Author: Stefano Arrigoni
-
Generative AI-Enhanced Sensor Fusion: Robust LiDAR-Radar Inpainting and Restoration in Extreme Weather
Introduction This thesis investigates the application of Generative AI for the intelligent fusion of heterogeneous sensor data. The goal is to develop a Generative AI framework, specifically leveraging Conditional Diffusion Models, for real-time point cloud restoration. The research focuses on using Radar data, which is inherently weather-resilient, as a “semantic and physical prior” to denoise
-
Resilient Autonomous Navigation via Generative AI: Virtual Sensor Synthesis for Real-Time Fault Recovery
Introduction What happens when an autonomous vehicle’s camera is covered by mud or a LiDAR module suffers a hardware failure? Currently, the car performs a “safe stop.” This thesis aims to replace this passive safety with Generative AI-driven resilience, creating a system that can “imagine” missing data. Goals The project focuses on building a Virtual
-
Optimality assessment of an NMPC-based GLOSA system
Introduction The assessment of solution optimality is a crucial aspect in the design of real-time control strategies for intelligent transportation systems. In the context of Green Light Optimal Speed Advisory (GLOSA), Nonlinear Model Predictive Control (NMPC) is widely adopted due to its ability to handle system constraints and nonlinear vehicle dynamics while operating in real
-
Development of a centralized eco-driving system
Introduction Centralized eco-driving systems exploit vehicle-to-infrastructure (V2I) communication to coordinate vehicle speed profiles based on traffic signal and network information. By enabling predictive and cooperative driving, it reduces energy consumption, emissions, and travel time, while improving traffic flow efficiency and driving comfort. Goals This thesis proposes the development of a smart traffic light control system
-
Manual-Driving Data Acquisition System for Dataset Creation in Unstructured Private Areas with VRU Interaction
Introduction High-quality datasets are a key enabler for data-driven methods in autonomous driving and mobile robotics. This is particularly true in unstructured environments and private areas (e.g., campuses, industrial sites, private roads), where vehicle behavior and interactions can differ substantially from public-road scenarios. In these contexts, collecting representative data of vehicle operation and VRU (Vulnerable
-
Industry-Driven Topics in AV Software, Teleoperation and Fleet Management with Tecnocad Group
Introduction This thesis “umbrella” is developed in interaction with Tecnocad Group, an Italian engineering company operating across mobility sectors and providing end-to-end engineering development from concept to production.Tecnocad Group delivered a guest presentation within the course, outlining several industry-relevant directions that can be shaped into a thesis topic depending on the interests and technical profile
-
ADS Safety, Monitoring and Reporting Topics in collaboration with JRC (Ispra)
Introduction The Joint Research Centre (JRC) is the European Commission’s science and knowledge service, providing independent scientific advice and evidence to support EU policymaking. JRC in Ispra (Italy), is one of the European Commission’s major research campuses.This thesis “umbrella” collects potential topics inspired by those lectures, with a focus on safety-oriented analysis for Automated Driving
-
Dataset-Driven Robust Ego Localization for Autonomous Vehicles in Challenging Sensing Conditions
Introduction Robust ego-vehicle localization is a core requirement for autonomous driving. In real-world operation, performance can degrade due to adverse weather and non-ideal sensor behavior, such as dropouts, partial failures, miscalibration, or temporal misalignment. This thesis topic focuses on dataset-driven, learning-based approaches to improve localization robustness in multi-sensor settings, with systematic experimental evaluation. Goals This
-
Adaptive Collision Avoidance in Mixed-Traffic Autonomous Driving via Hamilton-Jacobi Reachability and Learning-Based Techniques
Introduction When it comes to the safety of autonomous vehicles (AV), developing algorithms that balance conservatism and performance is the key to achieve methods that allow Avs to be integrated with human driven ones.One of the most promising techniques is the Hamilton-Jacobi (HJ) backward reachability analysis (based on dynamic programming and differential games), which allows
-
Agents for Competitive Multi-Agent Autonomous Vehicle Racing: Learning-based vs Model-base
High-speed, multi-agent autonomous vehicle racing demands agents that can make split-second decisions while strategically interacting with unpredictable opponents. This thesis will develop a learning-based racing agent and a model-based one both capable of long-term reasoning, balancing raw speed with tactical maneuvers, and benchmark their performance under head-to-head competition. Requirements and tools Contacts: Michael Khayyat, Stefano