Thesis title: Artificial intelligence and liability for its actions The artificial intelligence has recently become a ubiquitous phenomenon with a potential to change the world as we know it. Therefore, this thesis is concerned with the topic of artificial intelligence, specifically with a connection to a civil-law liability for its actions. It is absolutely clear that there will be more and more events in the future where damage will occur due to actions of artificial intelligence. Thus, the primary goal of this thesis is the determination of the person liable for damage caused in such cases under Czech law. Further goals of this thesis are an analysis of the question how is the dawn of autonomous cars influencing the legal instrument of liability for the damage caused by the operation of a means of transport as well as an introduction and examination of a recent resolution of the European Parliament which is supposed to serve as a basis for a future legal framework addressing the artificial intelligence in the area of the European Union. First, the paper focuses on the definition of the term artificial intelligence from a technical and legal viewpoint. This analysis shows that the definition of this term is not a straightforward one from neither of these viewpoints which can have serious...

Context. Building autonomous racing car controllers is a growing field of computer science which has been receiving great attention lately. An approach named Artificial Potential Fields (APF) is used widely as a path finding and obstacle avoidance approach in robotics and vehicle motion controlling systems. The use of APF results in a collision free path, it can also be used to achieve other goals such as overtaking and maneuverability. Objectives. The aim of this thesis is to build an autonomous racing car controller that can achieve good performance in terms of speed, time, and damage level. To fulfill our aim we need to achieve optimality in the controller choices because racing requires the highest possible performance. Also, we need to build the controller using algorithms that does not result in high computational overhead. Methods. We used Particle Swarm Optimization (PSO) in combination with APF to achieve optimal car controlling. The Open Racing Car Simulator (TORCS) was used as a testbed for the proposed controller, we have conducted two experiments with different configuration each time to test the performance of our APF- PSO controller. Results. The obtained results showed that using the APF-PSO controller resulted in good performance compared to top performing controllers. Also, the results showed that the use of PSO proved to enhance the performance compared to using APF only. High performance has been proven in the solo driving and in racing competitions, with the exception of an increased level of damage, however, the level of damage was not very high and did not result in a controller shut down. Conclusions. Based on the obtained results we have concluded that the use of PSO with APF results in high performance while taking low computational cost.

With self-driving cars on the horizon, vehicle autonomy and its problems is a hot topic. In this study we are using convolutional neural networks to make a robot car avoid obstacles. The robot car has a monocular camera, and our approach is to use the images taken by the camera as input, and then output a steering command. Using this method the car is to avoid any object in front of it. In order to lower the amount of training data we use models that are pretrained on ImageNet, a large image database containing millions of images. The model are then trained on our own dataset, which contains of images taken directly by the robot car while driving around. The images are then labeled with the steering command used while taking the image. While training we experiment with using different amounts of frozen layers. A frozen layer is a layer that has been pretrained on ImageNet, but are not trained on our dataset. The Xception, MobileNet and VGG16 architectures are tested and compared to each other. We find that a lower amount of frozen layer produces better results, and our best model, which used the Xception architecture, achieved 81.19% accuracy on our test set. During a qualitative test the car avoid collisions 78.57% of the time.

This thesis describes development of autonomous car for Carrera 143 racing track. Main objective of a car is to stop when traffic light shows red, or when there is an obstacle infront of a car. This paper also describes electric schemes used to control the car and their placement on the car. Algorithms developed for image processing are developed for processing unit Raspberry Pi Zero and are written in C/C++ programming language. OpenCV library is used for image processing. All source codes were developed in Microsoft Visual Studio 2019.

This thesis deals with the design and implementation of a self-driving model race car that is intended to attend an NXP Cup race. The work describes the selection of the platform, design process of algorithm, design process of printed circuit board for obstacle detection system, and various experiments with motion control.

Urbanization has been a growing trend in the past fifty years. Cities are now transforming into smart cities, spaces whose infrastructure comprises an embedded digital layer. Hardware collects real-time data in the urban environment and software elaborates it to improve all types of services, from traffic to waste management to well-being. One technology that is expected to use this digital layer to further change the urban environment is the autonomous car. The purpose of this dissertation is to explore what key design attributes future autonomous cars should possess if they have not only to co-exist with and be accepted by people in the landscape of tomorrow’s smart cities, but also what they should not possess in order not to cause any harm. In this sense, the dissertation recognizes calm technology to be necessary in the design of a future autonomous car to support a human-centered, as opposed to a car- or technology-centered, environment. A socio-technical and systemic lens is applied to the phenomenological investigation of nine companies carried out by means of twelve in-depth semi-structured interviews with experts working within the automotive sector, the smart city industry, and calm technology. Eight attributes (safety, on-demand, geo-tracking, sharing, multiple purposes, communication through smart devices, electrical care and IoT/connectedness) are identified as necessary for future autonomous cars to implement in order to take advantage of the smart city infrastructure and provide a human-centered experience. Additionally, six out of the eight calm technology principles recognized in literature are considered necessary when designing future autonomous cars.

This thesis deals with the design and implementation of a self-driving model race car that is intended to participate at NXP Cup race. First of all, the attention is given to the selection of a suitable platform, which will be used for the implementation of a motion control algorithm that enables autonomous navigation of the autonomous model race car along a specified track. Then, the details of motion control algorithm design and its implementation aspects are explained.  Also, the design of an obstacle detection system is presented. Finally, the work is concluded with presentation of various experiments with the implemented self-driving control algorithm and its evaluation.

博士(心理学) / Doctor of Philosophy in Psychology / 同志社大学 / Doshisha University

資訊科技(Information Technology, IT)對我們的生活與企業帶來極大的影響與改變。在企業中，資訊科技經常扮演不同的角色，這些不同的資訊科技角色(IT Role)可以自動化企業流程、支援決策制定、整合資源，甚至實現轉型與創新，對於企業的決策帶來不同層面的影響。而我們從近年來新興的資訊科技─大數據與人工智慧技術中，發現了不同於過去的新資訊科技角色。為了近一步了解這個新角色，本研究選擇人工智慧應用之一的自動駕駛車作為研究案例。本研究目的是探討自動駕駛車的資訊科技所扮演的新資訊科技角色；研究問題包含 (1) 自動駕駛車的資訊科技如何影響駕駛決策制定 (2) 在決策制定過程中，人與資訊科技分別扮演何種角色與職責。 本研究採用多個案研究法，分為兩個階段。首先，為解構資訊科技的決策制定流程，本研究依據決策理論與系統理論建構一研究架構。於文獻探討的章節中，本研究根據過往文獻與案例，提出四種企業常見的資訊科技角色─Automation、Supporter、Mentor與Enabler，並將研究架構應用於以上資訊科技角色以進行調整與驗證。接著，本研究選擇Google (Waymo)與Tesla作為自動駕駛車的研究個案，並將研究架構套用於兩個個案研發的自動駕駛車。由於不同的自動駕駛車研發理念與實現方式，Google與Tesla自動駕駛車的資訊科技分別扮演兩種不同的資訊科技角色─Autonomer與Smart Automation，本研究進一步比較所有資訊科技角色的研究架構結果，了解資訊科技角色的特性、影響與適用的決策類型。 自動駕駛的決策問題與環境與過去有極大的不同。為了實現安全的自動駕駛，資訊科技需要的資料類型更加多元，除了傳統數位類型資料，也需要收集周遭環境的3D影像等資料；另外，由於決策從過去的靜態問題轉移到動態與快速變化、擁有爆炸性資料與資訊的環境中，資訊科技需要更多的應變能力以制定更即時與適當的決策。由於資料、決策問題與環境的改變，企業對於資訊科技能力的需求也隨之改變，從自動駕駛車的個案中，本研究發現原本的資訊科技角色(Automation、Supporter、Mentor、Enabler)並不具備能應對如此動態與快速變化的決策問題與環境的能力，因而根據個案提出有能力實現動態即時決策制定的兩種新資訊科技角色。 使用人工智慧技術的Google無人駕駛車扮演著Autonomer的角色。資訊科技角色Autonomer能夠與外界進行互動，並且能夠不斷地追蹤、反饋與修正以實現自我成長；此外，面對各種駕駛決策情境，也能夠在無人為干預的情況下獨力完成駕駛決策的制定。資訊科技的學習能力是面對未知與難以預測的問題的最大優勢，而Autonomer的自我學習與決策制定能力也是與其他資訊科技角色最大的不同之處。使用大數據技術的Tesla自動駕駛車的Autopilot系統扮演著Smart Automation。資訊科技角色Smart Automation擁有更進步的資料收集與分析能力，能夠在動態與快速變化的環境中處理更為複雜的決策問題；此外，面對各種駕駛決策情境，Autopilot系統能在駕駛人保持監督的條件下進行自動駕駛以駕駛輔助的方式減輕駕駛人的負擔。最後，我們發現對於決策制定，資訊科技不僅能扮演一個完全獨立的角色，也能夠扮演一個與人互補的角色。大部分的人工智慧如同Google無人駕駛車做為一個Autonomer的角色，但同時更多企業目前使用的資訊科技屬於Supporter、Mentor與Smart Automation以支援或強化決策者的能力。 本研究探討在自動駕駛過程中不同資訊科技角色如何影響決策制定，以及駕駛人與資訊科技的角色與職責。並且從決策類型與資訊科技能力的角度，協助決策者與使用者全面地了解每個資訊科技角色的特性與適用的決策類型。此外，科技不斷在進步，本研究也提供一個了解各種資訊科技角色的基石，透過本研究的研究架構與方法，協助企業與決策者了解不同資訊科技對於決策的影響，本研究結果也能延伸應用於其他自動化、大數據與人工智慧相關領域，如無人工廠、吾人航空載具、工業4.0與金融科技(Fintech)。 / Information technology (IT) has brought great changes to people and business. In various applications, IT plays diverse roles that can automate business processes, support decision-making, integrate resources, and enable transformation and innovation and brings the impacts on different aspect of decision-making in enterprises. However, with the emerging technology of big data and artificial intelligence (AI), there is a new role for IT. To understand this role, we chose the autonomous car, an application of AI, as a study case. The objective of the research is to understand the new roles played by IT in the autonomous car. We focused on two questions: (1) how IT impacts decision-making in the autonomous car; and (2) what roles do IT and humans play during the decision-making process. This study applies a multiple case study in two phases. First, we built a conceptual framework, based on decision theory and system theory, to deconstruct the decision process of IT. To adjust and verify the framework, we applied it to actual cases and proposed IT roles of Automation, Supporter, Mentor and Enabler. Second, we applied the framework to the chosen autonomous car case studies, Google (Waymo) and Tesla, to explore the new role of IT in the autonomous car. Because of the different philosophies, there were two distinct roles played by IT in Google and Tesla’s autonomous cars, Autonomer and Smart Automation, respectively. We furthermore compared the frameworks of Google and Tesla, as well as the existing and new IT roles, explained the differences regarding the IT roles and decision types, and found out the applicable decision-making type of each IT roles.. Compared to the past, there were the great differences for the decision problems and environment of autonomous driving. To realize the safe autonomous driving, the data IT required became more diverse including non-text or non-digit data; besides, the decision-making also changed from static decision problems into dynamic and rapid decision environment with the explosive data and information that IT required more resilience to make decision. Due to the changes of the data, decision problems and environment, the demand for IT capability also changed. From the cases of the autonomous car, we found the original roles including Automation, Supporter, Mentor and Enabler was not enough – they did not possess the capability to make the dynamic and instantaneous decision. Therefore, we proposed two new IT roles – Smart Automation and Autonomer in this research that these two new IT roles which were applicable to the dynamic and instantaneous decision-making. The computer of the Google driverless car using AI technology acted as an Autonomer that was responsible for interacting with the surroundings and being self-growing with continuous tracking and adjustment; furthermore, under driving decision circumstances, this computer could assume the entire decision-making process without human intervention. The self-learning and decision-making ability of Autonomer is the characteristic most different from other IT roles; additionally, the learning ability was the greatest strength for dealing with unknown and unpredictable circumstances. The Autopilot system of the Tesla self-driving car, leveraging big data technology, acted as a Smart Automation that could process more complex decision problems in the dynamic environment with the advancement of data collection and analysis ability; furthermore, under the driving decision circumstances, the Autopilot system of the Tesla self-driving car could temporarily take over the driving control to decrease the driving burden and provide assistance to make driving easier. According to the research results, IT can not only play a totally independent role but also a complementary role. Most AI played the same IT role – Autonomer, such as the computer of the Google driverless car; meanwhile, much of the IT introduced by businesses acted as Supporter, Mentor and Smart Automation to assist and complement humans. This research provided a perspective for identifying how the different IT roles impact decision-making while driving an autonomous car and clarify the responsibility of humans and IT in the driving experience; moreover, from the perspective of decision problems and IT ability, it also provided a comprehensive and general understanding for realizing the characteristics of diverse IT roles and the applicable decision problems.

La génération de chemins lisses pour les voitures intelligentes est l’une des conditions les plus importantes pour faire accepter et faciliter la navigation autonome de ces véhicules. Cette thèse propose plusieurs méthodes de génération de chemins lisses pour les véhicules non-holonomes qui permet une continuité intrinsèque de la courbure de navigation et offre par ailleurs une flexibilité accrue pour diverses conditions aux limites. Le chemin de courbure continue est construit en composant plusieurs clothoids, comprenant notamment des segments de lignes et/ou d’arcs, et où chaque clothoid est obtenue par une régulation appropriée de ses paramètres. À partir de ces propriétés, le chemin obtenu est nommé pCCP (parametric Continuous Curvature Path). Le pCCP fournit un diagramme de courbure qui facilite une commande en orientation du véhicule, ce qui permet d'obtenir une évolution lisse de sa trajectoire. Le problème du pCCP local est défini par des configurations initiales et finales (caractérisées pour chacune par une posture et un angle de braquage). Le problème a été étendu pour être aussi général que possible en incluant plusieurs cas. La génération locale de pCCPs, pour des cibles statiques, est spécifiquement décrite, les problèmes ont été divisés en trois problèmes et chaque problème a été décomposé par la suite en plusieurs sous-classes possibles. Pour avoir une flexibilité importante des pCCPs proposés, des cibles dynamiques ont été considérées, obtenant ainsi le dynamic-pCCP (d-pCCP). Un cadre simple mais efficace pour analyser l'état futur de l'évitement des obstacles est appliqué en configuration 4D (3D avec l’ajout d’un axe temporel) en mettant en exergue deux manoeuvres d’évitement possibles, car les évolutions avant et arrière sont appliquées et validées avec plusieurs exemples. Selon une méthodologie similaire pour atteindre les critères de performance liés à la génération des pCCPs, le h-CCP (pour human-pCCP) est proposé en utilisant des modèles expérimentaux comportementaux d’échantillons de conducteurs humains. À partir de quelques sous-expériences, le modèle de conduite humain pour l’évitement d’obstacles, les changements de voie et les mouvements en virage sont extraits et ces modèles ont été inclus pour créer ainsi le h-CCP (obtenu d’une manière similaire au pCCP mais avec différents critères d’optimisation) qui permet d’améliorer considérablement le confort des passagers. / Smooth path generation for car-like vehicles is one of the most important requisite to facilitate the broadcast use of autonomous navigation. This thesis proposes a smooth path generation method for nonholonomic vehicles which has inherently continuity of curvature and having important flexibility for various boundary conditions. The continuous curvature path is constructed by composing multiple clothoids including lines and/or arc segments, and where each clothoid is obtained by parameter regulation. From those properties the path is named pCCP (parametric Continuous Curvature Path) and provides curvature diagram which facilitates a smooth steering control for path following problem. Local pCCP problem is defined by initial and final tuple configurations (vehicles posture and steering angle). The problem is expanded to be as general as possible by including several cases. The local pCCP generation for steady target pose is specifically described, where the problem is divided into three problems and each problem is also decomposed into several sub-cases. To give more flexibility to the proposed pCCP, dynamic target is considered to obtain dynamic-pCCP (d-CCP). A simple but efficient framework to analyze the future status of obstacle avoidance is applied in 4D (3D with the addition of time axis) configuration and two avoidance maneuvers as front and rear avoidance are applied and validated with several examples. Under the similar methodology in performance criteria of pCCP generation, the human-CCP (h-CCP) is derived from experimental patterns of human driver samples. From several subexperiments, human driving pattern for obstacle avoidance, lane change and cornering motion are extracted and those pattern were included to make the h-CCP (which is obtained with similar way as pCCP but with different optimization criteria) to enhance considerably the passenger comfort.