Meet EVENTS team-TUD

Delft University of Technology (TUD) is the largest and oldest technical university in the Netherlands, renowned for excellence in engineering, science, and design. It consistently ranks among the global leaders in engineering and technology — for example, 13th worldwide in Mechanical, Aeronautical & Manufacturing Engineering (QS 2025).

The Intelligent Vehicles (IV) Group at TUD conducts problem-driven research to enhance transportation safety, comfort, and sustainability through automated driving technologies. Its work spans the full self-driving stack — from machine perception and motion planning to control and human factors. The group’s primary experimental facilities include a moving-base driving simulator and a prototype research vehicle (a Toyota Prius). The IV Group is internationally recognized for pioneering research on vehicle interactions with pedestrians and cyclists in dense urban environments, as well as on high-dynamic vehicle maneuvers. The group is led by Prof. Dr. Dariu M. Gavrila.

Can you provide an overview of your role and involvement in the EVENTS project?

Within EVENTS, TUD focuses on on-board environment perception and motion planning & control for automated driving. The perception research addresses 3D object detection, with an emphasis on developing efficient techniques for data annotation. The motion planning and control work covers two key scenarios: a low-speed scenario involving interactions with multiple vulnerable road users (VRUs – pedestrians, cyclists, and other riders), and a high-speed scenario involving evasive maneuvers under low-friction (slippery) road conditions.

TUD serves as Work Package (WP) Leader for WP3 – Perception and Self-Assessment, and will demonstrate both VRU scenarios in experiments using its Toyota Prius vehicle.

What are the key challenges in achieving reliable automated driving?

Automated driving has advanced considerably in recent years. For example, highway autopilot systems now allow drivers to divert their attention and engage in secondary activities until prompted to retake control (conditional automation). Meanwhile, driverless vehicles (“robotaxis”) have begun providing commercial mobility services within geofenced areas in several U.S. and Chinese cities (high automation).

However, a major remaining challenge is enabling automated vehicles to safely and efficiently navigate in the presence of Vulnerable Road Users (VRUs) —such as pedestrians and cyclists—in dense urban environments. VRUs pose persistent difficulties due to their diverse appearances (e.g., varying articulations and clothing), high maneuverability, unpredictable behaviors, and occasional non-compliance with traffic rules.

Adverse weather conditions further increase the difficulty by reducing perception performance and making vehicle control more challenging.

How do you address the beforementioned challenge(s) in EVENTS ?

We aim to improve the perception of Vulnerable Road Users (VRUs) by training 3D object detection networks with larger and more diverse datasets. However, manual data annotation remains a costly bottleneck. Therefore, we are investigating unsupervised learning techniques to automatically annotate moving objects in large-scale traffic datasets. Furthermore, we are exploring novel sensing modalities such as 4D radar to achieve better perception performance under adverse weather conditions (e.g., rain). Finally, we are looking at the fusion of multiple sensor modalities (camera, LiDAR and 4D radar) to increase robustness.

In the area of motion planning, nonconvex optimization methods are widely used for real-time trajectory generation. Yet, they often converge to locally optimal solutions and may switch between local minima, resulting in inefficient or unsafe vehicle behavior. To address this, we are developing a topology-driven trajectory optimization framework for dynamic environments, capable of planning multiple distinct evasive trajectories to enhance both safety and efficiency.

In the area of control, we are exploring methods capable of handling highly nonlinear vehicle dynamics, such as those occurring during emergency evasive maneuvers involving VRUs on low-friction (slippery) road surfaces.

What does the future hold for Automated Driving?

The future of automated driving promises continued technological progress coupled with gradual, domain-specific deployment. In the coming decade, we can expect increased adoption of highly automated systems in controlled environments—such as highways, industrial zones, and dedicated urban corridors—where conditions are predictable and infrastructure is supportive. Advances in sensor fusion, machine learning, and high-definition mapping will enhance reliability, while vehicle-to-everything (V2X) communication will enable safer coordination among road users. At the same time, driver-assistance functions will become more capable, steadily narrowing the gap between conditional and high automation for private vehicles.

However, full autonomy in complex, mixed urban traffic remains a formidable challenge. Future progress will likely depend on combining technical innovation with regulatory evolution, infrastructure support, and societal acceptance.

Anything else you would like to mention or highlight?

TUD would like to thank all EVENTS partners for their excellent collaboration and collective achievements. TUD also gratefully acknowledges the European Commission for its funding support in advancing this key technological area.