Study Finds Humans Outperform Automated Technology When Driving at Twilight

The study also revealed that human drivers excel at navigating intersections.

Driver-assist systems with partial automation can help prevent collisions, but they aren’t quite ready for widespread use. A recent study, which analyzed over 37,000 vehicle collisions, found that humans are less likely to crash during turns or in low-light conditions compared to automated technology—and the differences are significant.

Published in Nature Communications, the study revealed that vehicles using automated systems experienced five times as many crashes during sunrise or sunset. In corners, automated systems had twice as many crashes as human-driven vehicles. The data is based on approximately 35,000 crashes involving human drivers and 2,100 with active automated systems.

In low-light conditions, the study points out the limitations of cameras and sensors, which struggle to adapt to changing conditions. For instance, early-morning or late-day shadows can be misinterpreted as objects, and fluctuating light can confuse algorithms, leading to system errors. Objects in shadows may not be detected at all, as supported by crash tests showing vehicles braking late or failing to stop for simulated pedestrians or animals.

Regarding turns, situational awareness is identified as a significant challenge for current automated systems. Sensors and cameras might not detect all obstacles in dynamic environments like intersections. The study highlights that current systems typically “see” only the area close to the vehicle. In contrast, a human driver might notice a fog bank half a mile away and take precautionary measures, while an autonomous car would continue at its current pace.

Evidence supports this claim. Analyzing actions taken before collisions, most vehicles under autonomous control were found to be driving straight and at a constant speed before emergency maneuvers were initiated. Conversely, human-driven cars often showed more instances of slowing down and changing lanes before impact.

The study considers numerous variables to reach these conclusions, but the message is clear. Current driver-assist systems are merely aids. While automation performs adequately in straightforward scenarios, significantly more data and research are needed before true hands-off/eyes-off Level 4 driving becomes feasible.

Author:

  • Ian Sawyer

    Growing up with a father who was a mechanic I had an appreciation for cars and motorcycles from an early age. I shared my first bike with my brother that had little more than a 40cc engine but it opened up a world of excitement for me, I was hooked. As I grew older I progressed onto bigger bikes and...

    View all posts