Driver errors play a role in just about every motor vehicle crash. That’s why some are hopeful that automated, so-called “self-driving” vehicles will enhance safety for everyone on the road.
However, a new study by the Insurance Institute for Highway Safety (IIHS) shows that automated systems may only reduce crashes by about a third.
Driver-related crash factors
Some have predicted that self-driving vehicles could make car crashes a thing of the past. But IIHS research says that’s unlikely. The study focused on five categories of driver-related errors contributing to crashes:
- Sensing and perceiving: These mistakes result from distracted driving, poor visibility and failing to recognize hazards in time to react.
- Predicting: Misjudging gaps in traffic, incorrectly gauging another vehicle’s speed or incorrectly assuming what other drivers do.
- Planning and deciding: These mistakes include going too slow or too fast for road conditions, failing to maintain enough space from the vehicle ahead or driving aggressively.
- Execution and performance: Making incorrect or inadequate evasive maneuvers, overcompensating or making other errors in controlling the vehicle.
- Incapacitation: Alcohol or drug use, medical problems or falling asleep behind the wheel cause many crashes.
The IIHS study noted that some crashes are unavoidable, such as when tires blowout, axles break, or other vehicle failures occur.
Results show strengths and weaknesses
According to the IIHS, automated vehicles will likely prevent more crashes caused by perception errors or incapacitated drivers. That’s because cameras and sensors are able to identify potential hazards faster and better than humans.
However, sensing and perceiving errors account for only 23% of all accidents, while incapacitation caused another 10%. The remaining two-thirds of all accident causes can only be avoided if automated vehicles are programmed to avoid prediction, performance and decision-making mistakes. Unfortunately, the study researchers concluded that the current technology has not yet reached that level to avoid prediction, performance and decision-making errors.