<p dir="ltr">Autonomous driving technology has rapidly advanced in recent years, leading to significant developments in its deployment and application. This paper presents the design and deployment of a real-world autonomous driving test platform with comprehensive capabilities, enabling the test and evaluation of autonomous driving technologies in real-world scenarios. The platform integrates multiple sensors, including LiDAR, radar, cameras, Global Navigation Satellite System (GNSS), and Inertial Measurement Unit (IMU), which collectively provide robust sensing, localization, and measurement capabilities. Built on the Autoware.AI framework, this test platform offers a flexible environment for diverse autonomous driving functionalities, including mapping, object detection, planning, and control. The use of ROS (Robot Operating System) enables seamless communication between various system modules, simplifies sensor integration, and provides extensive tools for debugging and visualization, making the platform highly adaptable for both research development and algorithm validation.</p><p dir="ltr">As a key demonstration of the platform’s capabilities, the paper introduces Talk2Drive, a Large Language Model (LLM)-based autonomous driving framework designed to enhance human-vehicle interaction. Talk2Drive leverages advanced AI techniques to interpret and execute verbal commands from the driver, enabling real-time adjustments to vehicle behavior and offering a personalized driving experience. This paper indicates the comprehensive integration and deployment process of the Talk2Drive framework. Additionally, through various experimental setups—including highway driving, intersections, and parking lot scenarios—the paper demonstrates how this autonomous driving platform evaluates the safety, performance, adaptability, and reliability of AI-driven frameworks like Talk2Drive under the real-world condition. The results underscore the platform's effectiveness in testing and validating autonomous systems. At the same time, the successful deployment of Talk2Drive also proves that the designed autonomous driving testing platform has the capabilities to examine the complex autonomous driving algorithm or systems' performance in real-world environments.</p>

Scenario-based testing will help to validate automated driving systems (ADS) and establish safer road traffic. To date, most data-driven test scenario generation methods rely primarily on one data source such as police accident data (PD), naturalistic driving studies, or video-based traffic observations (VOs). However, none of these data sources perfectly satisfies all the layers of the six-layer model for the description of test scenarios. Moreover, not all available data sources cover the same location and period of time. Therefore, we fused information from 1,648 scenarios extracted from a German VO with information from 74 scenarios extracted from German PD into a comprehensive new PD* database. Finally, PD* consisted of 74 accident scenarios extended, for example, by variables containing the dynamic information of the VO scenarios. Thus, PD* contained more than 350 variables, whereas PD contained only 269 and VO only 122 variables. For fusion, we followed the Find-Unify-Synthesize-Evaluation (FUSE) for Representativity (FUSE4Rep) process model using statistical matching. Subsequently, we derived three logical scenarios from PD* to test an autonomous emergency braking system (AEB) in a stochastic traffic simulation incorporating driver-behavior models. The quality of the fusion itself was satisfactory, and we propose improving the VO data collection process and observation time to obtain even better results.

Autonomous Driving Systems (ADS), a subcategory of Cyber-Physical Systems (CPS) are becoming increasingly popular with ubiquitous deployment. They provide advanced operational functions for perception and control, but this also raises the question of their safety capability. Such questions include if the vehicle can stay within its lane, keep a safe distance from the leading vehicle, or avoid obstacles, especially under the presence of uncertainties. In this master thesis, the operational safety of ADS will be addressed, more specifically on the Adaptive Cruise Control (ACC) system by modeling an optimal control problem based on Control Barrier Function (CBF) unified with Model Predictive Control (MPC). The corresponding optimal control problem is robust against measurement uncertainties for an Autonomous Vehicle (AV) driving on a highway, where the measurement uncertainties will represent the common faults in the perception system of the AV. A Kalman Filter (KF) is also added to the system to investigate the performance difference. The resulting framework is implemented and evaluated on a simulation scenario created in the open-source autonomous driving simulator CARLA. Simulations show that MPC-CBF is indeed robust against measurement uncertainties for well-selected horizon and slack variable values. The simulations also show that adding a KF improves the overall performance. The higher the horizon, the more confident the system becomes as the distance to the leading vehicle decreases. However, this may cause infeasibility where there are no solutions to the optimal control problem during sudden braking as the AV cannot brake fast enough before it crashes. Meanwhile, the smaller the slack variable, the more restrictive becomes CBF where it impacts more on the control input than desired which could also cause infeasibility. The results of this thesis will help to facilitate safety-critical CPS development to be deployed in real-world applications. / Autonoma körsystem (ADS), som är en del av cyberfysiska system (CPS), har blivit alltmer populär med allestädes närvarande användning. Det bidra med avancerade operativa funktioner för perception och styrning, men samtidig väcker detta också frågan om dess säkerhetsförmåga. Sådana frågor inkluderar om fordonet kan hålla sig inom sitt körfält, om det kan hålla ett säkert avstånd till det ledande fordonet eller om det kan undvika hinder, speciellt under osäkerheter hos systemet. I detta examensarbete kommer driftsäkerheten hos ADS att behandlas, mer specifik på adaptiv farthållare (ACC) genom att modellera ett optimalt kontrollproblem baserat på kontrollbarriärfunktion (CBF) förenat med modellförutsägande styrning (MPC). Motsvarande optimalt kontrollproblem är robust mot mätosäkerheter för ett autonomt fordon som kör på en motorväg, där mätosäkerheterna representerar vanliga fel i AV:s perceptionssystem. Ett Kalmanfilter (KF) läggs också till i systemet för att undersöka skillnaden i prestanda. Det resulterande ramverket implementeras och utvärderas på ett simuleringsscenario som skapats i den öppna källkodssimulatorn för autonom körning CARLA. Simulationer visar att MPC-CBF är robust mot mätosäkerheter för väl valda värden för horisont och slackvariabler. Det visar också att systemets prestanda förbättrats ännu mer om ett KF läggs till. Ju större horisont, desto mer självsäkert blir systemet när avståndet till det ledande fordonet minskar. Detta kan dock leda till att det inte finns några lösningar på det optimala kontrollproblemet vid plötslig inbromsning, eftersom fordonet inte hinner bromsa tillräckligt snabbt innan det kraschar. Ju mindre slackvariabeln är, desto mer restriktiv blir CBF som påverkar styrningen mer än vad som är önskvärt vilket också kan leda till olösbart optimalt kontrollproblem. Resultatet från detta examensarbete bär syftet att gynna utvecklingen av säkerhetkritisk CPS som ska användas i praktiska tillämpningar.