Skip to main content



Reporting period: 2020-06-01 to 2021-05-31

Car Simulators are widely used in transport safety research, despite the fact that simulator users behave quite differently in simulators. But how different is this? And can this be improved?

The SimuSafe project -Simulation of Behavioural Aspects for Safer Transport- investigated this so-called simulator bias. At the same time, it also aimed at reducing this bias by improving the realism of the simulations. Vulnerable road users were also included in the study: besides cars as well scooters, bicycles and pedestrians were included in the simulators.

The core of SimuSafe is research on using Road User Simulators as valid tools for studying Traffic Safety, with the following project objectives:
1. Develop Road User behavioural models based on Data Collection
2. Effective, multi-user, multi-agent Road User Simulation reproducing the Road User models
3. Societal Impact: initial steps towards standardisation, safety devices and novel training modules

Road user behaviour was modelled after holistic data collection. These were used in the simulators to make the behaviour of simulated traffic more realistic, e.g. crossing the street aside the crosswalk. Riskless interaction was made possible with the multi-user functionality.

Parallel to this, SimuSafe performed studies leading to specific recommendations on simulator use for driving training programmes. Study and design for future safety devices responding to driver and rider behaviour were initiated during the project.
The project was organised in three research cycles with integrated simulator system development, and parallel work on societal impact (objective 3).

The three research cycles consisted of pilots with external participants performing in real vehicles and in simulators. This involved an extensive data collection, covering the complete data triangle of user - vehicle - environmental data. Sensors ranged from cameras and biosignals to specifically designed devices. Additional important data come from interviews with the participants: pre-test for determining a profile of each person, and post-test self-confrontations with the users explaining motivations for their own behaviour. In all cycles, all four actor types are involved: car, scooter, bicycle and pedestrian.

The first cycle focused on naturalistic behaviour with tests on the public roads of Rome (IT) and Burgos (ES), followed by individual tests in the simulators.

The second cycle was performed on a test track in Cracow (PL), with an additional set of sensors not allowed on public roads. For the simulator system, an exact copy of the test track was created as virtual environment and exactly the same exercises were performed: this allowed for a precise comparison for bias assessment.

Behavioural rules resulting from these two cycles were introduced in the actor agents steering the simulated traffic in the simulator system. The multi-user simulator system was developed in parallel focussing on performance, flexibility and managing the complex data collection.

The third cycle was a first use case of the simulator system with multi-user functionality and advanced actor agents, focused on altered conditions in five different locations in Sweden, UK, France, Italy and Spain. The altered conditions included drug use, medication use and emotional state. The Covid-19 pandemic hindered execution of data analysis within the project, however, collected data has been uploaded to Open Access repository Zenodo. Thus not only the consortium partners but the entire research community can find, use, process and analyse these data.

The Ethics Committee safeguarded the privacy and safety of the participating volunteers. All shared data has been fully anonymised.

The study on standardisation and new training models comprised a broad study of novice driver training and driving instructor training in more than twenty European countries. It showed a very disperse landscape on regulations with still limited interest in simulator use. As example of resulting recommendations: the study identifies specific training where simulators can be useful (e.g. at night, bad weather, ADAS use, eco driving).

Initial steps towards safety devices for car and motorcycle were made. The most advanced one is an alcohol sensor that can be integrated in a motor helmet. A patent has been applied for by SME Hök instruments, who joined with larger company Senseair during the project. R&i performing SME Link Innova has ongoing with companies regarding data collection using simulators and sensor integration in helmets.

The series of workshops covering different aspects of the project’s research, reached many stakeholders from all over the world and from different sectors (industry, public administrations, academic).
The traffic of the SimuSafe simulator system shows non-standard behaviour which highly increases realism of the simulation. Actors have different profiles, resulting in some to behave more aggressive or risky, while others are more cautious. Additionally, rules extracted from the analyses of the collected data are used in the simulations, which becomes visible in e.g. cars speeding more on straight road sections and pedestrians deciding to cross despite a red traffic light.

The SimuSafe simulator system is exceptionally flexible, supporting e.g. four types of road users, multiple environments including road signs from five countries, single-user and multi-user predefined scenarios; and event-based simulation. It achieves very high performance ensuring smooth managing of large amounts of traffic and a large high-quality environment. And although in SimuSafe a maximum of 5 users were interacting with one another in the same environment, the software has no theoretical limit of users.

Several models have been created, documented and shared for risk-related actions like hurry and street crossing.

Behavioural data analyses have shown that gender and age are important factors for behaviour in traffic. For example, higher age relates to a higher responsibility and conscientiousness, lower need for excitement, but also to a lower reaction speed and lower perceptual abilities.

Data analyses also reveal some interesting results concerning differences in risky behaviour between the types of users. The drivers are the most distracted dividing attention between driving and social activities (e.g. talking with the passengers). And the most vulnerable road users, bicycle riders and pedestrians make the highest number of violations, especially failing to look properly before a manoeuvre.

Important knowledge has been generated regarding bias of behaviour when using a simulator. For example, neurometric analysis has shown that simulators generate extra workload to the user, probably because the simulator is something new to many people and they have to learn to use it. Also, the lower perception of risk leads to more speeding.

Many road safety research activities make use of simulators, using the generated data as if they were from the public road, even while aware that this is not the case. Now data is available on this simulator bias and can be taken into account, e.g. by adjusting values.
SimuSafe sensoried motorcycle
SimuSafe newsletter
SimuSafe website
Multiuser pilot
Naturalistic test data analysis
Actor modelling example: decision trees
Online consortium meeting during Covid-19 lockdown
The SimuSafe Team
Motorcycle simulator
Pedestrian video analysis
SimuSafe sensorised car on the test track of project partner Aptiv