Road safety related impacts within the Levitate project

Working paper of the road safety working group of the H2020 project LEVITATE
Author(s)
Weijermars, W.; Zwart, R. de; Mons, C.; Gebhard, S.; Cleij, D.; Sha, H.; Chaudhry, A.; Boghani, H.; Haouari, R.; Quddus, M.; Thomas, P.; Hula, A.
Year

LEVITATE (Societal level impacts of connected and automated vehicles) is a Horizon 2020 project that aims to forecast impacts of developments related to Cooperative, Connected and Automated Mobility (CCAM). Impacts are estimated for different so-called ‘sub use cases’ (SUCs) that reflect applications or interventions which can be implemented by policy makers. The impacts for the sub use cases are estimated by comparing the situation with intervention to the situation without intervention, i.e., the baseline scenario. The baseline scenario reflects the starting point for which increasing penetration levels of first cautious and later more ambitious automated vehicles (CAVs) are estimated over time.

One of the relevant impact areas of CCAM is road safety. This Working Document 1) discusses in which way road safety is impacted by increasing penetration levels of connected and automated vehicles (CAVs) and 2) quantifies the road safety impacts of increasing penetration levels of CAVs as far as possible.

Identified road safety impacts

First of all, it was investigated in which ways road safety is impacted by increasing penetration levels of CAVs. The impact diagram that was constructed in Deliverable 3.1 was used as a basis for this investigation and was further elaborated with a review of literature and expert knowledge. To check whether all relevant impacts were included, we discussed the identified impacts with experts outside Levitate and asked for additional input.

Deliverable 3.1 makes a distinction between primary or direct impacts and secondary or indirect impacts. Road safety is expected to be influenced both directly and indirectly by increasing penetration levels of CAVs.

Primary road safety impacts of CAVs

Many risks related to human drivers are expected to be prevented or decreased by CAVs. CAVs are assumed to obey traffic rules and are expected to be able to prevent most human driver errors. Moreover, they are expected to have lower reaction times and less variability in driving behaviour. Therefore, CAVs will likely have a lower risk of being involved in a crash than human driven vehicles.

On the other hand, some new potential risks might be introduced by CAVs. These potential new risks include:

  • Risks related to system failures or system degraded performance; problems related to sensors, software or other components of the system
  • Risk related to cyber-security; hacking, cyber attacks
  • Risks related to transition of control to human drivers or mode-confusion.

Secondary road safety impacts of CAVs

In addition, some rebound/indirect effects can be expected, caused by changes in broader factors that in turn affect road safety. Indirect impacts include changes in modal split, total distance traveled and route choice, as well as changes in traffic behaviour of other road users and infrastructural changes due to the introduction and increasing penetration levels of CAVs

Development of impacts over time

The impact of CAVs on road safety is not a static figure, but will likely evolve over time as CAVs are expected to become progressively safer, drivers and other road users will become more experienced in dealing with CAVs, and the penetration level of different types of CAVs is expected to increase over time. To account for developments in performance of CAVs over time, two types of CAVs are distinguished within Levitate; first generation, comprising more cautious CAVs, and second generation, comprising more ambitious/aggressive CAVs.

Quantification of road safety impacts

Concerning the quantification of road safety impacts, the available literature was first studied looking for quantitative information on road safety impacts. Second, road safety impacts were estimated by combining three approaches:
1. Impacts of decreased reaction times and driver variability on crash rates between motorized vehicles are estimated by means of microsimulation.
2. The impact of improved driving behaviour on crash rates between motorized vehicles and vulnerable road users is estimated by using crash data and assumptions concerning types of crashes that can be prevented by CAVs and reduced reaction times.
3. The estimated impacts on crash rates are combined with estimated impacts on distance travelled that are determined via other methods within LEVITATE to estimate the overall impact on the number of crashes.

Information from literature

Available literature provides some quantitative information on road safety impacts of increasing penetration levels of CAVs. Previous microsimulation studies estimated impacts of lower reaction times and less variations in driving behaviour and report reductions of safety critical events up to 99% at a 100% penetration rate of CAVs. By combining different studies it is possible to determine dose-response curves that provide estimated impacts for increasing penetration rates. Quantitative information on the size of potential new risks of CAVs was not found in literature.
Other types of studies that were considered in the literature review are studies combining crash data from human driven vehicles and CAVs, studies looking at disengagement reports, studies that use data from tests with CAVs and studies that apply naturalistic driving data. These studies however are not suitable for estimating future impacts of CAVs that are not fully developed yet. Driving simulator studies can be used for obtaining more information on specific impacts like transition of control or impacts of CAVs on human drivers.

Microsimulation approach

Impacts on crashes between motorized vehicles are estimated by means of a microsimulation model (AIMSUN) that is also used to estimate impacts on, for example, travel times and emissions within Levitate. Output from the AIMSUN microsimulation model is postprocessed using the software application SSAM (Surrogate Safety Assessment Model) to estimate changes in conflicts per distance travelled. A probabilistic method proposed by Tarko (2018) is then applied to estimate the change in crash rate based on the change in conflicts and their Time to Collision (TTC) values. Impacts on conflict and crash rates are estimated for three calibrated and validated networks: Manchester (UK), Leicester (UK), and Athens (GR). As light and heavy good vehicles appeared to have an unrealistically high share in the number of conflicts and may not be reasonably modelled in the test networks, it was decided to remove these vehicles from the analysis. Moreover, as there seemed to be a peak in conflicts with TTC ≤ 0.1 sec that is likely caused by limitations in the microsimulation and/or SSAM software, it was decided to exclude these conflicts from the analysis as well.

At 100% market penetration rate of CAVs, conflicts per 1000 vehicle-kilometer are estimated to be reduced by almost 90%. Results are comparable for the three networks. The expected change in crash rate differs between the three networks: at 100% penetration of CAVs, crash rates are estimated to decrease by 87% in the Manchester network, 92% in Leicester, and 68% in the Athens network, all in comparison to present values with 0% CAV market penetration rate.

Impacts on vulnerable road users

Unmotorized vulnerable road users (VRUs), comprised of pedestrians and cyclists, are not included in the microsimulation model and therefore, crashes involving VRUs are not taken into account in the impacts discussed above. As developments related to CCAM are expected to impact road safety of VRUs as well, another approach based on crash statistics was taken to estimate the impacts on crashes with VRUs.

This approach is based on three main assumptions:

  1. It is assumed that all crashes that were caused by human-driven vehicles (car is ‘at fault’) can be prevented by CAVs
  2. As CAVS are expected to have lower reaction times than human-driven vehicles, it is assumed that the remaining crashes (VRU is ‘at fault’) are less severe when CAVs are involved instead of human-driven vehicles
  3. The CAV systems are highly developed and reliably operational across all relevant real-world scenarios.

The share of crashes for which the pedestrian or cyclist is registered to be ‘at fault’ differs between cities and countries. Based on crash statistics from a number of countries, we assume that about 70% of the crashes is caused by human-driven vehicles. In that case, 70% of the VRU-car crashes can be prevented in case of a 100% penetration level of CAVs. Taking into account the extra reduction in (severe) crashes due to the reduced impact speed, it is estimated that 91% of all fatal crashes between VRUs and cars can be prevented in the case that all cars are fully automated.

Overall impact taking into account changes in modal split

To estimate the overall impact of increasing penetration levels of CAVs on road safety, the impacts on crash rates that are estimated above, are combined with estimated impacts on distance traveled with various transport modes. The impacts on distances traveled are estimated by means of System Dynamics/Mesosimulation and are discussed in more detail in Deliverable 5.3 (Roussou, Müller, et al., 2021) and Deliverable 6.3 (Sha et al., 2021).
The impacts on modal split for the baseline scenario are only available for the Athens network. In Athens, the distance traveled by car is expected to increase by around 8%, due to a decrease in active travel of 2% and a decrease in travel by public transport of 6%. An increase in travel by private car transport is expected to have a slight negative effect on road safety. Overall, this results in an expected decrease in crashes of 74% in comparison to present values with 0% CAV market penetration rate.

Discussion and limitations

It should be stressed that the quantification of impacts in this report is based on many assumptions. The results based on microsimulation for example depend on the parameter settings for the behaviour of human-driven and automated vehicles. Moreover, freight vehicles and TTC values ≤ 0.1 sec were removed from the analysis. The impact on VRU-car crashes is also based on assumptions concerning the share of crashes that can be prevented by CAVs and the expected reaction time of CAVs. Furthermore, for both impacts described above, it is assumed that CAVs function perfectly and that the human drivers don’t need to and are not able to take over control. In addition, issues related to hacking or cyber-attacks are not taken into account.

The impacts on vehicle kilometers travelled are used as fixed input for the estimation of the overall road safety impacts and are outside the scope of this report, yet also these estimations are based on assumptions. Further indirect impacts are not taken into account when estimating the overall road safety impacts.

Furthermore, the results are based on estimations for a limited number of cities. Impacts or crashes between cars are estimated for three networks and the proportion of remaining crashes appear to differ between the three networks. The estimated impacts on VRU is based on crash data from a limited number of countries and model split impacts were only available for Athens. Therefore it is not possible to determine to which extent the estimated total impacts are transferable to other networks.

Because of these limitations, the quantified impacts discussed in this report should be seen as a rough estimate of the potential road safety impacts of increasing penetration levels of CAVs. Unavailability of real-world data makes it impossible to validate the results.

In general, it should be noted that it is very difficult to quantify expected road safety impacts of increasing penetration levels of CAVs as CAVs will continue to develop over time and future vehicle specifications are not yet known. Moreover, policy makers can influence the road safety impacts of CAVs, for example by regulating the conditions that must be met by CAVs to be allowed on the public roads. Moreover, measures like urban shuttles, provision of dedicated lanes or automated freight consolidation have additional impacts on road safety. These impacts are also identified in LEVITATE and are discussed in Deliverables 5.4, 6.4 and 7.4 of the Levitate project.

Although the exact impacts are not known yet, CAVs are expected to improve road safety. They would not be accepted by road users and policy makers if they would not improve road safety. However, it should be stressed that CAVs cannot be expected to solve all road safety problems. First of all, even systems that are very well designed can fail, and cyber-attacks and manipulation of the software cannot be fully prevented. Secondly, it should also be noted that only crashes involving vehicles which can be replaced by AVs can be prevented. A significant amount of crashes do not involve motorized vehicles and these crashes -for example single bicycle crashes- cannot be prevented by CAVs.

LEVITATE has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 824361.

Pages
71
Publisher
European Commission, Brussels

SWOV publication

This is a publication by SWOV, or that SWOV has contributed to.