Cameras, radar, lidar, ultrasound – modern cars have an increasing number of sensors. They assist with parking, monitor blind spots, and help you keep your distance from the car in front. In autonomous vehicles, these sensors must go beyond just providing convenient assistance, and also reliably capture the entire environment. However, differing weather conditions pose huge challenges for the technology. For example, strong background light, snowfall or dense fog can mean that the system does not identify obstacles, other vehicles or people until too late or fails to do so at all. To enable automated driving functions to be used safely even under adverse environmental influences, researchers at the Heinz Nixdorf Institute at Paderborn University and the Fraunhofer Institute for Mechatronic Systems Design IEM are working with industry partners to improve the ability of sensors and sensor systems to withstand environmental conditions such as bad weather or pollution. To do so, they have developed virtual environments where they can test reliable environmental sensors for highly and fully autonomous vehicles so that they can then be improved on the basis of the results. As well as the scientists at Paderborn, the companies HELLA GmbH & Co. KGaA (network coordinator), dSPACE GmbH, RTB GmbH & Co. KG and Smart Mechatronics GmbH are also involved in the "robustness to environmental conditions of sensors and sensor systems for highly automated driving" project (rosshaf), which was launched in April and has received 2.81 million euros of funding from the German Federal Ministry of Economic Affairs and Energy (BMWi).
Only safe in good weather?
From driving assistants to self-driving cars, vehicles are now able to perceive their environment to a varying extent using numerous sensors. The higher the vehicle’s level of automation, defined by the SAE level set by the Society of Automotive Engineers, the higher the requirements placed on the car. As well as complex technical equipment, the systems in question must also be very robust. Environmental sensors in particular are a critical factor. "Environmental sensors are pivotal for identifying objects in the area surrounding the vehicle. With the rising level of automation, vehicles are becoming increasingly dependent on continuously functioning sensors", emphasises Nico Rüddenklau, a research associate in the "Control Engineering and Mechatronics" research group headed up by Prof. Dr. Ansgar Trächtler at the Heinz Nixdorf Institute.
For example, vehicles with an SAE level of 3, i.e. vehicles with conditional automation, have now become increasingly common on public roads. However, their use requires the driver to always be ready for action and able to take over control. "This is required if certain vehicle sensors are defective or impaired. As rain, snow and fog can cause malfunctions such as these, they are not a rare occurrence, particularly given the weather conditions in Germany. On the other hand, SAE level 5 (i.e. fully automated) vehicles must be able to function at any time and anywhere in the world, without requiring any intervention from the occupant", the Paderborn scientist explains.
Simulation in bad weather conditions
This is where the project comes in. "We are working to make autonomous vehicle sensors robust enough that they can be used at SAE levels 4 and 5, i.e. for highly and fully automated vehicles. With our project, we are seeking to help develop such vehicles and bring them to streets in the future", Rüddenklau explains – because designing automated driving functions to meet the requirements of SAE levels 4 and 5 still requires further research. In order to be able to intensively test and optimise environmental sensors, the project partners are virtually mapping out the individual sensors and simulating how they behave in bad weather conditions. They began by developing a simulation environment in order to achieve this. This enables specific driving situations and accident scenarios that could not be tested in real life for safety reasons to be incorporated into the investigation. In addition, the simulations mean that the tests are quick to implement, cost-efficient and reproducible, Rüddenklau emphasises.
