Study Finds Humans May Outperform Autonomous Driving in Some Aspects

TapTechNews June 26 - A study published this week in Nature Communications finds that humans may perform better in the field of autonomous driving than we think. The research analyzed data from 2,100 self-driving car accidents and 35,133 human-driven car accidents between 2016 and 2022, and found that human drivers are less likely to have accidents in dim light and on turning sections compared to autonomous driving systems.

The study report concludes that the probability of accidents for vehicles using autonomous driving systems at sunrise or sunset is five times that of human drivers, and the accident rate on turning sections is twice that of human drivers.

The research highlights the limitations of cameras and sensors in low light environments and their inability to adapt to complex weather. For example, shadows in early morning or evening are easily mistaken for objects, and the constantly changing light can also affect the signal processing of sensors, causing confusion in the system algorithms and inability to accurately identify obstacles. In severe cases, the system may even completely fail to detect objects in the shadows. This is also supported by crash tests, in which vehicles in autonomous driving mode often brake too late or even fail to brake at all when encountering simulated pedestrians or animals.

For turning sections, the research report points out that current autonomous driving systems may have flaws in situation awareness. Sensors and cameras may not be able to detect all obstacles in a dynamic environment such as an intersection, but the problem doesn't stop there. Research shows that current systems usually only see the relatively close area around the vehicle. A human driver can see thick fog half a mile ahead and take preventive measures, while a car in autonomous driving mode will just keep moving forward.

The researchers further supported this finding by analyzing the behavior of vehicles before the collision. Before the accident, most autonomous vehicles maintained a straight and uniform driving state until an emergency braking was required. In contrast, human drivers will take more actions such as deceleration and lane changing before the collision.

This study considered a large number of variables to draw a conclusion, but the result is clear: the current driving assistance systems at this stage are only assistance systems. TapTechNews needs to point out that since the data collection time is from 2016 to 2022, and now the autonomous driving technology has also made great progress, and specific comparisons with human drivers need updated research.