Waymo Compiles the Largest Dataset on Pedestrian and Cyclist Injuries.
There have been few studies on road users who are in vulnerable situations.
The integration of autonomous vehicles into society requires special attention to the safety of all road users, especially those who are more vulnerable, such as pedestrians, cyclists, and motorcyclists. Every year, thousands of these individuals lose their lives in traffic accidents. Recently, Waymo, Alphabet's robotaxi company, released a study that examines hundreds of incidents involving these users, which has been dubbed "the largest dataset of its kind in the U.S."
Approximately 40,000 people die annually in the country due to vehicle collisions, and although automakers have significantly improved safety measures for those inside vehicles, they have neglected the protection of those outside. Furthermore, the academic community has shown little interest in researching the injuries suffered by vulnerable road users, prompting Waymo to focus its efforts on this aspect, emphasized John Scanlon, a safety researcher at the company. The goal was to shed light on a poorly studied area of traffic research, hoping that the findings would help make their autonomous driving technology safer and even benefit competitors.
The new analysis comes in an alarming context, where violence against pedestrians and cyclists in the U.S. remains high. In 2022, there were 7,522 reported pedestrian deaths and more than 67,000 injuries in accidents, according to the National Highway Traffic Safety Administration. “It is essential to have a precise and in-depth understanding of the specific risks facing these groups to develop effective strategies to reduce injuries and fatalities,” Scanlon stated.
To conduct their study, Scanlon's team needed footage of hundreds of car accidents, leading them to collaborate with the dashcam company Nexar. Using anonymized data from 500 million miles of driving captured by Nexar, they were able to reconstruct 335 collisions involving vulnerable users in six U.S. cities. However, the analysis showed that 80% of the incidents occurred in New York. The anonymous individuals in the dataset suffered injuries ranging from moderate to severe, although fortunately none of the accidents studied resulted in fatalities. This work is presented as “the largest documented dataset on naturalistic driving in the U.S.,” according to the company.
By analyzing this information, Waymo aims to better understand how, when, and why vulnerable users are injured by vehicle drivers. By focusing on the "frequency and severity" of collisions, they were able to draw relevant conclusions. On one hand, it is evident that pedestrians and cyclists are more likely to be injured when they "surprise" drivers, such as when attempting to cross the street against a red light. Additionally, "geometric occlusions," such as trees or buildings, increase the risk of injuries, as does the trajectory of the vehicle.
To further enhance risk assessment, Waymo partnered with VUFO, a traffic research group in Germany, and used anonymized data from the German Accident Study, which covers over 20 years of information on accidents involving vulnerable users, considered as "the most relevant data available in today's world" for assessing injury risk.
Waymo's autonomous vehicles operate in cities like San Francisco, Los Angeles, and Phoenix, making over 150,000 paid trips per week, and they plan to expand their service to Austin and Atlanta. Each day, these vehicles must navigate environments with numerous vulnerable users. A single mistake could have tragic consequences. In the past, there have been reports of incidents where autonomous vehicles were involved in collisions, such as when a cyclist was injured in San Francisco after coming out from behind a truck that obstructed visibility. Following these events, the company faces ongoing challenges.
Scanlon mentioned that a better understanding of these types of collisions could allow autonomous vehicle operators to recreate these incidents in both simulations and real-world tests, which could lead to safer decisions. “This analysis can serve as a starting point for identifying the driving risks associated with collisions involving vulnerable users in dense urban areas, which in turn will facilitate testing and performance assessments of autonomous vehicles,” he concluded.
