Optimizing AI at the Edge: from network topology design to MCU deployment

The first topic analyzed in the thesis will be Neural Architecture Search (NAS). I will focus on two different tools that I developed, one to optimize the architecture of Temporal Convolutional Networks (TCNs), a convolutional model for time-series processing that has recently emerged, and one to optimize the data precision of tensors inside CNNs. The first NAS proposed explicitly targets the optimization of the most peculiar architectural parameters of TCNs, namely dilation, receptive field, and the number of features in each layer. Note that this is the first NAS that explicitly targets these networks. The second NAS proposed instead focuses on finding the most efficient data format for a target CNN, with the granularity of the layer filter. Note that applying these two NASes in sequence allows an "application designer" to minimize the structure of the neural network employed, minimizing the number of operations or the memory usage of the network. After that, the second topic described is the optimization of neural network deployment on edge devices. Importantly, exploiting edge platforms' scarce resources is critical for NN efficient execution on MCUs. To do so, I will introduce DORY (Deployment Oriented to memoRY) -- an automatic tool to deploy CNNs on low-cost MCUs. DORY, in different steps, can manage different levels of memory inside the MCU automatically, offload the computation workload (i.e., the different layers of a neural network) to dedicated hardware accelerators, and automatically generates ANSI C code that orchestrates off- and on-chip transfers with the computation phases. On top of this, I will introduce two optimized computation libraries that DORY can exploit to deploy TCNs and Transformers on edge efficiently. I conclude the thesis with two different applications on bio-signal analysis, i.e., heart rate tracking and sEMG-based gesture recognition.

1° level of automation: the effectiveness of adaptive cruise control on driving and visual behaviour

The research activities have allowed the analysis of the driver assistance systems, called Advanced Driver Assistance Systems (ADAS) in relation to road safety. The study is structured according to several evaluation steps, related to definite on-site tests that have been carried out with different samples of users, according to their driving experience with the ACC. The evaluation steps concern: •The testing mode and the choice of suitable instrumentation to detect the driver’s behaviour in relation to the ACC. •The analysis modes and outputs to be obtained, i.e.: - Distribution of attention and inattention; - Mental workload; - The Perception-Reaction Time (PRT), the Time To Collision (TTC) and the Time Headway (TH). The main purpose is to assess the interaction between vehicle drivers and ADAS, highlighting the inattention and variation of the workloads they induce regarding the driving task. The research project considered the use of a system for monitoring visual behavior (ASL Mobile Eye-XG - ME), a powerful GPS that allowed to record the kinematic data of the vehicle (Racelogic Video V-BOX) and a tool for reading brain activity (Electroencephalographic System - EEG). Just during the analytical phase, a second and important research objective was born: the creation of a graphical interface that would allow exceeding the frame count limit, making faster and more effective the labeling of the driver’s points of view. The results show a complete and exhaustive picture of the vehicle-driver interaction. It has been possible to highlight the main sources of criticalities related to the user and the vehicle, in order to concretely reduce the accident rate. In addition, the use of mathematical-computational methodologies for the analysis of experimental data has allowed the optimization and verification of analytical processes with neural networks that have made an effective comparison between the manual and automatic methodology.

The future of dubbing: an overview of the new technologies

Throughout the years, technology has had an undeniable impact on the AVT field. It has revolutionized the way audiovisual content is consumed by allowing audiences to easily access it at any time and on any device. Especially after the introduction of OTT streaming platforms such as Netflix, Amazon Prime Video, Disney+, Apple TV+, and HBO Max, which offer a vast catalog of national and international products, the consumption of audiovisual products has been on a constant rise and, consequently, the demand for localized content too. In turn, the AVT industry resorts to new technologies and practices to handle the ever-growing workload and the faster turnaround times. Due to the numerous implications that it has on the industry, technological advancement can be considered an area of research of particular interest for the AVT studies. However, in the case of dubbing, research and discussion regarding the topic is lagging behind because of the more limited impact that technology has had on the very conservative dubbing industry. Therefore, the aim of the dissertation is to offer an overview of some of the latest technological innovations and practices that have already been implemented (i.e. cloud dubbing and DeepDub technology) or that are still under development and research (i.e. automatic speech recognition and respeaking, machine translation and post-editing, audio-based and visual-based dubbing techniques, text-based editing of talking-head videos, and automatic dubbing), and respectively discuss their reception by the industry professionals, and make assumptions about their future implementation in the dubbing field.

Augmented Reality for Airport Control Tower: Analysis and Implementation of Safety Nets

With the development of new technologies, Air Traffic Control, in the nearby of the airport, switched from a purely visual control to the use of radar, sensors and so on. As the industry is switching to the so-called Industry 4.0, also in this frame, it would be possible to implement some of the new tools that can facilitate the work of Air Traffic Controllers. The European Union proposed an innovative project to help the digitalization of the European Sky by means of the Single European Sky ATM Research (SESAR) program, which is the foundation on which the Single European Sky (SES) is based, in order to improve the already existing technologies to transform Air Traffic Management in Europe. Within this frame, the Resilient Synthetic Vision for Advanced Control Tower Air Navigation Service Provision (RETINA) project, which saw the light in 2016, studied the possibility to apply new tools within the conventional control tower to reduce the air traffic controller workload, thanks to the improvements in the augmented reality technologies. After the validation of RETINA, the Digital Technologies for Tower (DTT) project was established and the solution proposed by the University of Bologna aimed, among other things, to introduce Safety Nets in a Head-Up visualization. The aim of this thesis is to analyze the Safety Nets in use within the control tower and, by developing a working concept, implement them in a Head-Up view to be tested by Air Traffic Control Operators (ATCOs). The results, coming from the technical test, show that this concept is working and it could be leading to a future implementation in a real environment, as it improves the air traffic controller working conditions also when low visibility conditions apply.

Fleet Management System for Search and Rescue Operations

Over one million people lost their lives in the last twenty years from natural disasters like wildfires, earthquakes and man-made disasters. In such scenarios the usage of a fleet of robots aims at the parallelization of the workload and thus increasing speed and capabilities to complete time sensitive missions. This work focuses on the development of a dynamic fleet management system, which consists in the management of multiple agents cooperating in order to accomplish tasks. We presented a Mixed Integer Programming problem for the management and planning of mission’s tasks. The problem was solved using both an exact and a heuristic approach. The latter is based on the idea of solving iteratively smaller instances of the complete problem. Alongside, a fast and efficient algorithm for estimation of travel times between tasks is proposed. Experimental results demonstrate that the proposed heuristic approach is able to generate quality solutions, within specific time limits, compared to the exact one.