Self-Driving Cars Face Challenges in Specific Scenarios, Study Finds
A recent comprehensive accident study suggests that self-driving cars generally demonstrate better safety than human drivers in many situations. However, the research reveals that autonomous vehicles still struggle in specific scenarios, particularly during low-light conditions and when making turns.
A self-driving car navigating a city street.
The findings come at a critical time, as autonomous vehicles are actively operating in several cities across the US. The GM-owned company Cruise is attempting to resume driverless car testing following an incident in March that led to a temporary suspension of its operating permit in California. Meanwhile, Google spin-off Waymo continues to expand its robotaxi services in various cities, including Austin, Los Angeles, Phoenix, and San Francisco.
“It is important to improve the safety of autonomous vehicles under dawn and dusk or turning conditions,” says Shengxuan Ding at the University of Central Florida. “Key strategies include enhancing weather and lighting sensors and effectively integrating sensor data.”
Data Analysis Reveals Safety Disparities
Ding and his colleague Mohamed Abdel-Aty, also at the University of Central Florida, analyzed data from 2,100 accidents in California and from the National Highway Traffic Safety Administration (NHTSA) involving vehicles with some level of automated self-driving or driver assistance technologies. Additionally, they gathered data on over 35,000 accidents involving unassisted human drivers.
Using statistical matching, the researchers identified pairs of accidents that occurred under similar circumstances, controlling for factors like road conditions, weather, time of day, and location (intersection or straight road). The analysis specifically focused on 548 self-driving car crashes reported in California, excluding vehicles with only driver-assistance systems.
Overall, the results suggest autonomous vehicles “generally demonstrate better safety in most scenarios,” according to Abdel-Aty. However, the analysis also revealed that self-driving cars had a crash risk five times greater than human drivers during dawn and dusk conditions. Furthermore, they experienced almost double the accident rate of human drivers when making turning maneuvers.
Challenges and Limitations
One of the key obstacles to this research is the “autonomous vehicle accident database is still small and limited,” acknowledges Abdel-Aty. He and Ding emphasize the need for “enhanced autonomous vehicle accident reporting,” a critical point echoed by independent experts.
Missy Cummings, at George Mason University, describes the number of self-driving car crashes as “so low that no sweeping conclusions can be made.” She also cautions against potential biases in reporting from self-driving car companies, noting that video footage sometimes has contradicted the companies’ narratives, which often place blame on human drivers.
Eric Teoh from the Insurance Institute for Highway Safety points out that many minor accidents, or “fender benders,” may not be reported to police. Therefore, any comparative analysis of autonomous vehicle and human driver crashes must consider this factor. A 2017 study by Teoh on Google’s early self-driving car tests showed that only three out of ten specific crashes were documented in police reports.
Junfeng Zhao at Arizona State University adds, “Both California and NHTSA do not require comprehensive data reporting for autonomous vehicle testing and deployment.” Furthermore, he explains that, because robotaxis often operate in specific areas and environments, generalizing the findings can be difficult.
Journal Reference: Nature Communications DOI: 10.1038/s41467-024-48526-4
