Human error is responsible for virtually all auto accidents. In fact, a survey of police-reported crashes in the U.S. found that driver error ultimately leads to more than nine out of every 10 collisions on the road.

It’s widely believed that by eliminating human error with self-driving cars, crashes would eventually become a thing of the past. However, new research from the Insurance Institute of Highway Safety (IIHS) suggests that autonomous vehicles (AV) aren’t yet living up to their promise for safer driving.

While self-driving cars will likely soon identify hazards on the road better than people, researchers found that this alone will not prevent a majority of accidents. Only about one-third of crashes on the road are the result of a mistake that an AV would be capable of avoiding. To prevent the other two-thirds of accidents, AVs need specific programming to avoid other uniquely human driving errors.

What driver errors contribute to accidents?

After examining over 5,000 police-reported crashes, the IIHS broke up the driver-related factors that contributed to the accidents into five distinct categories:

  1. Sensing and perceiving errors – These errors can include driving distracted, impaired visibility or failing to see a road hazard before it was too late.
  2. Incapacitation errors – Any errors involving impairment from drugs, alcohol, medical conditions or falling asleep at the wheel.
  3. Predicting errors – Includes mistakes in which a driver makes an incorrect assumption about how fast another driving is going or what another driver is going to do.
  4. Planning and deciding errors – These driving errors cover driving too fast or slow for the road conditions, driving aggressively or tailgating a vehicle in front of you.
  5. Execution and performance errors – Includes errors such as inadequate or incorrect evasive maneuvers, over-correcting on the road and other issues controlling the vehicle itself.

If all cars on the road were self-driving, the researchers believe that only the first two categories – sensing and perceiving and incapacitation errors – would be prevented by AV technology. These two categories alone account for about 34% of the total crashes on the road. AVs will need the ability to avoid prediction, performance, and decision-making errors to prevent the other 66% of accidents and drive better than humans.

The bottom line

Until AVs can account for all mistakes made on the road, they’ll likely only prevent about one-third of car accidents. There may come a day where self-driving cars end collisions for good, but science still has a long way to go.