The National Highway Traffic Safety Administration (NHTSA) estimates that in 2022 alone, 7,522 pedestrians and 1,105 cyclists will be killed in traffic accidents. However, detailed data on incidents involving such VRU populations is scarce. Many road accidents involving VRUs are not reported to the police or insurance companies. In addition, the United States has limited data collection on injuries in traffic accidents involving VRUs compared to collisions between vehicles. Even large traffic accident databases lack sufficient information about VRU crashes, which is important to fully understand and mitigate the associated injury risks.

Over the years, Waymo has conducted studies analyzing real-world collisions between VRUs and human drivers to gain more insight into the risks VRUs face and explore ways to improve their safety. Waymo’s research has covered a wide variety of scenarios, from reconstructing bike crashes in dense urban areas to studying injuries caused by door invades into bike lanes. In addition, Waymo has studied the movement behavior of micro-vehicles such as scooters and unicycles; The distribution of pedestrian injury severity was evaluated. The attitude of VRU before collision and the consequences after collision are described. An injury risk model for pedestrians and cyclists was constructed.

In Waymo’s latest study, in collaboration with Nexar, it reconstructed and analyzed 335 crashes captured by dashcams involving VRUs (pedestrians, cyclists, and motorcyclists) and human-driven vehicles in six major U.S. cities. This anonymised dataset is the largest traceable natural driving dataset in the United States, providing unprecedented insights into real-world traffic accident scenarios. With Nexar’s more than 5 billion miles of driving data, Waymo is able to capture driving events and environmental conditions to provide a more comprehensive view of the VRU than ever before.

In the process of reconstructing crashes, Waymo analyzed human-driven vehicles, VRU speeds, and collision configurations to study the estimated risk of injury in each accident. The analysis showed that human driver response to unexpected VRU behavior, lack of visibility, and vehicle trajectory (e.g., whether the vehicle is going straight or turning) were associated with collision speed and injury risk estimates.

Waymo worked with research partner VUFO to develop the model for injury risk assessment. This study helps to fill the current gap in the field of risk modeling for VRU injury risk modeling. Existing models are not only limited in number, but are generally limited to associating vehicle speed or approach speed with injury outcomes, ignoring other factors such as the VRU’s age, gender, and body size relative to the vehicle. In addition, Waymo uses anonymized data from the German in-Depth Accident Study (GIDAS), which contains information on thousands of VRU crashes over 20 years and is the world’s strongest data for estimating VRU injury risk associated with vehicle collisions.

Waymo’s research could help provide a more complete picture of the frequency and severity of crashes occurring in VRUs, rather than simply counting the number of crashes or assessing the risk individually. Autonomous vehicle manufacturers can use such data to test the performance of autonomous driving systems in a simulated environment before deploying them and assess their safety impact compared to human drivers. In addition, the results of this study will guide the continued development of Waymo Driver to further improve its safety when using the road with pedestrians and other road users.