Advanced Driver Assistance Systems (ADAS) allow to reduce the number of vehicle accidents in the road, increase the efficiency of the logistics industry and revolutionize the urban transport. To develop further and better ADAS functionalities, there is a need to integrate LiDAR systems in the vehicles. Among the LiDAR subsystems, lie the laser subsystem, which is responsible for emitting the laser signals to map and comprehend the surrounding environment. This thesis explores the performance and suitability of three laser systems, for LiDAR applications within autonomous vehicles. The systems under investigation include a Vertical-Cavity Surface-Emitting Laser array (VCSEL), an Edge Emitting Laser (EEL), and a compact LiDAR system prototype. The research studies the advantages and limitations of these laser systems in the context of LiDAR technology. Measurement results highlighted the VCSEL array superior optical power range, efficiency, and spectral stability, making it ideal for Flash LiDAR applications. In contrast, the SPL PL90 EEL system, while slightly less efficient, showed promise for MEMS and mechanical LiDARs due to its appropriate optical power range and laser emission characteristics. The outcome not only sheds light on the laser systems’ contributions to LiDAR technology, but also assists for informed decision-making when integrating LiDAR in autonomous vehicles.

With rapid urbanization and a sharp increase in vehicular density, cities face several mobility-related challenges such as traffic congestion, high accident rates, and environmental impact from vehicular emissions. Autonomous driving technologies have the potential to mitigate these problems by improving traffic safety and efficiency. However, to fully realize these benefits, autonomous vehicles must effectively coordinate their movements and communicate seamlessly. This Dissertation addresses the research and implementation of the Maneuver Cooperation Service (MCS), enabling real-time, coordinated vehicle maneuvers through Vehicle-to-Everything communication. The MCS architecture, based on standards established by the European Telecommunications Standards Institute, employs a modular and multi-technology design to support diverse communication standards for efficient and reliable inter-vehicle communication. The MCS contributes meaningfully to Cooperative Intelligent Transportation Systems by reducing maneuver uncertainty and enabling secure, real-time decision-making for autonomous vehicles. To increase system robustness, an information recovery mechanism using acknowledgments was designed, implemented and evaluated. The system’s effectiveness, validated through emulation, showed its ability to support safety and synchronized maneuvers across various scenarios. The results demonstrate that the implemented MCS and the proposed acknowledgment mechanism significantly enhances road safety and traffic flow by coordinating maneuvers efficiently, by considerably reducing the time needed to prepare the execution of a cooperative maneuver. This research supports smart city initiatives by presenting a scalable, cooperative solution for sustainable and safe urban mobility integration.

This thesis advances the field of end-to-end Autonomous Driving (AD) in urban environments, focusing primarily on Reinforcement Learning (RL) techniques to address existing limitations. The research begins with a comprehensive review of current end-to-end AD systems, highlighting the strengths and weaknesses of Imitation Learning (IL) and RL approaches, and identifying critical areas for improvement and promising avenues for future research. To address these challenges, we introduced RLAD, the first Reinforcement Learning from Pixels (RLfP) method for urban AD. Several techniques were introduced to enhance the performance of state-of-the-art methods. First, we developed an image encoder that utilizes both image augmentations and Adaptive Local Signal Mixing (A-LIX) layers. Additionally, we introduced WayConv1D, a waypoint encoder that captures the 2D geometrical information of waypoints using 1D convolutions. Furthermore, we designed an auxiliary loss function to emphasize the significance of traffic lights in the latent representation of the environment. RLAD demonstrated good performance on the NoCrash benchmark but required further integration of demonstrations to match state-of-the-art AD systems. Consequently, we developed RLfOLD (Reinforcement Learning from Online Demonstrations), which combines IL and RL by incorporating online demonstrations into RL training. We proposed a policy network that outputs two standard deviations, enabling adaptive control for exploration and IL training while considering uncertainty in both domains. Additionally, we incorporated an uncertainty-based technique guided by an online expert to enhance the exploration process. RLfOLD achieves state-of-the-art results on the NoCrash benchmark with enhanced efficiency and resource utilization. We further tackled the CARLA Leaderboard 2.0 benchmark’s challenges by developing PRIBOOT, an expert agent leveraging privileged information and limited human driving logs to navigate demanding scenarios. PRIBOOT’s novel techniques, such as the bird’s-eye view (BEV) representation and processing the BEV as an RGB image instead of a set of masks, significantly improved performance. This approach provided an expert capable of navigating this challenging benchmark, enabling researchers to generate extensive datasets and potentially resolving the data availability issues that have hindered progress. Collectively, our work presents significant contributions to the field of AD, offering insights and tools that pave the way for safer and more reliable autonomous systems in urban environments.

Advanced Driver Assistance Systems (ADAS) and Autonomous Driving (AD) use a variety of sensors that capture different kinds of data to build a representation of the world surrounding the vehicle. The integrated and contextual visualization of these data is essential to understand these systems and develop the algorithms which control them. The use of Virtual Reality (VR) for scientific visualization may bring benefits to ADAS/AD data visualization. This dissertation details the development of a web-based platform for ADAS/AD data visualization in collaboration with VORTEX-CoLab, based on an existing desktop platform. Using emerging web technologies to create VR experiences, such as libraries based on the WebXR API, the platform can be accessed from the browser of standalone VR devices, with no other installation requirements. The developed platform represents multiple data types simultaneously, including images from cameras, overlaying each other, which permits direct comparisons between them. User tests showed that although there are aspects to improve in the interaction with the platform, users considered the developed application to be easy to learn and use. However, employees of VORTEX-CoLab did not find that the platform would be particularly useful in their job. Not all employees who participated in the user tests deal with segmentation validation in their work, so the utility of the platform in a workplace setting is still in question. Other users found the platform interesting, so it may be a way to generate interest in ADAS in an amusing way.

Every 23 seconds, one more person dies in a traffic accident, with this being the leading cause of death for children and young adults. The main reason for traffic accidents is the human factor, with about 90% of all road casualties being rooted in human error. Autonomous Driving (AD) is, therefore, highly suggested as one of the most promising solutions to this problem. The ultimate goal would be to have a fully autonomous vehicle, without the need of any human intervention. If the vehicle is to take over the steering wheel, it must be able to perform all the tasks that the human was performing before, with one of the critical ones being the ability to reliably perceive its surroundings. LiDAR sensors have proven to have a key role in achieving such requirements, as they are the only type of sensor able to provide a detailed point cloud of the vehicle’s surroundings. LiDARs can be divided in two categories: full-waveform and TDC-based LiDARs. Contrary to TDC-based LiDARs, in full-waveform LiDARs, the full received pulse is captured, which enables precise range estimation, and discrimination of interfering or ghost pulses. However, fast and accurate ADCs are required, which, if resorting to conventional ADCs may result in a prohibitive Size, Weight, Power and Cost (SWaP-C). In this work, an All-Digital Analog-to-Digital Converter (AD-ADC) is custom-designed to enable a precise reconstruction of the LiDAR waveforms, while keeping SWaP-C low. An ideal model of the proposed ADC was first obtained, with a final AD-ADC model being constructed from this first one, by successively adding and analysing new impairments, including realistic time resolution and reference curve. The influence of finite slew rate and timing jitter was also addressed. All ADC models were applied to an experimental dataset of LiDAR waveforms, followed by a pre-selected range estimation method. In the end, a centimeter precision and accuracy were obtained over the majority of the waveforms. The results also suggest that the estimation performance may be further improved, by overcoming the proposed AD-ADC inherent drawback of performing non-uniform sampling, and by optimizing the considered reference curve. In conclusion, this study demonstrates that the proposed digital ADC may configure an advantageous part of an automotive LiDAR, being a viable option for the automotive industry.

The number of road accidents still remains tragically high, being estimated that human error is the main reason behind 90% of the accidents. As such, there is a rising interest in self-driving vehicles, as these shall eliminate human error and ensure safe driving. A self-driving vehicle first needs to perceive its surroundings to drive safely. This is achievable by using vision sensors. Among these sensors, LiDAR presents the unique advantage of acquiring a high-resolution 3D representation of the surroundings. Such rich 3D data, in the form of point clouds, enables accurate 3D object detection. The success obtained from the first and current LiDAR generation has motivated the development of a second-generation LiDAR, now resorting to coherent detection. Besides estimating the coordinates of each point, such a LiDAR also estimates the radial velocity. Therefore, it is expected that the use of the radial velocity enhances 3D object detection, as it helps differentiating moving objects from static background points. This work has the main objective of analysing whether LiDAR-based 3D object detection can be improved by considering the additional feature of radial velocity. As, to the best of the author’s knowledge, no LiDAR dataset including radial velocity information is yet available, the first step taken in this work was to generate a synthetic dataset. Three use cases were created to evaluate the impact of raw and processed radial velocity features, added to ideal and flawed range estimation. Considering the radial velocity, either raw or processed, did not introduce noteworthy benefits to the detection of objects with rich geometrical information. Conversely, for objects represented by few and possibly flawed points, which are typically more challenging to detect, this information was able to improve object detection. As a conclusion, this work suggests that a second-generation LiDAR may indeed contribute to enhanced road safety.

According to the World Health Organization, around 1.19 million people die every year as a result of road accidents caused essentially by human error. For this reason, advanced driver-assistance systems have been developed for several years, with the aim of increasing road safety. In order to perceive the surrounding environment, these systems rely on vision sensors such as cameras, RaDARs and LiDARs. Of all these sensors, LiDAR is the only one that can capture a high-resolution point cloud, enabling 3D object detection. However, a LiDAR is made up of high-precision components, which makes it expensive. One way to circumvent this barrier is to use active stereoscopic vision to build a LiDAR. As a result, the proposed LiDAR consists of affordable and mass-produced components such as cameras and lasers. However, this approach has its challenges, such as noise, correspondence, calibration and centroid estimation. This thesis focuses on noise. We propose an alternative LiDAR sensor based on active stereoscopic vision, featuring a dot projector capable of increasing its range. In this work, the difference in 3D object detection performance between an ideal LiDAR, a conventional active stereo rig and the proposed LiDAR is evaluated using the same set of synthetically generated ideal point clouds. The results obtained show that the extreme simplification assumed when building the ray optics model did not allow to prove that the proposed LiDAR is more accurate than conventional stereoscopic LiDAR when it comes to detecting 3D objects.

Currently, a wide range of research and development activities are taking place in the field of autonomous vehicles, and many companies and organizations are working on developing and testing vehicles with various levels of automation. Autonomous vehicles require multiple sensors and advanced control systems to perceive the surrounding environment and make decisions without resorting to a human driver. LiDAR (Light Detection and Ranging), highly utilized in the development of autonomous vehicles, stands out for its potential to provide long-range highresolution 3D vision. Utilizing lasers and photodetectors to measure the time it takes for light pulses to bounce back, LiDAR produces precise and accurate 3D point clouds of a vehicle’s surrounding environment. The development of a secondgeneration LiDAR, coherent LiDAR, promises better reliability and the capability to directly measure relative radial velocity alongside distance for each point. This direct velocity measurement has the potential to revolutionize perception tasks in autonomous vehicles, including Multiple Object Tracking (MOT), paving the way for significant advancements in 3D perception. The fundamental motivation of this thesis is to investigate whether LiDAR-based 3D MOT can be improved by taking advantage of the radial velocity in each point of the point cloud provided by a coherent LiDAR. Due to the novelty of this sensor and the lack of a publicly-available MOT dataset with velocity information, a synthetic dataset was generated with a modified version of PreSIL, displaying 19 training and 20 testing sequences, with over 16k frames and 80k annotations. By adapting a state-of-the-art MOT model, it is was possible to conclude that, with velocity information, this model yields more accurate object proposals, at the cost of worse trajectory associations. These results were influenced by limitations of the data generation code, such as the lack of inertial data, problematic car models and poor performance in high inclination driving scenarios. When only taking into account sequences where these problems are absent, the proposed solution obtains improved performance when compared to the original implementation, showing that the additional velocity information has a positive effect on MOT tasks. This work thus points out that a second-generation LiDAR may indeed improve the perception systems, that are critical to the development of safe autonomous vehicles.

This dissertation examines the impact of Vehicle-to-Everything (V2X) communications on vehicle cooperation, focusing on increasing the robustness and feasibility of Cooperative, Connected Autonomous Vehicles (CCAVs). V2X communications enable CCAVs to obtain a holistic environmental perception, facilitating informed decision-making regarding their trajectory. This technological innovation is essential to mitigate accidents resulting from inadequate or absent communication on the roads. As the importance of vehicle cooperation grows, the European Telecommunications Standards Institute (ETSI) has been standardizing messages and services for V2X communications, in order to improve the synchronization of CCAVs actions. In this context, this work explores the usage of Maneuver Coordination Messages (MCMs), under standardization by ETSI, for cooperative path planning. These messages, as well as the associated Maneuver Coordination Service (MCS), were developed in a V2X simulation framework combining both network communications and traffic simulations. The obtained results prove the effectiveness of using V2X communications to ensure the safety and efficiency of Cooperative Intelligent Transportation Systems (C-ITS).

Road traffic injuries are the leading cause of death for people aged 5-29 years old. Given that the main cause of these casualties is human error, if we could offset the driving tasks from driver to an autonomous vehicle, many lives would be saved. As such, the development of autonomous vehicles is rising in interest. Autonomous vehicles are equipped with vision sensors that allow for the perception of their surroundings. There are multiple types of vision sensors, such as cameras, RaDARs and LiDARs. Among these, LiDAR has the differentiating feature of capturing a high-resolution point cloud, thus enabling 3D object detection. However, LiDAR is still an immature and expensive sensor, being difficult to mass-produce. This work proposes an alternative LiDAR concept based on active stereoscopic vision and dot projection. The proposed LiDAR is comprised of mature massproduced components, thus being affordable, and overcoming the limitations of currently accessible LiDAR sensors. However, the proposed LiDAR has four challenges that need to be carefully analyzed: noise, correspondence, calibration and centroiding. This thesis focuses on the noise and calibration challenges. It aims to analyze how much impact the introduction of systematic error, introduced by miscalibration, and non-systematic error, introduced by camera noise, have on 3D object detection. In this work, the difference in performance in 3D object detection between an ideal LiDAR, a commercially available LiDAR and the proposed LiDAR is assessed drawing from a synthetic dataset comprised of ideal point clouds. The obtained results show that the differences between the three types of LiDAR are small, which means that calibration and noise do not put insurmountable limits to the performance of a LiDAR based on active stereo.