Home Program Directions

Workshop Program

8:50am Introduction
Introduction to the workshop and overview of the day's program.
9:00am The Era of End-to-End Autonomy: Transitioning to Large Driving Models and the Future of Automated Vehicles
Em. Prof. Eduardo Nebot, Emeritus Professor at University of Sydney
Automated driving is undergoing a fundamental transition from modular, rule-based architectures to end-to-end (E2E) learning systems that map sensor observations directly to vehicle motion. This presentation examines the technologies driving that transition and their implications for drivers, vehicle manufacturers, and the broader deployment of automated mobility. Particular attention is given to the emerging category of supervised E2E driving, often described as E2E Supervised or L2++, in which the vehicle performs a substantial portion of the Dynamic Driving Task (DDT), while the human driver remains responsible for continuous supervision and fallback intervention. Using a current deployed technology, Tesla FSD Supervised as a primary case study, the presentation explains how large driving models combine imitation learning, reinforcement learning, fleet-scale data collection, simulation, and targeted edge-case analysis. Their operation in right-hand-drive and left-hand-drive environments around the world demonstrates the potential for natural driving behaviour, rapid responses, road-rule compliance, and improved handling of long-tail scenarios. The presentation distinguishes supervised Level 2 automation from Level 4 robotaxi operation and considers related E2E technologies being developed by other vehicle manufacturers. While these systems offer significant opportunities for continuous improvement through fleet-scale learning, the presentation also introduces the unresolved challenges associated with their deployment, including validation, transparency, safety assurance, human-machine interaction, driver attention, liability, and regulatory assessment. Finally, drawing on the technologies, operational experience, and challenges discussed, the presentation outlines how future vehicle systems are likely to evolve. Advanced sensing, centralized computing, continuous connectivity, over-the-air updates, and fleet-based learning are expected to become core capabilities of next-generation vehicles.
9:30am Road users’ behavior prediction as an advanced driving assistance system
Prof. Miguel Angel Sotelo, Full Professor at the Department of Computer Engineering at the University of Alcalá, Spain
Automated Vehicles (AVs) have experienced a booming development in the latest years, having achieved a large degree of maturity. Their scene recognition capabilities have improved in an impressive manner, especially thanks to the development of Deep Learning techniques and the availability of immense amounts of data contained in well-organized public datasets. But still, AVs exhibit limited ability to deal with certain types of situations that become natural to human drivers, such as entering a congested round-about, predicting the presence of occluded pedestrians at cross-walks, dealing with cyclists, or giving way to a vehicle that is aggressively merging onto the highway from a ramp lane. Not to mention their limitations to interpret implicit communication -when interacting with other road users- and to anticipate dynamic scenarios, especially those dealing with near-crash situations. All these tasks require the development of advanced capabilities that rely on contextual reasoning in order to: a) anticipate the most likely behaviours of all traffic agents around the ego-car; b) adapt the AV’s own actions to the anticipated road users’ behaviours in a socially-accepted manner, and c) reason about the main factors that caused the decisions taken by the AV. These features will open the gate to expanding the operational design domain of AVs and to contribute to their societal acceptance. In this talk, latest results achieved at UAH in the framework of contextual and causal reasoning for behaviour understanding and prediction will be presented and discussed, with a special focus on the prediction of occluded pedestrians and the anticipation of near-crash scenarios.
10:00am What End-to-End Driving Still Needs From the Human Driver
Univ.-Prof. Dr. Cristina Olaverri Monreal Full Professor, Head of Department Intelligent Transport Systems at Johannes Kepler University in Linz, Austria
Supervised end-to-end (E2E) driving systems can now perform most of the driving task while legal responsibility remains with the human driver. This transition introduces interconnected challenges related to safety validation, driver supervision, human–machine interface (HMI) design, road-environment performance, and ethics. Although each of these areas has been studied extensively, they are rarely examined together as part of a unified framework for the safe deployment of supervised E2E driving. Drawing on research in driver monitoring, human factors, in-vehicle interfaces, vulnerable road user interactions, perception, and vehicle safety, this talk synthesizes current evidence on supervised E2E driving. It highlights established findings, identifies key knowledge gaps, and discusses the remaining challenges in driver supervision, safety validation, liability, and regulation.
10:30am Coffee & Tea Break
11:00am Interesting Title of a Talk
Javier Ibanez Guzman (Renault)
Abstract 3.
11:30am Transitioning to Objective Metrics for Real-World ADAS Evaluation
Dr. Mao Shan, Senior Research Fellow at the University of Sydney
Traditional assessment of Advanced Driver Assistance Systems (ADAS), such as Lane Departure Warning (LDW) and Lane Keep Assist (LKA), relies heavily on standardised, closed-track protocols (e.g., Euro NCAP) or subjective driver feedback. However, track testing is frequently costly, limited in its coverage of diverse road geometries, and insufficient to capture the complexities of everyday naturalistic driving. This presentation details a research project at the University of Sydney aimed at the objective evaluation of lane-support ADAS within real-world driving environments. Central to this framework is a sensor system designed to provide an accurate, continuous reference for a vehicle’s lateral position relative to the lane boundary. This framework is intended to inform superior system design, influence OEM safety standards, and enhance overall road safety through precise, naturalistic performance metrics.
12:00am Interesting Title of a Talk
Pannel?
Abstract 3.
12:30pm Lunch