Select Page
  • UDRIVE database

    The UDRIVE database constitutes a very rich and detailed set of naturalistic driving data. Between January 2015 and May 2017 almost 100.000 hours of data were collected from three different vehicle types (cars, trucks and powered two-wheelers) in six European countries.

  • Data Acquisition System (DAS)

    UDRIVE provides a unique opportunity to study safety and eco-driving behavior in relationship with personal driving style, infrastructure, congestion and vehicle technology. Data was acquired by using a tailor-made Data Acquisition System (DAS) which collected up to 270 measures including CAN data, footage from up to eight high-resolution cameras, and a smart camera (MobileEye). The picture below shows the camera view for the different vehicle types, followed by a screenshot of the captured video footage from a car.

  • The value of a European naturalistic driving dataset

    What is the added value the UDRIVE dataset in relation to the U.S. SHRP2 dataset?  Naturalistic driving data can be used to gain insight into driver behaviour and develop diversified and targeted safety measures. But people from different countries with other cultural backgrounds express different behaviours. Therefore, a European dataset is extremely valuable because it enables us to compare driver behaviour between not only Europe and the U.S., but also between the different countries within Europe. Insights gained from the UDRIVE data can help develop tailored and targeted policy measures, and provide new scientific insight to support the development of automated driving.

  • How can I use the data?

    The UDRIVE Data User Group is a joint initiative of partners to maintain the remote access to the central data center. The partners are open for collaborations. If you are interested to work with them, please find the contact information below:

    SWOV Nicole van Nes
    DLR Mandy Dotzauer
    IFSTTAR Helene Tattegrain
    University of Leeds Oliver Carsten
    Loughborough University Ruth Welsh
    SAFER Helena Gellerman


    Some partners host the data at their own premises and are open for collaboration. If you are interested to work with them, please find the contact information below:

    SWOV (all data) Nicole van Nes
    CEESAR (car data only) Clement Val


  • The findings in terms of time headway are rather alarming

    Although it is strongly recommended to increase time headway with speed (even measured in time, not in meters), people obviously do the opposite as confirmed by both UDRIVE data and Austrian site-based speed data. This is more or less consistent through all the countries and all vehicle categories and it is valid up to a speed of 80 km/h per hour. National differences are very small.

    Assistance systems like distance warning and adaptive cruise control are the most effective intervention. These systems should support optimised time headway with respect to speed. Public campaigning could also raise awareness on the issue. However, campaigns should on no account focus on time headway in general, but on its adoption to speed. Further research on how exactly drivers choose time headway is needed.

    In any case, PTW riders in general do not have to worry about being followed too closely by cars, which also means that further research must investigate other reasons than short time headway for rear-end collisions.

  • Safety by numbers applies for pedestrians

    Expectancy is crucial: when pedestrians are present, drivers are better prepared for pedestrians in general and more ready for responding performing unexpected actions (e.g. stepping out).

    The notion of “surprise” plays an important role: when Pedestrian & Cyclist Detection and Collision Warnings (PCWs) were preceded by other events involving pedestrians (PCW or Danger Zone (DZ) warnings), drivers significantly reduced their speeds and were better prepared for a potential conflict with pedestrians.

  • The blind spot was rarely checked before turning manoeuvres

    Failure to perform appropriate visual checks at intersections may have contributed to the 2069 cyclist fatalities in the European Union in 2015. We investigated whether and when drivers perform visual checks for potentially encroaching cyclists during right-turn manoeuvres (UK: left turn). Data was collected from 69 car drivers in France, the Netherlands, Poland, and UK, and from 12 Dutch truck drivers. The car dataset consisted of 961 manoeuvres at intersections (trucks: 159) and 852 manoeuvres at roundabouts (trucks: 209).

    In the six seconds before the manoeuvre, car drivers checked their blind spot in 4% of the intersections and roundabouts. Including the manoeuvre, this frequency increased to 7% at intersections and 5% at roundabouts. The prevalence of cyclist facilities was highest in the Netherlands, which is also the country where blind spot checks were performed most often (27% at intersections, 19% at roundabouts). Dutch truck drivers checked their blind spot at 19% of the intersections and 27% of the roundabouts. When not checking the blind spot, car and truck drivers mostly looked toward the road they were turning into. Our findings may inform driver training to increase awareness of cyclists at urban intersections and roundabouts.

    Manoeuvre Anatomy


  • Vehicle dynamics of powered two-wheelers (PTWs)

    Vehicle dynamics with powered two-wheelers (PTWs) are much more difficult than for cars: there’s one degree of freedom more, and, according to the findings of UDRIVE, riders’ preferences in terms of dynamics strongly differ. That means that design of assistance systems for PTWs is more difficult than for cars and the individual skills and preferences of the rider have to be considered.

    It was found that despite the use of a high-quality data acquisition system, detection of high-risk events with powered two-wheelers remains like searching a needle in a stack of hay. Everyday riding behaviour does not detectably differ from emergency behaviour. This is most likely caused by poor skills of riders and a lack of willingness to perform highly dynamic manoeuvres – being aware of the particular danger of falling off the vehicle. Dynamic preferences of riders strongly differ within the population and even within subjects. And this tolerance for high dynamics seems to influence emergency behaviour as well as everyday behaviour.

    This could be changed by training, but the everyday behaviour would most likely be changed at a time. For naturalistic studies, this phenomenon suggests the use of an incident button, trip records or periodic interviews. For the design of rider assistance systems, this automatically means that systems will have to be highly adaptive with respect to the preferences of a rider.

  • Further reading

    D44.1 – Interactions with vulnerable road users

    This deliverable analyses the interactions of pedestrians, cyclists and PTWs with passenger cars and trucks. The aim was to identify and understand the everyday behavioural patterns in these interac-tions as well as the circumstances of safety critical events (SCE). For cyclists, identified SCEs were caused by a combination of infrastructure (a curve or a too narrow road), manoeuvre (often over-taking), the presence of other traffic, and an error or unexpected behaviour of the cyclist (slowing down). For pedestrians, in around three quarters of SCEs, the driver him- or herself had spotted the pedestrian in time. In the remaining situations, a warning system could have been of help. For PTWs, the data did not show that car drivers tend to follow them closer than cars or trucks.

  • Overall involvement in secondary tasks for trucks is much higher than for cars, about 20% of the total time

    · Most time is spent in food and phone (5% each)
    · Most frequent is operations of vehicle control: 8 (7.78) times/h
    · Out of 5% of phone use, half (2.4 %) was texting, reading or other visual interactions (talking on the phone was relatively small proportion)

  • Car drivers were involved in distracting activities for 10% of the time spend driving · The most common distracting activity is hand-held mobile phone use; this is in total 4.2% of driving time · In Poland, mobile phone use was significantly higher than in other countries (UK, France, Germany)
    This is significantly less than in the U.S. (as reported by Dingus et al) and methodological differences between the U.S. and European studies cannot account for that difference
    There are indications that car drivers attempt to self-regulate their activity: there is considerably more activity when stopped or moving slowly


    Supporting documents 

    D41.1 Synthesis of analysis results

    This deliverable presents key results of the analysis performed in UDRIVE Sub-project 4: Data analysis. It also describes the UDRIVE dataset. The analysis in UDRIVE was facilitated by tools such as the quality assurance procedures and data tracking, the SALSA data processing tool, the UDRIVE annotation codebook and high-quality manual annotation of video. The analysis itself is described in short in this report, while details are presented in separate UDRIVE deliverables. In summary, a large variety of analyses was performed on the UDRIVE naturalistic driving data (NDD). Although the efforts and results have been significant and already impact safety measure design and development, the UDRIVE project has only scratched the surface of the analysis potential.


  • Auto annotation – if properly trained, automated identification of distracting activities is feasible
  • Further reading

    D43.1 – Driver distraction and inattention

    This deliverable provides better understanding of whether and how drivers manage their secondary task activities. An automated procedure has been applied to provide candidate cases of secondary tasks to manual annotators. The automatic annotation tool is based on deep learning algorithms. The focus of the research questions was on self-regulation, on how drivers manage their secondary task activity in the context of the dynamics of the traffic and road situation. That man-agement includes the determination not to engage in such tasks in the first place or only to engage in some particular activities. The deliverable finds that car drivers spent 10.2% of their driving time engaged in some kind of secondary tasks. The total time spent in all the secondary tasks for truck drivers sums up to about 20%.  The duration of secondary task was affected by complexity of manoeuvre. There are thus indications of some self-regulation by drivers.

  • Anticipating traffic when driving, correlated with safe driving, results in less braking and therefore better eco-driving behaviour

    The more headway a driver keeps, the less he needs to brake because he has time to anticipate the traffic. Indeed, this figure shows that more headway corresponds with less braking, when comparing the points for one velocity range, e.g. 0-50 km/h. These results are retrieved for one driver after binning the data points per trip in three velocity ranges. Each point represents one trip. We also see that the higher the velocity, the lower the average time headway and the lower the braking energy. Since braking is the main source of energy loss in urban driving, it should be avoided whenever possible by anticipating the traffic situation.

  • Different driving styles can be identified by different patterns in gear shifting and braking behaviour. The difference in fuel consumption and emissions due to driving behaviour can be up to 25% in urban areas

    Since some drivers shift gear much earlier than others, gear shifting behavior is a good indicator for eco-driving behaviour. Drivers that shift gear at low velocities and moderate accelerations enter third gear around 30 km/h, whereas others only reach third gear at 45 km/h in the same type of vehicle.

    In terms of fuel consumption, the lower the engine speed the better, and the eco-driver advice is to change gear between 2000 and 2500 RPM. The average gear shifting RPM for different drivers however ranges from 1400 to 3000 RPM, depending on the vehicle type and the gear, but mostly on the driver behaviour. This large bandwidth means there is quite some room for improvement by better eco-driving behaviour. The estimated difference in fuel consumption due to different engine speeds can be up to 20-25%.


  • Further reading

    D45.1 – Potential of eco-driving

    This deliverable offers the analysis of possibilities of the naturalistic driving data to provide more in-sight in different (normal) driving styles and eco-driving. Unique to UDRIVE is the augmentation of the velocity data with driving circumstances, like road type, speed limits, headway, and in-vehicle information. This allows placing the driver behaviour in context and distinguishing personal driving styles from behaviour forced by traffic conditions. To assess the fuel consumption and CO2 emis-sion reduction potential associated with adopting an eco-driving style, it is crucial to separate per-sonal driving style from infrastructure and from congestion while driving.

  • Cruise Control (CC) and Speed Limiter (SL) use

    Cruise Control (CC) is used more often than the Speed Limiter (SL), but all together they are used only in approximately 10% of the trips. 17% of the equipped drivers didn’t even know they have these functions. Drivers generally don’t use both functions, they specialize either in  CC use or SL use. The road types where CC and SL are used are equivalent.

  • Seatbelt use differs between countries

    Globaly speaking, 87,3% of the trips are driven with the seatbelt on from start to end. This percentage is different from one country to another, with a lower rate for Poland (76%) and a higher for the Netherlands (95,6%). The driver country is the most significant parameter explaining the behaviour of driving for some distance without wearing the seatbelt. The second most significant factor is gender: more male drivers do not wear seatbelts. For whole trips without driver seatbelt use, the most significant variable is trip distance: not wearing the seatbelt happens more frequently during very short trips (<325m) and especially at night.

  • Further reading

    D42.1 – Risk factors, crash causation and everyday driving

    This deliverable reports the results of normal and risky driving behaviour. The safety critical event definition section explains the procedure of creating safety critical event triggers (SCE). Actual crashes are very rare, even in a data collection of over 21 months and 200 vehicles. Thus, it is almost impossible to investigate crashes directly. Surrogate measures are used instead to identify and assess potential risk factors. Hard braking, sudden steering, and accelerations are used as surrogates for collisions. While it is reasonable to assume a connection between these surrogates and real crashes, researchers are still uncertain whether SCEs and crashes follow the same patterns. None-theless, SCEs are still the best option to investigate how crashes are caused. In addition to SCEs, episodes with a high relevance to road safety were investigated. On rural roads, more crashes occur than on highways and they are more severe than in cities. This makes them a highly relevant research area in respect to traffic safety.