Modern camera-based vehicle safety systems heavily rely on machine learning and consequently require large amounts of training data to perform reliably. However, collecting and annotating the needed data is an extremely expensive and time-consuming process. In addition, it is exceptionally difficult to collect data that covers critical scenarios. This thesis investigates to what extent synthetic data can replace real-world data for these scenarios. Since only a limited amount of data consisting of such real-world scenarios is available, this thesis instead makes use of proxy scenarios, e.g. situations when pedestrians are located closely in front of the vehicle (for example at a crosswalk). The presented approach involves training a detector on real-world data where all samples of these proxy scenarios have been removed and compare it to other detectors trained on data where the removed samples have been replaced with various degrees of synthetic data. A method for generating and automatically and accurately annotating synthetic data, using features in the CARLA simulator, is presented. Also, the domain gap between the synthetic and real-world data is analyzed and methods in domain adaptation and data augmentation are reviewed. The presented experiments show that aligning statistical properties between the synthetic and real-world datasets distinctly mitigates the domain gap. There are also clear indications that synthetic data can help detect pedestrians in critical traffic situations / <p>Examensarbetet är utfört vid Institutionen för teknik och naturvetenskap (ITN) vid Tekniska fakulteten, Linköpings universitet</p>

People in cities have a need of recreation. Leisure activities are a big part of people’s health and well- being. Many leisure activities are practiced outside in the nature. Today many citizens do not own a car and public transportation fail to meet the peoples need to move out from the city center out into the wild nature. Public transportation run according to a certain timetable and an established route. They also come with restrictions of what you can bring on board. In other words, there is a need for agile transportation that transport people and their equipment from the city out to the nature; and the need will only increase as the urbanization continue. The whole idea of the project originated from the question, “How might future mobility adapt to fit people’s need for recreation?” The project started with a research. The author studied trends how cities will develop, what defines generation Z, how 6G can be used in the future transportation industry, how electrification changes the terms of car design, the current status of autonomous technology etc. The research also included Canoo, a car company that design, develop and build electric vehicles with focus on lifestyle, utility and sustainability. After completed research, the project moved into a creative phase which included analog sketches of the vehicle, testing of proportions in scale 1: 1 and a storyboard that describes how the vehicle can be used. When thecreative phase was done, the project moved to its final phase; visualization. A 3D model was constructed in Autodesk Maya, a polygonal modelling software, and rendered images of the 3D model was conducted with Autodesk VRED, a 3D visualization software. The project resulted in Canoo Link; Link, targeting year 2035, is an electric autonomous utility vehicle that you subscribe to. It can carry 4 passengers and has storage space to haul equipment and supplies for canoeing, mountain biking etc. With its robust design and high ground clearance it is ready to take on some tough terrain. The subscription offers the customer full disposal of the vehicle during the activities for convenience and security. It acts like a hub for your activities, not just as a vehicle for commuting. To summarize; Link is a design proposal of a vehicle that connect people living in cities to the nature. It is not just a car that takes you from one point to another, it is a lifestyle.

Background: The transport industry is facing new challenges such as increased competition between the actors and an increasing shortage of truck drivers. Implementing new technologies such as autonomous driving can represent a solution for companies to increase their competitiveness and gains. However, implementing such an innovative solution leads to a certain resistance to change that has to be dealt with, as well as concerns about the current jobs within the industry. Purpose: The purpose of this thesis is to assess the resistance to change linked to implementing this technology within Logistics Service Providers, provide solutions to overcome this resistance, as well as assessing the impact on jobs. Method: An inductive multiple case study has been used to conduct this research. The data was gathered from 12 semi-structured interviews with experts related to the transport industry. Then, thematic data analysis has been used to provide insights. Conclusion: The results show that the resistance is characterized by barriers to the technology and the resistance from the people, that support and communication is the key factor for successful implementation and that the truck driving professions will evolve considerably.

Human driving decisions are the leading cause of road fatalities. Autonomous driving naturally eliminates such incompetent decisions and thus can improve traffic safety and efficiency. Deep reinforcement learning (DRL) has shown great potential in learning complex tasks. Recently, researchers investigated various DRL-based approaches for autonomous driving. However, exploiting multi-modal fusion to generate pixel-wise perception and motion prediction and then leveraging these predictions to train a latent DRL has not been targeted yet. Unlike other DRL algorithms, the latent DRL algorithm distinguishes representation learning from task learning, enhancing sampling efficiency for reinforcement learning. In addition, supplying the latent DRL algorithm with accurate perception and motion prediction simplifies the surrounding urban scenes, improving training and thus learning a better driving policy. To that end, this Ph.D. research initially develops LiCaNext, a novel real-time multi-modal fusion network to produce accurate joint perception and motion prediction at a pixel level. Our proposed approach relies merely on a LIDAR sensor, where its multi-modal input is composed of bird's-eye view (BEV), range view (RV), and range residual images. Further, this Ph.D. thesis proposes leveraging these predictions with another simple BEV image to train a sequential latent maximum entropy reinforcement learning (MaxEnt RL) algorithm. A sequential latent model is deployed to learn a more compact latent representation from high-dimensional inputs. Subsequently, the MaxEnt RL model trains on this latent space to learn a driving policy. The proposed LiCaNext is trained on the public nuScenes dataset. Results demonstrated that LiCaNext operates in real-time and performs better than the state-of-the-art in perception and motion prediction, especially for small and distant objects. Furthermore, simulation experiments are conducted on CARLA to evaluate the performance of our proposed approach that exploits LiCaNext predictions to train sequential latent MaxEnt RL algorithm. The simulated experiments manifest that our proposed approach learns a better driving policy outperforming other prevalent DRL-based algorithms. The learned driving policy achieves the objectives of safety, efficiency, and comfort. Experiments also reveal that the learned policy maintains its effectiveness under different environments and varying weather conditions.

Optimal control of autonomous systems is a fundamental and challenging problem, especially when many stringent safety constraints and tight control limitations are involved such that solutions are hard to determine. It has been shown that optimizing quadratic costs while stabilizing affine control systems to desired (sets of) states subject to state and control constraints can be reduced to a sequence of Quadratic Programs (QPs) by using Control Barrier Functions (CBFs) and Control Lyapunov Functions (CLFs). Although computationally efficient, this method is limited by several factors which are addressed in this dissertation. The first contribution of this dissertation is to extend CBFs to high order CBFs (HOCBFs) that can accommodate arbitrary relative degree systems and constraints. The satisfaction of Lyapunov-like conditions in the HOCBF method implies the forward invariance of the intersection of a sequence of sets, which can then guarantee the satisfaction of the original safety constraint. Second, under tight control bounds, this dissertation proposes an analytical method to find sufficient conditions that guarantee the QP feasibility. The sufficient conditions are captured by a single state constraint that is enforced by a CBF and then added to the QP. Third, for complex safety constraints and systems in which it is hard to find sufficient conditions for feasibility, machine learning techniques are employed to learn the definitions of HOCBFs or feasibility constraints. Fourth, when time-varying control bounds and noisy dynamics are involved, adaptive CBFs (AdaCBFs) are proposed, which can guarantee the feasibility of the QPs if the original optimization problem itself is feasible. Finally, for systems with unknown dynamics, adaptive affine control dynamics are proposed to approximate the real unmodelled system dynamics which are updated based on the error states obtained by real-time sensor measurements. A set of events required to trigger a solution of the QP in order to guarantee safety is defined, and a condition that guarantees the satisfaction of the HOCBF constraint between events is derived. In order to address the myopic nature of the CBF method, a real-time control framework that combines optimal trajectories and the computationally efficient HOCBF method providing safety guarantees is also proposed. The HOCBFs and CLFs are used to account for constraints with arbitrary relative degrees and to track the optimal state, respectively. Eventually, an optimal control problem based on the proposed framework is always reduced to a sequence of QPs regardless of the formulation of the original cost function. Another contribution of the dissertation is to apply the above proposed methods to solve complex safety-critical optimal control problems, such as those arising in rule-based autonomous driving and optimal traffic merging control for Connected and Automated Vehicles (CAVs).

This document is intended to give an overview of the human subject study in a driving simulator that was conducted by the Chair of Traffic and Transportation Psychology (Verkehrspsychologie – VPSY) of the Technische Universität Dresden (TUD) to provide the Chair of Automotive Engineering (Lehrstuhl Kraftfahrzeugtechnik – LKT) of TUD with the necessary input for the validation of a stochastic traffic simulation, especially for the parameterization, consolidation, and validation of driver behaviour models. VPSY planned, conducted, and analysed a driving simulator study. The main purpose of the study was to analyse driving behaviour and gaze data at intersections in urban areas. Based on relevant literature, a simulated driving environment was created, in which a sample of drivers passed a variety of intersections. Considering different driver states, driving tasks, and traffic situations, the collected data provide detailed information about human gaze and driving behaviour when approaching and crossing intersections. The collected data was transferred to LKT for the development of the stochastic traffic simulation.

No description available.

付記する学位プログラム名: デザイン学大学院連携プログラム / 京都大学 / 新制・課程博士 / 博士(工学) / 甲第24607号 / 工博第5113号 / 新制||工||1978(附属図書館) / 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 小森 雅晴, 教授 泉井 一浩 / 学位規則第4条第1項該当 / Doctor of Philosophy (Engineering) / Kyoto University / DFAM

Autonomous vehicles will help society if they can easily support a broad range of driving environments, conditions, and vehicles. Achieving this requires reducing the complexity of the algorithmic system, easing the collection of training data, and verifying operation using real-world experiments. Our work addresses these issues by utilizing a reflexive neural network that translates images into steering and throttle commands. This network is trained using simulation data from Grand Theft Auto V~\cite{gtav}, which we augment to reduce the number of simulation hours driven. We then validate our work using a RC car system through numerous tests. Our system successfully drive 98 of 100 laps of a track with multiple road types and difficult turns; it also successfully avoids collisions with another vehicle in 90\% of the trials.

Due to the mission-critical nature of the autonomous driving application, underlying algorithms for scene understanding should be given special care during their development. Mostly, they should be designed with precise consideration of accuracy and run-time. Accuracy should be considered strictly which if compromised leads to faulty interpretation of the environment that may ultimately result in accidental scenarios. On the other hand, run-time holds an important position as the delayed understanding of the scene would hamper the real-time response of the vehicle which again leads to unforeseen accidental cases. These factors come as the functions of several factors such as the design and complexity of the algorithms, nature of the encountered objects or events in the environment, weather-induced effects, etc. In this work, several novel scene understanding algorithms in terms- of semantic segmentation are devised. First, a transfer learning technique is proposed in order to transfer the knowledge from the data-rich domain to a data-scarce off-road driving domain for semantic segmentation such that the learned information is efficiently transferred from one domain to another while reducing run-time and increasing the accuracy. Second, the performance of several segmentation algorithms is assessed under the easy-to-severe rainy condition and two methods for achieving the robustness are proposed. Third, a new method of eradicating the rain from the input images is proposed. Since autonomous vehicles are rich in sensors and each of them has the capability of representing different types of information, it is worth fusing the information from all the possible sensors. Forth, a fusion mechanism with a novel algorithm that facilitates the use of local and non-local attention in a cross-modal scenario with RGB camera images and lidar-based images for road detection using semantic segmentation is executed and validated for different driving scenarios. Fifth, a conceptually new method of off-road driving trail representation, called Traversability, is introduced. To establish the correlation between a vehicle’s capability and the level of difficulty of the driving trail, a new dataset called CaT (CAVS Traversability) is introduced. This dataset is very helpful for future research in several off-road driving applications including military purposes, robotic navigation, etc.