Driving Automation and Vulnerable Road Users: Peanut Butter & Jelly or Oil & Water?

Jesse Eisert

The Federal Highway Administration (FHWA) prioritizes safety for all road users, which is why FHWA is committed to Vision Zero and its goal of zero roadway deaths. To achieve this goal, FHWA has adopted a Safe System Approach for the Nation’s transportation network (Abel, Lindley, & Paniati 2021). The Safe System Approach takes a holistic view of the entire transportation network rather than a silo approach of one part of the system. Two key elements of the Safe System Approach are safe road users and safe roads (Abel, Lindley, & Paniati 2021). With these two elements in mind, researchers completed a literature review and gap analysis on the unique challenges and potential infrastructure-based solutions for various levels of automation interacting with vulnerable road users (Weaver et al. 2022).

One area of concern highlighted by this research effort is the limited ability of driving automation systems (DAS) to detect vulnerable road users. Currently, vulnerable road users are often obscured from view, particularly in dense rural environments. However, infrastructure can potentially overcome this limitation. FHWA has several initiatives and projects examining how to implement various technologies at intersections to improve the detection of vulnerable road users and disseminate that information to other road users, including DAS.

One challenge in developing and assessing these emerging intersection safety systems is testing them safely and meaningfully. At the Turner-Fairbank Highway Research Center (TFHRC), FHWA has developed a vulnerable road user test bed, including fully controllable intersections and a midblock crossing that can be used with pedestrian and bicyclist dummies and instrumented field research vehicles. TFHRC also houses an advanced highway driving simulator (HDS) and a virtual reality (VR) laboratory as part of its Human Factors Laboratory. FHWA staff at TFHRC are currently working on integrating the CARMA℠ ecosystem into the HDS (FHWA, 2023). The CARMA ecosystem is a collection of open-source software for cooperative driving automation research developed by FHWA. Additionally, FHWA is developing a connected distributed simulation platform that will allow the HDS and VR laboratory to operate in the same simulated scenario. These new and emerging capabilities will allow FHWA to safely conduct research supporting the Vision Zero goal (Abel, Lindley, & Paniati 2021) and ensure interactions between vulnerable road users and automated vehicles go together like peanut butter and jelly rather than oil and water.

Laura Sandt

There is a wide spectrum of road users who are made vulnerable to injury by their close proximity to larger, faster moving vehicles. This includes older pedestrians, children, people using canes and other assistive mobility devices, people with cognitive and sensory disabilities, people using micromobility devices, as well as often marginalized groups such as low-income populations, women, Black, Indigenous, and people of color who are more likely to be walking, bicycling, and using transit. Pedestrian and bicyclist fatalities represent a growing proportion of all road deaths, and the role of automated technology in these events is rarely established. At the same time, innovations in automated and connected hardware, software, remote application technologies, data systems, and integration techniques have combined to usher in many new forms of travel, including automated delivery bots, various forms of automated shuttles, mini-vehicles, buses/trucks, and other devices. As a variety of connected and automated innovations spread, agencies are quickly seeing the need to better understand how these technologies and related systems interact with and affect people walking and bicycling, and how to understand, plan for, and evaluate multimodal safety, access, and equity. In this panel, we will discuss how AV research is accounting for bicycle and pedestrian user behaviors and safety; how current AV deployments affect bicycle and pedestrian behaviors and other safety and equity outcomes; what realistic scenarios can be expected with respect to AV and nonmotorized road users interactions; and what research is needed to enhance bicycle and pedestrian safety and mobility.

David Kidd

Crash avoidance technologies like automatic emergency braking (AEB) are foundational technologies of Automated Driving Systems. The struggles that crash avoidance technologies face today provide insight into the challenges that automated driving systems must overcome. Existing crash avoidance technologies struggle to detect and respond to vulnerable road users like pedestrians and bicyclists, but the potential safety benefits are immense. Pedestrian and bicyclist motor vehicle crash deaths accounted for about 20 percent of all crash deaths in 2020.

Fatal pedestrian and bicyclist vehicle crashes are overrepresented in dark conditions. A recent Insurance Institute for Highway Safety (IIHS) study found that pedestrian AEB systems reduce pedestrian crash risk by 32% during daylight, but these safety benefits were not found for dark, unlit conditions. Lack of color in dark conditions and confusing vulnerable road users with roadside furniture make feature extraction and object recognition by cameras and radars a challenge.

Vehicles are turning in a large proportion of police-reported crashes with pedestrians and bicyclists. This crash configuration poses multiple sensor challenges for advanced vehicle technologies too. Forward-looking sensors may have a limited field-of-view that is not wide enough to identify and track pedestrians and bicyclists crossing roadways before and during a turn. Pedestrians and bicyclist movements can be sudden and unpredictable which may befuddle path planning algorithms.

Vehicle testing programs play a key role in highlighting opportunities for improving crash avoidance technologies and motivating manufacturers to make improvements. The IIHS recently extended its daytime pedestrian AEB evaluation to dark conditions to encourage manufacturers to overcome the technical challenges of dark conditions. Early testing results from 12 vehicles illustrated how most pedestrian AEB systems that performed well during daylight struggled at night. Multiple vehicle testing organizations started evaluating AEB systems in turning configurations with vehicles and vulnerable road users which will improve AEB performance at intersections. Motivating manufacturers to develop crash avoidance technology that reliably detects pedestrians, bicyclists, and other vulnerable road users in the most common crash circumstances will not only save lives but facilitate safe interactions between vulnerable road users and Automated Driving Systems.

Chris Monk

Advanced driver assistance system (ADAS) technologies, including automatic emergency braking (AEB), forward collision warning, lane departure warning, and SAE Level 2 automation capabilities, continue to achieve greater vehicle market penetration each year. The National Highway Traffic Safety Administration (NHTSA) has introduced guidance documents to define performance metrics and testing methodologies to help ensure these systems have safety benefits while not impeding their technology development. However, as with many industries that see rapid technological development driven by consumer demand, it takes time for regulation to catch up, especially given the diversity of ADAS implementation strategies. U.S. regulatory efforts are accelerating; NHTSA released a request for comment in 2022 to update NCAP test and evaluation procedures to adopt additional ADAS, reflecting broader consensus on the effectiveness of certain ADAS technologies. In addition, the recent Bipartisan Infrastructure Law mandates that NHTSA complete new regulations requiring AEB, lane departure warning, lane keeping assist, and forward collision warning on passenger vehicles, and AEB on heavy vehicles as well. However, other ADAS technologies, including those addressing vulnerable road users like pedestrian AEB, are still evolving. The regulatory landscape for ADAS and future automated vehicles is dynamic, and NHTSA has utilized a range of approaches to facilitate deployment, including agreements with automakers to voluntarily equip all vehicles AEB systems by 2022. While the European Union has mandated certain ADAS technologies on passenger vehicles starting in 2022, NHTSA is in the process of initiating rulemaking activities on the installation and performance of ADAS technologies.

Bahar Dadashova

AVs have the potential of improving highway safety, since they may help to overcome the driver error that contributes to 94% of crashes (NHTSA, 2017). Our 2021 study (Mousavi et al., 2021) found that the AVs have the potential to improve safety at unsignalized intersections as well at driveways. Because majority of bicyclist and pedestrian fatalities occur at these locations (NTSB, 2019), AVs may seem to have a great potential of decreasing vulnerable road user fatalities. However, due to limited data on AV and VRU crashes, conducting predictive analysis is not yet feasible. A study by Kutela et al., (2022) used text mining techniques to assess the AV crashes involving VRUs. Based on the limited crash data, it was found that the VRU crashes are likely to occur when the AV is in autonomous mode. Due to the gaps in the detection of vulnerable users by AVs, the full-extend of their expected implications for VRU safety are not clear. VRUs pose one of the most complex problems that AVs face: whether AVs can detect, classify, and predict the intention of bicyclists in real-time is one of the primary concerns (Dai and Dadashova, 2021). A study by Combs et al (2019) found that current automated driving system (ADS) technologies vary considerably in detecting and avoiding the pedestrian fatalities from less than 30% to over 90%. There have also been other investigations on how AVs can be configured and improved to ensure pedestrian and bicyclist safety, including developing a simulation model for vehicle-pedestrian/bicyclist crash-testing scenarios (Chen et al., 2018). However, VRU safety is still not at the forefront of AV discussions and research. As the number of bicyclist, pedestrian and e-scooter users increase every year, there is a need to address the potential implications of AVs on vulnerable road user safety. More research is needed to 1) identify ADS and ADAS targeting VRUs; 2) assess their effectiveness and failure risks; 3) identify the gaps and limitations of AVs in sensing and detecting the VRUs; and 4) propose new and upgraded technologies for improving communication and interaction between AV and VRUs.

Charlie Klauer

With the development of SAE Level 4+ Automated Vehicles (L4+ AVs), concerns have been raised with regard to L4+ AV and pedestrian interactions. How will pedestrians know and/or understand the intentions of the L4+ vehicle as it approaches intersections, parks to pickup/drop off passengers, etc. This concern has led to the development of broad variety of external vehicle communication systems. Most research on these types of vehicles has been conducted using one vehicle and/or in a virtual environment. However, it is vital to understand how pedestrians will communicate with multiple L4+ vehicles in a real-world environment. Our research examined how multiple L4+ AVs can best communicate their intentions (e.g., turning, stopping, yielding) to VRUs on a controlled test track. Participants without any prior knowledge of the communication systems experienced different external communication displays in a variety of common traffic scenarios. Our research suggested that participants were overwhelmed when multiple vehicles with external communication displays were in their vicinity and found it challenging to prioritize attention. These results suggest that training may be necessary for pedestrians, given that many of the participants did not understand the displays after multiple exposures. Additionally, these displays may be more useful in some traffic settings than in others.