Study: Are autonomous cars safer to drive than humans?

The debate is as old as the technology itself: Are cars that are driver-assisted or autonomous safer on the road than those with a human at the wheel? A research team has looked at accidents and concludes: it depends.

Mohamed Abdel-Aty and Shengxuan Ding from the University of Central Florida examined the documentation of 35,113 accidents in which a human was driving and 2,100 accidents involving cars controlled by advanced driving systems (Advanced Driving Systems and Advanced Driver Assistance Systems). The data sets included accidents caused by humans and those caused by fully autonomous driving systems (ADS, autonomous driving without restrictions such as external influences) and assisted driving systems (ADAS). The researchers had data from vehicles that were driving at level 2 (partially automated driving, such as lane keeping and cruise control) and level 4 (fully automated driving, such as autonomous driving in good weather). These levels are steps on the SAE scale, which divides autonomous driving into six categories, from unassisted (Level 0) to fully automated (Level 5).

Humans perform better at dusk and when turning

The team wanted to find out how the accidents differed depending on who was driving the car. The researchers concluded that vehicles with advanced autonomous driving are less likely to cause an accident than human-driven cars in the same scenario.

At the same time, the technically controlled cars performed worse at dusk or when turning than those driven by a human: under these conditions, the probability of an accident was 5.25 times (dusk) or 1.98 times (turning) higher. "Possible reasons for this could be a lack of situational awareness in complex driving scenarios and the limited driving experience of AVs (autonomous vehicles)," according to the research report published by the team in the journal Nature Communications.

The researchers considered the type of accident, the road and surroundings, the conditions before the accident such as vehicle manufacturer, AV driving mode and direction of movement and the course of the accident to ensure comparability. The team also considered the day of the week or the time of day, as these typically influence the risk of an accident. The scientists believe that the generally better performance of automated vehicles is due to the vehicles' object recognition and avoidance, precise control and better decision-making.

External conditions make the difference between autonomous and assisted systems

Depending on the external conditions, however, autonomous vehicles (ADV) also differed from assisted driving systems (ADAS). For example, ADAS recorded 23.34 percent fewer accidents under clear skies. In the rain, however, they recorded 13.65 percent more than ADS. If the vehicles were driving straight ahead before the accident, the ADAS were involved in 27.91 percent more accidents than the autonomous vehicles. The latter fared worse when turning: ADAS were involved in three percent fewer accidents than ADS. When analyzing the movements before the accident, ADAS show a 27.91 percent higher number of accidents when driving straight ahead, while they report 3 percent fewer accidents when turning than ADS.

The researchers hope that their findings will provide important insights for the further development of autonomous vehicles. The results of the latest systems may be different from those of the study. For further research, it is important to include data on priority regulations at intersections, stop signs, traffic lights and priority signals to improve the comparison.

(are)