Meet ICT4CART: the Interview series, 15th appointment

  1. Meet Ulm University – Institute of Measurement, Control and Microtechnology (UULM)

Ulm University enjoys an excellent reputation for innovative research, interdisciplinary education, and successful training. In recent years, it has been ranked continuously among the world’s top 15 young universities and as one of the best in Germany (Times Higher Education “150 below 50”).

Besides other methodic and application-oriented research topics, the Institute of Measurement, Control an Microtechnology has been conducting research in the field of ADAS ( automated and connected driving ) for more than 20 years. Developed methods are tested in real traffic with own test vehicles and, for connected driving, on a pilot site with intelligent infrastructure at a junction in a suburban area of Ulm, Germany.

UULM in ICT4CART: what is your role in the project?

In ICT4CART, we have been the work package leader for WP3 "ICT Architecture for Connected and Automated Traffic", that means we are responsible for the development of the ICT4CART high-level reference architecture. Furthermore, we coordinate the German test site in Ulm. On the technical side, we operate a pilot installation in real traffic environment. This pilot provides an environment model from infrastructure sensors to the connected automated vehicles; we input the information collected from the environment model in our automated vehicles to cross the junction efficiently and safely. Additionally, together with the other project partners of the German test site we perform tests related to precise positioning and hybrid communication with our vehicle.

What were the main factors to take into account when designing the ICT4CART reference architecture?

One of the most important factors was the integration of hybrid communication. This means that the infrastructure provides its services via a cellular network (LTE or 5G) as well as via ITS-G5, an ad-hoc network similar to WiFi. With that, we can ensure that all connected road users are able to use the services, no matter which kind of connectivity they have installed. Furthermore, we integrated the concept of Multi-access Edge Computing (MEC), which means that the data is processed without large communication delays, either directly on site or at the base station of the cellular network.

What are the main results you have obtained from the implementation of the environmental perception model for improving intersection crossing for road users?

Our environment model is built on detection of camera and fixed-beam lidar sensors mounted on lampposts along the streets close to the junction area. The information from the sensors about the road users are used for an overall environment model of the junction area is built using multi-object fusion and tracking methods. From that, we retrieve our environment model, a mathematical description about the size, position, orientation and movement of the road users. Then, additionally, we apply a prediction network to those data to allow latency and processing time compensation, as well as predictive planning aboard the vehicle based on the environment model. This network generates forecasts regarding the possible movement of objects for the following few seconds. This network is based on deep learning technologies and won the Argoverse Motion Forecasting Challenge in 2019.

What have been the main challenges for UULM? And the main achievements?

As for all partners, the CoViD pandemic was a big challenge: it hit us exactly when the intensive work of integration and testing at the pilot site and in the vehicles should have started. However, despite this challenge, we managed to perform our tasks and are currently in the final phase of implementation and testing to prepare our final event, which will take place in October at the ITS World Congress in Hamburg, Germany. At the final event we want to demonstrate our achievements in videos that show our use cases, the scenarios we have been working on and highlight the advantage of the ICT4CART system.

Another challenge is that operating a prototypical infrastructure sensor system on public roads in general comes with lots of additional efforts besides the technical challenges, such as GDPR compliance.

On the technical side, we found out that the current drafts of standardized messages are not sufficient for the specific use cases we are looking at. For example, the previously mentioned predictions have not yet been included in respective message formats, and this is why we proposed respective extensions.

One of our main achievements was the successful implementation of the ICT4CART architecture on our pilot installation and on our vehicles. For that, we have developed several new methods, like the aforementioned multi-object prediction network, a software simulation for developing planning algorithms based on this prediction network, a distributed implementation for a fusion and tracking filter, generic sensor interfaces for the environment model, easy-to-apply calibration methods for our infrastructure sensors, and methods for estimating the reliability of the communicated data, to name the most outstanding ones. Since these results have, so far, led to the publication of not less than three journal articles and several additional conference papers, for us ICT4CART is also a very successful project from the scientific point of view.

Anything else you would like to mention or highlight?

We hope to see all readers at our project’s final event on 15th October 2021 at the ITS World Congress 2021 in Hamburg, where we would be happy to explain them our achievements in more details.