US study finds autonomous vehicle systems can eliminate only one-third of accidents caused by driver error

Computers are better at spotting hazards and don’t get distracted, but making autonomous cars better than human drivers remains a ‘huge challenge’.

A Rideflux self-driving car in South Korea. Picture: AFP

Self-driving systems will fail to prevent most car accidents despite being lauded as the holy grail of road safety, according to the latest US research.

The study, by the Insurance Institute for Highway Safety, shows that while human error plays a role in almost all crashes self-driving cars will struggle to avoid the same mistakes.

Cameras and sensors can spot hazards better than human drivers and are not vulnerable to distraction.

But making cars that can drive as well as people remains a huge challenge, the IIHS says, and autonomous cars would have to drive even better to make accidents a thing of the past.

“It’s likely that fully self-driving cars will eventually identify hazards better than people, but we found that this alone would not prevent the bulk of crashes,” says study coauthor Jessica Cicchino.

The analysis, of 5000 crashes in the US, finds driver error is involved in 90 per cent.

However, about two-thirds involved mistakes unrelated to simple perception or distraction.

Other factors included faulty predictions, such as misjudging a gap in traffic or what another road user was going to do.

Decision errors included driving at an inappropriate speed for the conditions, or following too close to a vehicle in front.

Some accidents were unavoidable due to mechanical failure of tyre blow-out.

“Crashes due to only sensing and perceiving errors accounted for 24 per cent of the total, and incapacitation accounted for 10 per cent,” says the study.

“Those crashes might be avoided if all vehicles on the road were self-driving — though it would require sensors that worked perfectly and systems that never malfunctioned.

“The remaining two-thirds might still occur unless autonomous vehicles are also specifically programmed to avoid other types of predicting, decision-making and performance errors.”

Self-driving systems would need to become much better at reading the uncertain behaviour of other road users, such as in busy pedestrian areas, and adapting to difficult driving conditions, such as bad weather.

To eliminate accidents altogether, they would have to prioritise safety above every other consideration and as a result, would drive much more slowly than humans.