Better traffic simulation will improve safety assessment of self-driving vehicles

Illustration self-driving cars.
Photo: Akarat Phasura/Mostphotos

A major European research project, i4Driving, will lay the foundation for a new standard method for evaluating the safety impacts of self-driving vehicles. VTI is one of 17 partners that, among other things, will contribute to bringing more behavioural science aspects into traffic simulation models.

One of the goals of self-driving vehicles is to make them safer than those driven by humans. It may sound obvious, but difficult follow-up questions soon arise. When making the claim that self-driving cars should be safer, what should you compare them with? In order to make such a comparison, you have to know how human drivers react in different situations. Furthermore, how will traffic with a mix of self-driving and human-driven vehicles impact safety?

To be able to research this, good models are required to simulate human drivers.

“There exists many such models, develop by engineers for assessing traffic efficiency from a traffic planning purpose. We want to further develop them so that they better describe the behavioural aspects that can lead to incidents and accidents. This may include, for example, lack of attention, fatigue or distraction. The goal is to create a library of driver models that can be used to simulate different situations,” says Johan Olstam, research leader at the Swedish National Road and Transport Research Institute (VTI).

The work will be carried out within the framework of a large, three-year EU project called i4Driving. VTI is one of 17 partners from seven countries. Johan Olstam is the project manager for the VTI group, which initially includes five more researchers (see fact box).

The way each driver responds, for instance, varies depending on their level of experience. With the aid of driving simulator experiments, VTI will provide significant expertise to investigate the extent of the behaviour variance among drivers.

When i4Driving ends in 2026, the researchers will hopefully have laid a solid scientific foundation to develop methods for the safety testing of self-driving vehicles. The project also hopes to offer a simulation environment to safely evaluate different algorithms for self-driving vehicles. It is not enough to simulate how self-driving vehicles behave in different situations; you must be able to simulate a mixed traffic environment where human drivers are also included. A change in the behaviour of a self-driving vehicle does not necessarily lead to increased safety throughout the traffic system. The safe behaviour of some of the vehicles could, in certain situations, lead to unsafe behaviour of some of the vehicles driven by human beings.

“If self-driving vehicles start operating with shorter gaps between vehicles, it could also be possible that human drivers do the same. If that is the case, the sum of the individual gains in each specific situation will not necessarily have a positive effect on overall safety. To evaluate this, traffic simulation models are needed that can capture the variation in human driver behaviour,” says Johan Olstam.

When the researchers have developed better driver models, they conduct a variation of the famous Turing test. They will investigate whether human drivers experience any difference between being driven by a driver model and a human driver. They shall also examine whether human drivers can determine whether the vehicle they interact with in traffic is driven by a human or a driver model.

“It will be interesting to try to evaluate the driver models and get an idea as to how realistic they are perceived. And it will be an exciting challenge to design such experiments.”

Footnote: The article was first published in VTI Aktuellt 1/2023

Text: Johan Sievers/ Redakta

Translation: CBG

Fact box: VTI-researchers in the project

  • Johan Olstam
    Method and model development for the analysis and simulation of dynamic properties of traffic.
    Fredrik Johansson
    Method and model development for the analysis and simulation of dynamic properties of traffic.
    Katya Kircher
    Attention in traffic with respect to different groups of road users, from pedestrians and cyclists to vehicle drivers.
    Christer Ahlström
    Physiological measurements, biomedical signal processing and data analysis with respect to the state of vehicle drivers (for example, fatigue, attention and distraction).
    Anders Andersson
    Driving and pedestrian simulator technology, including co-simulation of such simulators.
    Ary Pezo Silvano
    Speed management, vulnerable road users and autonomous driving.

Don´t miss out on our news – subscribe now!

Stay informed with the latest research and news from VTI. Sign up for newsletters, sent by e-mail four times a year.