. Dans le cadre de cette thèse, des stratégies de détection et de tolérance aux fautes pour la perception et la localisation sont mise en œuvre. En outre, des stratégies de détection et d'estimation de défauts pour les capteurs proprioceptifs sont par ailleurs proposées. L'objectif est d'avoir une localisation fiable de défaut et assurer un fonctionnement avec des performances acceptables. Par ailleurs, vue l'imprédictibilité et la variété des scènes routières, une fusion tolérante aux fautes à base des algorithmes de vote est élaborée pour une meilleure perception. La fusion tire profit des technologies actuelles de détection d'obstacles (détection par radio, faisceaux lumineux ou par caméra) et l'algorithme de vote assure une sortie qui s'approche le plus de la réalité. Des tests avec des données réelles issues d'un véhicule démonstrateur sont utilisés pour valider les approches proposées dans cette thèse. / Driverless vehicle offers several advantages: comfort, reduced stress, and reduced road mortality. Nevertheless, fatal accidents involving its responsibility, have highlighted its limitations and imperfections. These accidents raise questions about autonomous vehicle reliability, and voices expressed a strong concern for the safety of users of the autonomous vehicle. Furthermore, the tasks of perception and localization of autonomous vehicles may have inconsistencies leading to errors that would affect the stability of the vehicle. The sources of these inconsistencies can be of different natures and act both on the sensor itself (Hardware), or on the post-processing of information (Software). In this context, several difficulties must be overcome to secure the interaction of automated driving systems with human drivers facing these problems, the adoption of a fault-tolerant control strategy is paramount. In this thesis, a fault detection and fault tolerant control strategies for perception and localization are implemented. In addition, fault estimation strategies for proprioceptive sensors are also proposed. The purpose is to have a reliable fault localization and ensure acceptable performance. Moreover, given the unpredictability and variety of road scenes, a fault-tolerant fusion based on voting algorithms is developed for a better perception. The fusion takes advantage of current obstacle detection technologies (radio, light beam or camera detection) and the voting algorithm provides an output that is closest to reality. Tests with real data from a demonstrator vehicle are used to validate the approaches proposed in this thesis.

<div> <div> <div> <p>For pedestrians and autonomous vehicles (AVs) to co-exist harmoniously and safely in the real-world, AVs will need to not only react to pedestrian actions, but also anticipate their intentions. In this thesis, we propose to use rich visual and pedestrian-environment interaction features to improve pedestrian crossing intention prediction from the ego-view. We do so by combining visual feature extraction, graph modeling of scene objects and their relationships, and feature encoding as comprehensive inputs for an LSTM encoder-decoder network. </p> <p>Pedestrians react and make decisions based on their surrounding environment, and the behaviors of other road users around them. The human-human social relationship has already been explored for pedestrian trajectory prediction from the bird’s eye view in stationary cameras. However, context and pedestrian-environment relationships are often missing in current research into pedestrian trajectory, and intention prediction from the ego-view. To map the pedestrian’s relationship to its surrounding objects we use a star graph with the pedestrian in the center connected to all other road objects/agents in the scene. The pedestrian and road objects/agents are represented in the graph through visual features extracted using state of the art deep learning algorithms. We use graph convolutional networks, and graph autoencoders to encode the star graphs in a lower dimension. Using the graph en- codings, pedestrian bounding boxes, and human pose estimation, we propose a novel model that predicts pedestrian crossing intention using not only the pedestrian’s action behaviors (bounding box and pose estimation), but also their relationship to their environment. </p> <p>Through tuning hyperparameters, and experimenting with different graph convolutions for our graph autoencoder, we are able to improve on the state of the art results. Our context- driven method is able to outperform current state of the art results on benchmark dataset Pedestrian Intention Estimation (PIE). The state of the art is able to predict pedestrian crossing intention with a balanced accuracy (to account for dataset imbalance) score of 0.61, while our best performing model has a balanced accuracy score of 0.79. Our model especially outperforms in no crossing intention scenarios with an F1 score of 0.56 compared to the state of the art’s score of 0.36. Additionally, we also experiment with training the state of the art model and our model to predict pedestrian crossing action, and intention jointly. While jointly predicting crossing action does not help improve crossing intention prediction, it is an important distinction to make between predicting crossing action versus intention.</p> </div> </div> </div>

Two controller performances are assessed for generalization in the path following task of autonomously backing up a tractor-trailer. Starting from random locations and orientations, paths are generated to loading docks with arbitrary pose using Dubins Curves. The combination vehicles can be varied in wheelbase, hitch length, weight distributions, and tire cornering stiffness. The closed form calculation of the gains for the Linear Quadratic Regulator (LQR) rely heavily on having an accurate model of the plant. However, real-world applications cannot expect to have an updated model for each new trailer. Finding alternative robust controllers when the trailer model is changed was the motivation of this research. Reinforcement learning, with neural networks as their function approximators, can allow for generalized control from its learned experience that is characterized by a scalar reward value. The Linear Quadratic Regulator and the Deep Deterministic Policy Gradient (DDPG) are compared for robust control when the trailer is changed. This investigation quantifies the capabilities and limitations of both controllers in simulation using a kinematic model. The controllers are evaluated for generalization by altering the kinematic model trailer wheelbase, hitch length, and velocity from the nominal case. In order to close the gap from simulation and reality, the control methods are also assessed with sensor noise and various controller frequencies. The root mean squared and maximum errors from the path are used as metrics, including the number of times the controllers cause the vehicle to jackknife or reach the goal. Considering the runs where the LQR did not cause the trailer to jackknife, the LQR tended to have slightly better precision. DDPG, however, controlled the trailer successfully on the paths where the LQR jackknifed. Reinforcement learning was found to sacrifice a short term reward, such as precision, to maximize the future expected reward like reaching the loading dock. The reinforcement learning agent learned a policy that imposed nonlinear constraints such that it never jackknifed, even when it wasn't the trailer it trained on.

This thesis discusses autonomous vehicles. At first it contains describing development of these type of vehicles, how they work and discuss their future development. Further it describe tools which can be used for create model of autonomous vehicle. The thesis includes design, description of the development and testing of the model in the UPPAAL Stratego verification environment. The resulting model is a system of intercommunicating timed automata. The analysis of the model properties is based on the method of statistical verification. The model allows us to investigate behavior of an autonomous vehicle in situations which correspond to regular traffic.