Recommended metrics for traffic interactions - D2.5 & D2.14 Summary
Work package 2 focuses on improving simulation tools to understand the impact of automated vehicles. In order to do so, metrics are being developed to identify safety-critical interactions and assess their severity. Typically, critical interactions are identified by applying specific thresholds to the measures of vehicle separation, such as space, time, or speed. While Deliverable 2.5 reviewed the literature on the topic and identified the further development necessary for the project, Deliverable 2.14 presents the defined metrics and their use.
In the context of this research, an interaction occurs when two or more road users have to adapt their behaviours to avoid occupying the same space at the same time. An interaction becomes critical if a severe intervention is necessary to prevent a collision. SAFE-UP Deliverable 2.5 uses the term "severe traffic conflict" to describe a situation where two or more traffic participants come into conflict and may need to take evasive action to avoid a collision. Critical interactions are more severe than severe traffic conflicts and require more urgent action to avoid a collision.
The metrics used to quantify the severity of traffic interactions are called surrogate measures of safety (SMoS) since they act as “surrogates” for the probability of collisions. There are many SMoS available in the literature with different threshold values, but there is no consensus on what exact value to use. The recommended practice is to use a range for the threshold and a set of different metrics, but sometimes the SMoS can contradict each other and show different levels of severity. Another problem is that some metrics can only be used in certain situations, such as Time-To-Collision, which can only be applied when the following vehicle is faster than the leader. Some solutions to improve these metrics include developing a threshold-independent method or deriving a metric that takes into account multiple metrics to avoid contradiction.
The criticality of a driving interaction depends on the type of participants involved. For instance, a critical interaction between a car and a motorcycle may not be critical between two cars, or a car-pedestrian interaction may be non-critical to the vehicle, but critical to the pedestrian. Thus, the three different perspectives of motor vehicles, powered two-wheelers (PTW), and vulnerable road users (VRU) need to be considered.
From the perspective of cars, the partners developed three metrics/approaches to identify safety-critical scenarios. Two of the metrics were developed by TNO, who used naturalistic driving data to develop and validate unsupervised anomaly detection and driver profiling. The first identifies abnormal driving situations in a vehicle following to differentiate between safe and unsafe behaviour, while the second one creates clusters of driver behaviour based on how safe their interactions are in general. TUD named their approach Probabilistic Driving Risk Field (PDRF) because it takes into account the probability of motion of each vehicle to identify the severity of crashes and the probability of collision.
From the perspective of PTWs, specific SMoS are necessary for a complete assessment of the interaction in mixed traffic, since they have usually been assessed as cars in traffic simulation. UNI adapted the Car Potential index defined in (Cunto & Saccomanno, 2008) to create the Motorcycle Collision Potential index (MCPI), considering a range of metrics for different PTW performances. They also used unsupervised classification to identify the safety-critical breaking events.
From the perspective of VRUs, IKA identified that psychological and subjective aspects are as important as the technical factors to identify (un)safe behaviour. In that sense, existing car metrics were used and adapted for pedestrians and cyclists. The deceleration to safety time (DST), Post encroachment time (PET) and Time Head Way (THW) were picked and adapted for pedestrian interactions and considered the pedestrian as an object on the road or crossing the road. For cyclists, a long list of metrics was identified and adapted to cover both longitudinal and lateral collisions, like Time To Collision (TTC) or Lateral Time To Collision (LTTC), for example.
Not all the defined metrics are mentioned in this summary, but the full list can be found in D2.14. All the developed metrics will be used in T2.5’s simulation to assess the severity of interactions between traffic participants.
The SAFE-UP simulation environment consists of behavioural models of different road users (human-driven car, autonomous car, pedestrian, cyclist and PTW-rider), developed by the partners and integrated into Aimsun Next.
Check out our full list of public deliverables on our Resources page.
Want to get in touch with the SAFE-UP team? Email us at: contact@safe-up.eu
