Modern automobiles and particularly those with off-road lineage possess subsystems that can be configured to better negotiate certain terrain types. Different terrain classes amount to different adherence (or surface grip) and compressibility properties that impact vehicle ma-noeuvrability and should therefore incur a tailored throttle response, suspension stiffness and so on. This thesis explores prospective terrain recognition for an anticipating terrain response driver assistance system. Recognition of terrain and road terrain is cast as a semantic segmen-tation task whereby forward driving images or point clouds are pre-segmented into atomic units and subsequently classified. Terrain classes are typically of amorphous spatial extent con-taining homogenous or granularly repetitive patterns. For this reason, colour and texture ap-pearance is the saliency of choice for monocular vision. In this work, colour, texture and sur-face saliency of atomic units are obtained with a bag-of-features approach. Five terrain classes are considered, namely grass, dirt, gravel, shrubs and tarmac. Since colour can be ambiguous among terrain classes such as dirt and gravel, several texture flavours are explored with scalar and structured output learning in a bid to devise an appropriate visual terrain saliency and predictor combination. Texture variants are obtained using local binary patters (LBP), filter responses (or textons) and dense key-point descriptors with daisy. Learning algorithms tested include support vector machine (SVM), random forest (RF) and logistic regression (LR) as scalar predictors while a conditional random field (CRF) is used for structured output learning. The latter encourages smooth labelling by incorporating the prior knowledge that neighbouring segments with similar saliency are likely segments of the same class. Once a suitable texture representation is devised the attention is shifted from monocular vision to stereo vision. Sur-face saliency from reconstructed point clouds can be used to enhance terrain recognition. Pre-vious superpixels span corresponding supervoxels in real world coordinates and two surface saliency variants are proposed and tested with all predictors: one using the height coordinates of point clouds and the other using fast point feature histograms (FPFH). Upon realisation that road recognition and terrain recognition can be assumed as equivalent problems in urban en-vironments, the top most accurate models consisting of CRFs are augmented with composi-tional high order pattern potentials (CHOPP). This leads to models that are able to strike a good balance between smooth local labelling and global road shape. For urban environments the label set is restricted to road and non-road (or equivalently tarmac and non-tarmac). Ex-periments are conducted using a proprietary terrain dataset and a public road evaluation da-taset.

A segurança no tráfego de carros é um assunto em foco nos dias de hoje e, dentro dele, podem-se citar os sistemas de auxílio ao motorista que vêm sendo desenvolvidos com a finalidade de reduzir o grande número de fatalidades em acidentes de trânsito. Tais sistemas de auxílio buscam mitigar falhas humanas como falta de atenção e imprudência. Visto isso, o projeto SENA, desenvolvido pelo Laboratório de Robótica Móvel da Escola de Engenharia de São Carlos, busca contribuir com a evolução dessa assistência ao motorista. O presente trabalho realiza um estudo sobre uma técnica de inteligência artificial chamada de Redes Bayesianas. Essa técnica merece atenção em virtude de sua capacidade de tratar dados incertos em forma de probabilidades. A rede desenvolvida por esse trabalho utiliza, como dados de entrada, os classificadores em desenvolvimento no projeto SENA e tem como resposta um comportamento que o veículo deve executar, por um ser humano ou por um planejador de trajetórias. Em função da alta dimensionalidade do problema abordado, foram realizados dois experimentos em ambiente simulado de duas situações distintas. A primeira, um teste de frenagem próximo a um ponto de intersecção e a segunda, um cenário de entroncamento. Os testes feitos com a rede indicam que classificadores pouco discriminantes deixam o sistema mais propenso a erros e que erros na localização do ego-veículo afetam mais o sistema se comparado a erros na localização dos outros veículos. Os experimentos realizados mostram a necessidade de um sistema de tempo real e um hardware mais adequado para tratar as informações mais rapidamente / The safety of cars in traffic scenarios is being addressed on the past few years. One of its topics is the Advanced Driver-Assistance Systems which have been developed to reduce the fatality numbers of traffic accidents. These systems try to decrease human failures, such as imprudence and lack of attention while driving. For these reasons, the SENA project, in progress on the Mobile Robotics Laboratory at the Sao Carlos School of Engineering (EESC), aims to contribute for the evolution of these assistance systems. This work studies an artificial intelligence technique called Bayesian Networks. It deserves our attention due to its capability of handling uncertainties with probability distributions. The network developed in this Masters Thesis has, as input, the result of the classifiers used on SENA project and has, as output, a behavior which has to be performed by the vehicle with a driver or autonomously by the means of a path planner. Due to the high dimensionality of this issue, two different tests have been carried out. The first one was a braking experiment near a intersection point and the other one was a T-junction scenario. The tests made indicate that weak classifiers leaves the system more instable and error-prone and localization errors of the ego-vehicle have a stronger effect than just localization errors of other traffic participants. The experiments have shown that there is a necessity for a real-time system and a hardware more suitable to deal quickly with the information

京都大学 / 0048 / 新制・課程博士 / 博士(工学) / 甲第18689号 / 工博第3967号 / 新制||工||1611(附属図書館) / 31622 / 京都大学大学院工学研究科機械理工学専攻 / (主査)教授 椹木 哲夫, 教授 西脇 眞二, 教授 松原 厚 / 学位規則第4条第1項該当 / Doctor of Philosophy (Engineering) / Kyoto University / DFAM

Advanced Driver Assistance Systems (ADAS) and Autonomous Ground Vehicles (AGV) have the potential to increase road transportation safety, environmental gains, and passenger comfort. The advent of Electric Vehicles has also facilitated greater flexibility in powertrain architectures and control capabilities. Path Tracking controllers that provide steering input are used to execute lateral maneuvers or model the response of a vehicle during cornering. Direct Yaw Control using Torque Vectoring has the potential to improve vehicle's transient cornering stability and modify its steady state handling characteristics during lateral maneuvers. In the first part of this thesis, the transient dynamics of an existing baseline Path Tracking controller is improved using a transient Torque Vectoring algorithm. The existing baseline Path Tracking controller is evaluated, using a linearized system, for a range of vehicle and controller parameters. The effect of implementing transient Torque Vectoring along with the baseline Path Tracking controller is then studied for the same parameter range. The linear analysis shows, in both time and frequency domain, that the transient Torque Vectoring improves vehicle response and stability during cornering. A Torque Vectoring controller is developed in Linear Adaptive Model Predictive Control framework and it's performance is verified in simulation using Simulink and CarSim. The second part of the thesis analyzes the tradeoff enabled by steady state Torque Vectoring between improved limit handling capability through optimal tire force allocation and drivability demonstrated by understeer gradient. Optimal tire force allocation prescribes equal usage in all four tires during maneuvers. This is enabled using steering and Torque Vectoring. An analytical proof is presented which demonstrates that implementation of this optimal tire force allocation results in neutralsteering handling characteristics for the vehicle. The optimal tire force allocation strategy is formulated as a minimax optimization problem. A two-track vehicle model is simulated for this strategy, and it verified the analytical proof by displaying neutralsteering behavior. / Master of Science / Advanced Driver Assistance Systems (ADAS) and Autonomous Ground Vehicles (AGVs) have the potential to increase road transportation safety, environmental gains, passenger comfort and passenger productivity. The advent of Electric Vehicles (EVs) has also facilitated greater flexibility in powertrain configurations and capabilities that facilitate the implementation of Torque Vectoring (TV), which is a method of applying differential torques to laterally opposite wheels to enhance the cornering performance of ground vehicles. Path Tracking (PT) controllers that provide steering input to the vehicles are traditionally used for lateral control in AGVs and ADAS features. The goal of this thesis is to develop Torque Vectoring algorithms to improve a vehicle's stability and shape its steady state behaviour through a corner during low lateral acceleration maneuvers. An existing baseline Path Tracking controller is selected and evaluated. The effect of implementing Torque Vectoring along with this Path Tracking controller is studied and it is found to improve the stability of the vehicle during cornering. This is verified in simulation by designing and implementing the Torque Vectoring algorithm. Finally, a Torque Vectoring strategy is proposed to manage the handling of the vehicle during low acceleration cornering.

Intersection crashes are among the most frequent and lethal crash modes in the United States. Intersection Advanced Driver Assistance Systems (I-ADAS) are an emerging active safety technology which aims to help drivers safely navigate through intersections. One primary function of I-ADAS is to detect oncoming vehicles and in the event of an imminent collision can (a) alert the driver and/or (b) autonomously evade the crash. Another function of I-ADAS may be to detect and prevent imminent traffic signal violations (i.e. running a red light or stop sign) earlier in the intersection approach, while the driver still has time to yield for the traffic control device. This dissertation evaluated the capacity of I-ADAS to prevent U.S. intersection crashes and mitigate associated injuries. I-ADAS was estimated to have the potential to prevent up to 64% of crashes and 79% of vehicles with a seriously injured driver. However, I-ADAS effectiveness was found to be highly dependent on driver behavior, system design, and intersection/roadway characteristics. To generate this result, several studies were performed. First, driver behavior at intersections was examined, including typical, non-crash intersection approach and traversal patterns, the acceleration patterns of drivers prior to real-world crashes, and the frequency, timing, and magnitude of any crash avoidance actions. Second, two large simulation case sets of intersection crashes were generated from U.S. national crash databases. Third, the developed simulation case sets were used to examine I-ADAS performance in real-world crash scenarios. This included examining the capacity of a stop sign violation detection algorithm, investigating the sensor detection needs of I-ADAS technology, and quantifying the proportion of crashes and seriously injuries that are potentially preventable by this crash avoidance technology. / Ph. D.

abstract: Driving is already a complex task that demands a varying level of cognitive and physical load. With the advancement in technology, the car has become a place for media consumption, a communications center and an interconnected workplace. The number of features in a car has also increased. As a result, the user interaction inside the car has become overcrowded and more complex. This has increased the amount of distraction while driving and has also increased the number of accidents due to distracted driving. This thesis focuses on the critical analysis of today’s in-car environment covering two main aspects, Multi Modal Interaction (MMI), and Advanced Driver Assistance Systems (ADAS), to minimize the distraction. It also provides deep market research on future trends in the smart car technology. After careful analysis, it was observed that an infotainment screen cluttered with lots of small icons, a center stack with a plethora of small buttons and a poor Voice Recognition (VR) results in high cognitive load, and these are the reasons for the increased driver distraction. Though the VR has become a standard technology, the current state of technology is focused on features oriented design and a sales driven approach. Most of the automotive manufacturers are focusing on making the VR better but attaining perfection in VR is not the answer as there are inherent challenges and limitations in respect to the in-car environment and cognitive load. Accordingly, the research proposed a novel in-car interaction design solution: Multi-Modal Interaction (MMI). The MMI is a new term when used in the context of vehicles, but it is widely used in human-human interaction. The approach offers a non-intrusive alternative to the driver to interact with the features in the car. With the focus on user-centered design, the MMI and ADAS can potentially help to reduce the distraction. To support the discussion, an experiment was conducted to benchmark a minimalist UI design. An engineering based method was used to test and measure distraction of four different UIs with varying numbers of icons and screen sizes. Lastly, in order to compete with the market, the basic features that are provided by all the other competitors cannot be eliminated, but the hard work can be done to improve the HCaI and to make driving safer. / Dissertation/Thesis / Date collected about reaction time in the experiment_Excel / Masters Thesis Computer Science 2015

abstract: The environmental impact of the fossil fuels has increased tremendously in the last decade. This impact is one of the most contributing factors of global warming. This research aims to reduce the amount of fuel consumed by vehicles through optimizing the control scheme for the future route information. Taking advantage of more degrees of freedom available within PHEV, HEV, and FCHEV “energy management” allows more margin to maximize efficiency in the propulsion systems. The application focuses on reducing the energy consumption in vehicles by acquiring information about the road grade. Road elevations are obtained by use of Geographic Information System (GIS) maps to optimize the controller. The optimization is then reflected on the powertrain of the vehicle.The approach uses a Model Predictive Control (MPC) algorithm that allows the energy management strategy to leverage road grade to prepare the vehicle for minimizing energy consumption during an uphill and potential energy harvesting during a downhill. The control algorithm will predict future energy/power requirements of the vehicle and optimize the performance by instructing the power split between the internal combustion engine (ICE) and the electric-drive system. Allowing for more efficient operation and higher performance of the PHEV, and HEV. Implementation of different strategies, such as MPC and Dynamic Programming (DP), is considered for optimizing energy management systems. These strategies are utilized to have a low processing time. This approach allows the optimization to be integrated with ADAS applications, using current technology for implementable real time applications. The Thesis presents multiple control strategies designed, implemented, and tested using real-world road elevation data from three different routes. Initial simulation based results show significant energy savings. The savings range between 11.84% and 25.5% for both Rule Based (RB) and DP strategies on the real world tested routes. Future work will take advantage of vehicle connectivity and ADAS systems to utilize Vehicle to Vehicle (V2V), Vehicle to Infrastructure (V2I), traffic information, and sensor fusion to further optimize the PHEV and HEV toward more energy efficient operation. / Dissertation/Thesis / Masters Thesis Mechanical Engineering 2016

The aim of this diploma thesis is the research of building polygons for the testing of autonomous vehicles, from the point of view of road technology and also designing aspects. In the thesis are mentioned 9 most important world test polygons and their description of design parameters. There are described particular stages of automation from foreign organizations which are concerned with research and development in the automotive industry. In addition, there are described basic advanced driver assistance systems and connectivity between vehicles and infrastructure. Conclusion also contains the assessment of existing aerodrome test areas for autonomous vehicles.

abstract: Object detection is an interesting computer vision area that is concerned with the detection of object instances belonging to specific classes of interest as well as the localization of these instances in images and/or videos. Object detection serves as a vital module in many computer vision based applications. This work focuses on the development of object detection methods that exhibit increased robustness to varying illuminations and image quality. In this work, two methods for robust object detection are presented. In the context of varying illumination, this work focuses on robust generic obstacle detection and collision warning in Advanced Driver Assistance Systems (ADAS) under varying illumination conditions. The highlight of the first method is the ability to detect all obstacles without prior knowledge and detect partially occluded obstacles including the obstacles that have not completely appeared in the frame (truncated obstacles). It is first shown that the angular distortion in the Inverse Perspective Mapping (IPM) domain belonging to obstacle edges varies as a function of their corresponding 2D location in the camera plane. This information is used to generate object proposals. A novel proposal assessment method based on fusing statistical properties from both the IPM image and the camera image to perform robust outlier elimination and false positive reduction is also proposed. In the context of image quality, this work focuses on robust multiple-class object detection using deep neural networks for images with varying quality. The use of Generative Adversarial Networks (GANs) is proposed in a novel generative framework to generate features that provide robustness for object detection on reduced quality images. The proposed GAN-based Detection of Objects (GAN-DO) framework is not restricted to any particular architecture and can be generalized to several deep neural network (DNN) based architectures. The resulting deep neural network maintains the exact architecture as the selected baseline model without adding to the model parameter complexity or inference speed. Performance results provided using GAN-DO on object detection datasets establish an improved robustness to varying image quality and a higher object detection and classification accuracy compared to the existing approaches. / Dissertation/Thesis / Doctoral Dissertation Electrical Engineering 2020

Navigation systems are one of the most commonly found electronic gadgets in modern vehicles nowadays. Alongside navigation units this technology is made readily available to individuals in everyday devices such as a mobile phone. Digital maps which come preloaded on these devices accommodate within them an extensive dataset of spatial information from around the globe which aids the driver in achieving a well guided driving experience. Apart from being essential for navigation this sensor information backs up other vehicular applications in making intelligent decisions. The quality of this information delivered is in direct relation to the underlying dataset used to produce these maps. Since we live in a highly dynamic environment with constantly changing geography, an effort is necessary to keep these maps updated with the most up to date information as frequently as possible. The digital map of interest in this study is OpenStreetMap, the underlying data of which is a combination of donated as well as crowdsourced information from the last 10 years. This extensive dataset helps in building of a detailed digital map of the world using well defined cartographic techniques. The information within OpenStreetMap is currently enhanced by a large group of volunteers who willing use donated satellite imagery, uploaded GPS tracks, field surveys etc. to correct and collect necessary data for a region of interest. Though this method helps in improving and increasing the quality and quantity of the OpenStreetMap dataset, it is very time consuming and requires a great deal of human effort. Through this thesis an effort is made to automatically enrich this dataset by preprocessing crowdsourced sensor data collected from the navigation system and driver assistance systems (Traffic Sign Recognition system and a Lane Detection System) of a driving vehicle. The kind of data that is algorithmically derived includes the calculation of the curvature of the underlying road, correction of speed limit values for individual road segments being driven and the identification of change in the geometry of existing roads due to closure of old ones or addition of new ones in the Nuremberg region of Bavaria, Germany. Except for a small percentage of speed limit information on roads segments, other information is currently not available in the OpenStreetMap database for use in safety and comfort related applications. The navigation system has the ability to deliver geographical data in form of GPS coordinates at a certain frequency. This set of GPS coordinates can grouped together to form a GPS track visualizing the actual path traversed by a driving vehicle. A large number of such GPS tracks repeatedly collected from different vehicles driving in a region of interest gives all GPS points which lie on a particular road. These points, after outlier elimination methods are used as a dataset to scientifically determine the underlying curvature of the road with the aid of curve fitting techniques. Additional information received from the lane detection system helps identify curves on a road for which the curvature must be calculated. The fusion of information from these sources helps to achieve curvature results with high accuracy. Traffic sign recognition system helps detect traffic signs while driving, the fusion of this data with geographical information from the navigation system at the instance of detection helps determine road segments for which the recognized speed limit values are valid. This thesis successfully demonstrates a method to automatically enrich OpenStreetMap data by crowdsourcing raw sensor data from multiple vehicles equipped with driver assistance systems. All OpenStreetMap attributes were 100% updated into the database and the results have proven the effectiveness our system architecture. The positive results obtained in combination with minimal errors promise a better future for assisted driving.