Florian Lüttner, Fraunhofer Institute for High-Speed Dynamics
© Fraunhofer EMI
Despite declining accident figures, over 3,000 people still die in road traffic accidents in Germany every year. It is in everyone's interest to reduce this figure further and at the same time reduce the severity of accidents. Autonomous driving can make a significant contribution in this context. Scientists at the Fraunhofer Institute for High-Speed Dynamics, Ernst Mach Institute (EMI), are working on predictive traffic simulations that take into account the increasing complexity of road traffic due to automated driving functions.
Statistically, an autonomous test vehicle would have to travel over two billion kilometers on nationwide roads to provide just 50 percent evidence that it causes only half as many accidents as a human-driven vehicle. Since such test procedures are extremely time-consuming and expensive, the development of predictive traffic simulations plays a key role. For this reason, the Daimler and Benz Foundation has funded a research project at the Fraunhofer Institute for High-Speed Dynamics, Ernst-Mach-Institut (EMI), headed by Florian Lüttner, over a period of three years: “Traffic simulation model as a basis for forecasting traffic accident statistics in future traffic events.”
Data and goals
The researchers first compiled a valid database that statistically records real road traffic with its relevant scenarios – whereby critical situations and accidents play a special role. On this basis, simulation models can be examined to determine how precisely they emulate real traffic and how they can be optimized depending on the scenario. In parallel, existing models that simulate traffic flows were examined for their suitability and were appropriately further developed. The most important goal for the researchers was to generate an optimization procedure that can automatically adapt the simulation environment to the observed data.
Critical scenarios
Particular attention was given to critical traffic scenarios: While extensive information is generally available on normal road traffic and accidents actually occurring, there is hardly any data relating to the critical area of transition – i.e. scenarios in which potential accidents are avoided. On the other hand, these in particular are an important prerequisite for carrying out realistic traffic flow simulations. The researchers therefore developed a flexible system that uses just one numerical value to determine whether and to what extent a traffic situation is critical. It is thus now possible to identify hazardous scenarios on the basis of the few data sources available.
Optimization for the future
Finally, during the three-year funding period, a so-called automated optimization procedure was realized and is now already being used for simple scenarios. Compared to previous approaches to optimization, the required computing time has now been reduced by a factor of four. This was made possible by data analysis and pre-processing carried out prior to the actual optimization. The simulation variables that can be derived from the data – e.g. traffic volume, vehicle category or driver type – are identified directly.
The results of this funded project bring us closer to the goal of reliably determining the probability of accidents in a time- and cost-efficient manner from the observation and analysis of traffic situations alone – both in conventional road traffic and in mixed traffic with autonomous vehicles. The research project will be continued after conclusion of the funding period: As a new aspect, in particular vulnerable road users such as cyclists and pedestrians are to be included in the simulation of inner-city traffic.