Il presente lavoro nasce con l’obiettivo di indagare le possibili ripercussioni penali derivanti dallo sviluppo dell’intelligenza artificiale e più nello specifico dallo sviluppo delle auto a guida autonoma. In particolare, nel momento in cui si sta scrivendo questo lavoro, le auto a guida autonoma stanno già circolando per le strade di tutto il mondo, pur essendone autorizzata la circolazione, ad oggi, solo per le finalità di test scientifici. In breve tempo, tuttavia, tali veicoli potranno essere venduti ed utilizzati; questa circostanza impone una ricerca scientifica che indaghi quali potranno essere i più rilevanti problemi per il diritto penale; problemi che il legislatore, prima, ed i giudici, poi, si troveranno ad affrontare. Premessa tale finalità, la ricerca, metodologicamente, deve seguire necessariamente più direzioni. A fronte di una prima ricostruzione del fenomeno dell’intelligenza artificiale e delle auto a guida autonoma, ove si farà riferimento ai principali testi “normativi” che si sono finora interessati al tema, sarà necessario soffermarsi su due diverse “macro” aree della responsabilità penale: la responsabilità individuale del conducente dei veicoli a guida autonoma (fintanto che ci sarà la necessità di un conducente) e la responsabilità derivante dalla produzione dei veicoli. Medio tempore, tuttavia, fil rouge del lavoro sarà anche quello di comprendere se, per rispondere ai problemi che questo nuovo settore porrà, sia davvero necessario o consigliato l’utilizzo dello strumento penale o se, invece, non sia preferibile l’implementazione di strumenti giuridici diversi.

The mining industry uses vehicles for a wide range of applications, including excavation and transportation of rock and soil. Currently, this requires a lot of human labour, mainly drivers, but efforts are being made to increase automation, e.g. using autonomous vehicles. In order for a vehicle to reach any level of autonomy, it needs to be aware of its surroundings, for instance by using sensors. The placement of the sensors is a difficult problem. The goal of this project was to create a tool that would simplify the sensor placement process. The tool should simulate sensors on autonomous vehicles, both by visualizing their field of view and by generating synthetic data. The tool was created, including simulation environments, models of different types of sensors and tools to analyze the results of the simulation. Both the field of view visualization and the data analysis tools were shown to be powerful tools for evaluating sensor placements. All of the sensor models are able to generate data, with different levels of realism. The radar model and the camera model give a good estimation of what the sensors can detect, while the lidar model is capable of generating realistic data. / Gruvindustrin använder fordon till ett stort antal olika uppgifter, bland annat till att gräva ut och förflytta sten och jord. Detta kräver för tillfället mycket manuellt arbete, framförallt förare, men försök att automatisera delar av arbetet utförs, till exempel genom att använda autonoma fordon. För att ett fordon ska kunna bli autonomt krävs det att det kan känna av sin omgivning, exempelvis genom att använda sensorer. Sensorplacering är ett svårt problem. Målet med projektet var att skapa ett verktyg för att förenkla sensorplaceringsprocessen. Verktyget skulle simulera sensorer på autonoma fordon, både genom att visualisera deras synfält och genom att generera syntetisk data. Verktyget skapades, inklusive simuleringsmiljöer, modeller av olika typer av sensorer, och verktyg för att analysera genererad data. Både synfältsvisualiseringen och datagenereringen visades vara användbara verktyg för att utvärdera sensorplaceringar. Alla sensormodellerna kan generera data, med olika realistiska resultat. Radarmodellen och kameramodellen ger en bra uppskattning av vad sensorerna kan detektera, medan lidarmodellen kan generera realistisk data.

Automatic detection of weeds could be used for more efficient weed control in agriculture. In this master thesis, weed detectors have been trained and examined on data collected by RISE to investigate whether an accurate weed detector could be trained on the collected data. When only using annotations of the weed class Creeping thistle for training and evaluation, a detector achieved a mAP of 0.33. When using four classes of weed, a detector was trained with a mAP of 0.07. The performance was worse than in a previous study also dealing with weed detection. Hypotheses for why the performance was lacking were examined. Experiments indicated that the problem could not fully be explained by the model being underfitted, nor by the object’s backgrounds being too similar to the foreground, nor by the quality of the annotations being too low. The performance was better when training the model with as much data as possible than when only selected segments of the data were used.

The research analyzes the possibile development of road traffic liability, product liability and car insurance in the face of the progressive automation of the global vehicle fleet. Moving from a historical overview of the rules currently governing road traffic liability on the European continent, the research shows how this subject has developed over the decades, looking for a difficult balance between effective protection of injured parties and economic efficiency. As a result, road traffic liability regimes have some unique characteristics, which are expected to remain relevant even when the global vehicle fleet is composed of highly automated vehicles. The challenges posed by automation change with the degrees of technological advancement. By seeking to enhance the dialogue with technology, the research shows that there are realistic evolutions and utopian prospects, and that the excessive focus on the latter has contributed to an underestimation of the impact that more basic forms of autonomy have on road traffic. In fact, the Advanced Driver Assistance Systems (ADAS) – although they do not require amendments to existing civil liability systems – have so far been developed by requiring humans to adapt to the machine, and not the other way around. In contrast, safety needs would dictate that the software of the future be designed following a human-centered philosophy. More advanced levels of automation raise question about whether current traffic liability regimes should be amended, especially given the fact that the vehicle owner is no longer in the best position to manage risks. Preliminarily highlighting the need for a liberal and confident approach toward innovation, hypotheses for regulation that have emerged in the debate have been analyzed: given the difficulty of finding a clearly winning compromise between protecting third parties and incentivizing innovation, it is believed that the key lies in a simple, technically aware, sector-specific regulation that is adaptable to the multiple mobility scenarios of the future. The soft-law can be a useful tool for managing the technology's market debut, while waiting for evidence to suggest in which direction the technology will evolve and – consequently – the most appropriate civil liability framework.

No description available.

The utilization of image analysis and object detection spans various industries, serving purposes such as anomaly detection, automated workflows, and monitoring tool wear and tear. This thesis addresses the challenge of achieving precise robot navigation between fixed start and end points by combining GPS and image analysis. The underlying motivation for tackling this issue lies in facilitating the creation of immersive videos, mainly aimed at individuals with disabilities, enabling them to virtually explore diverse locations through a compilation of shorter video clips.  The research delves into diverse models for object detection frameworks and tools, including NVIDIA Detectnet, and YOLOv5. Through a comprehensive evaluation of their performance and accuracy, the thesis proceeds to implement a prototype system utilizing an Elegoo Smart Robot Car, a camera, a GPS module, and an embedded NVIDIA Jetson Nano system.  Performance metrics such as precision, recall, and map are employed to assess the models' effectiveness. The findings indicate that the system demonstrates high accuracy and speed in detection, exhibiting robustness across varying lighting conditions and camera settings

The proposed master's thesis investigates the adaptability and efficacy of self-supervised representation learning (SSL) in medical image analysis, focusing on Mammographic Imaging to develop robust representation learning models. This research will build upon existing studies in Mammographic Imaging that have utilized contrastive learning and knowledge distillation-based self-supervised methods, focusing on SimCLR (Chen et al 2020) and SimSiam (Chen et al 2020) and evaluate approaches to increase the classification performance in a transfer learning setting. The thesis will critically evaluate and integrate recent advancements in these SSL paradigms (Chhipa 2023, chapter 2), and incorporating additional SSL approaches. The core objective is to enhance robust generalization and label efficiency in medical imaging analysis, contributing to the broader field of AI-driven diagnostic methodologies. The proposed master's thesis will not only extend the current understanding of SSL in medical imaging but also aims to provide actionable insights that could be instrumental in enhancing breast cancer detection methodologies, thereby contributing significantly to the field of medical imaging and cancer research.

This thesis explores the potential of Neural Radiance Fields (NeRF) for generating orthophotos in forestry applications. Traditional orthophoto production methods, such as those implemented in Pix4D, require high image overlap and significant data collection. NeRF, a novel 3D scene reconstruction technique, shows potential for reducing these requirements by effectively reconstructing scenes with lower image overlaps. This study compares the orthophotos produced by NeRF and Pix4D using various degrees of image overlap, evaluating the results based on geometric accuracy, image quality, and robustness to data variations. The findings indicate that NeRF can produce orthophotos from low-overlap images with geometric accuracy comparable to orthophotos produced by Pix4D from high-overlap images, though with some trade-offs in image sharpness. These results suggest potential cost savings and operational efficiencies in forestry applications, providing a viable alternative to traditional photogrammetric techniques.

The global population is growing, and more people than before are moving to cities. This creates a need for increased building efficiency and possibility to work in remote environments. On today’s construction sites, there is a need to able to organize the site in a better way. In the future, autonomous vehicles will instead find it difficult to localize materials on a construction site. The autonomous vehicles can localize themselves with cameras and sensors, but they do not know how to localize the materials and items. This report is based on a project where Volvo Construction Equipment acted as a customer and the project was performed by students from Blekinge Institute of Technology and Stanford University. The prompt for this project was “From elephants to ants – from Earth to Mars” and would later be interpreted as finding a solution for the future that will be able to function without human’s intervention. From this project, this report was created. The following research questions for this report were: • How can workers locate building materials on today’s construction sites? • How will autonomous vehicles be able to locate material without human assistance in future construction sites? To solve these problems a design-process started, using an engineering design method. This method was chosen because of the type of problem. In engineering, the problem is identified to create a solution to the problem, comparing to when studying science, a question should be answered. The outcome from this report is a Product Service System (PSS) for a tracking system and a device for materials on today’s and future construction sites. When this solution was created no economic aspects were considered. Also, the focus of this report is the first steps of going from today’s construction sites to the future construction sites where autonomous vehicles will be used. The result from this research shows that the same problem of organizing a construction site is a pattern that can be seen in the majority of the sites that were visited during field works. Also, the workers today have little trust in the autonomous vehicles which is a result of lacking information and communication within companies. Furthermore, to be able to move to an autonomous future the mindset and attitude has to be changed. The collected data was analysed, and the outcome was a tracing system that will enable, both humans and machines, to localize materials on today’s and future construction sites. With this solution, today’s workers can track their materials wherever it is placed, without any need of changing the site. The autonomous vehicles will be able to use the tags to localize materials when there are no humans around. / Den globala befolkningen växer och fler flyttar till städerna än tidigare. Detta skapar ett behov av ökad effektivitet i byggbranschen och möjlighet till arbete i avlägsna miljöer. På dagens byggarbetsplatser är det nödvändigt att kunna organisera platsen på ett bättre sätt. I framtiden kommer de autonoma fordonen få det svårare att lokalisera material på en byggarbetsplats. De autonoma fordonen kan lokalisera sig med kameror och sensorer, men de vet inte hur man lokaliserar material och föremål. Rapporten bygger på ett projekt där kunden var Volvo Construction Equipment och projektet utfördes av studenter från Blekinge Tekniska Högskola och Stanford University. Prompten för projektet löd "Från elefanter till myror - från jorden till mars" och som senare tolkades till att finna en lösning för framtiden som kommer att kunna fungera utan mänsklig påverkan. Från detta projekt skapades denna rapport. Följande forskningsfrågor skulle besvaras: • Hur kan arbetare lokalisera byggmaterial på dagens byggarbetsplatser? • Hur kommer autonoma fordon kunna lokalisera material utan mänsklig hjälp på de framtida byggarbetsplatserna? För att lösa dessa problem startades en designprocess, med vald ingenjörsmetod. Denna metod valdes på grund av typen av problem. I ingenjörsmetoden identifieras problemet för att skapa en lösning till problemet, jämfört men en vetenskaplig metod, där en fråga besvaras. Resultatet från denna rapport är ett produkttjänstesystem (PSS) för ett spårningssystem för att kunna spåra material på dagens och framtida byggarbetsplatser. När denna lösning skapades togs det ingen hänsyn till de ekonomiska aspekterna. Fokus på denna rapport är de första stegen för att gå från dagens byggarbetsplatser mot de framtida byggplatserna där autonomiska fordon kommer att användas. Resultatet av forskningen visade att det finns ett stort behov av att organisera de olika byggarbetsplatserna som besöktes under studiebesöken. Arbetarna har idag ett litet förtroende för de autonoma fordonen som är ett resultat av bristande information och kommunikation inom företagen. För att kunna gå till en autonom framtid måste tankesätt och attityd ändras. Den samlade data analyserades och resultatet var ett spårningssystem som gör det möjligt för både människor och maskiner att lokalisera material på dagens och framtida byggarbetsplatser. Med denna lösning kan dagens arbetare enkelt spåra materialet, utan att behöva omstrukturera arbetsplatsen. De autonoma fordonen kommer kunna använda spårningssystem för att kunna lokalisera material när det inte finns några människor till hands.

In the field of autonomous driving a common scenario is to apply deep learningmodels on camera feeds to provide information about the surroundings. A recenttrend is for such vision-based methods to be centralized, in that they fuse imagesfrom all cameras in one big model for a single comprehensive output. Designingand tuning such models is hard and time consuming, in both development andtraining. This thesis aims to reproduce the results of a paper about a centralizedvision-based model performing 3D object detection, called BEVDet. Additionalgoals are to ablate the technique of class balanced grouping and sampling usedin the model, to tune the model to improve generalization, and to change thedetection head of the model to a Transformer decoder-based head. The findings include a successful reproduction of the results of the paper,while adding depth supervision to BEVDet establishes a baseline for the subsequentexperiments. An increasing validation loss during most of the training indicatesthat there is room for improvement in the generalization of the model. Severaldifferent methods are tested in order to resolve the increasing validation loss,but they all fail to do so. The ablation study shows that the class balanced groupingis important for the performance of the chosen configuration of the model,while the class balanced sampling does not contribute significantly. Without extensivetuning the replacement head gives performance similar to the PETR, themodel that the head is adapted from, but fails to match the performance of thebaseline model. In addition, the model with the Transformer decoder-based headshows a converging validation loss, unlike the baseline model.