Are Autonomous Vehicles Safer Drivers Than Humans?
Autonomous vehicles may be susceptible to occasional crashes, but those crashes tend to be less severe and are usually the fault of human drivers running into them. A recent study found self-driving vehicles have lower crash rates than cars driven by humans. Researchers say there isn’t enough data to produce conclusive results but the data they do have shows computers may drive better than us.
The latest study, “Automated Vehicle Crash Rate Comparison Using Naturalistic Data,” was conducted by Virginia Tech Transportation Institute (VTTI) and commissioned by Google. Myra Blanco, director of VTTI’s Center for Automated Vehicle Systems, says the report compared data from the Google Self Driving car project with national crash data from the National Highway Traffic Safety Administration. It found that self-driving cars have a rate of 3.2 crashes per million miles compared to a crash rate of 4.2 million miles for driver-operated vehicles.
Because data from self-driving vehicles is limited, and because a large number of crashes go unreported, VTTI researchers had to extrapolate and standardize the data. “As a research study, it’s not perfect and there are limitations, but it’s a way to start looking at the trends,” says Blanco.
The results contrast to another study released in October 2015 by the University of Michigan. “A Preliminary Analysis of Real-World Crashes Involving Self-Driving Vehicles,” found that for every million miles driven, self-driving vehicles had 9.1 crashes, compared to only 1.9 crashes for manned vehicles. Brandon Schoettle, Project Manager at the Transportation Research Institute at the University of Michigan, says it doesn’t necessarily mean the vehicles are less safe because humans were to blame in almost all of those crashes.
Schoettle says most autonomous vehicle crashes were minor “bumps” resulting from a human driver rear-ending them at slow speeds. He says humans aren’t accustomed to how self-driving vehicles drive because they “conservatively err on the side of caution” in almost any situation. Autonomous vehicles typically drive slower, brake earlier, and accelerate at a slower pace. “They drive like you might say an old lady drives. They drive extra safe and most [human drivers] don’t anticipate it,” says Schoettle.
Authors of both studies caution there could be large margins of error and that it remains difficult to make comparisons with limited data on self-driving vehicles. Researchers at the University of Michigan only had 1.2 million miles of data for 50 such vehicles, compared to 3 trillion annual miles with 269 million traditional vehicles.
Authors of both studies say there are anomalies in the data as autonomous vehicles are required to report every accident in California no matter how small whereas according to a survey by the NHTSA, 30% of accidents go unreported.
David Alexander, senior analyst at Navigant Research, says self-driving vehicles have the potential to be safer than human drivers. Autonomous features, such as intuitive cruise control, lane departure warning, and blind spot detection, are already on the market and improve drivers’ response times and awareness. “These things make cars safer because they’re operating all the time and can eliminate some human errors,” he says.
Both studies revealed that self-driving vehicles may have lower rates of serious accidents. Schoettle says there hasn’t been an incident of an autonomous vehicle involved in a head-on collision or a collision with a pedestrian or cyclist. “These vehicles seem to be good at avoiding the most serious types of crashes,” he said.
A primary benefit of autonomous features is that computers can’t get tired, distracted, or careless, says Alexander. According to the National Highway Traffic Safety Administration, 94% of all crashes are attributed to human error. Computers can also, within nanoseconds, take precise measurements around speed, time, distance, breaking power, and turning power to instantly engage in the most optimal maneuvers. “Individuals suffer from all sorts of distractions, from weather, children, dogs, the radio, eating or texting. Technology doesn’t get distracted or careless,” says Alexander.
Insurance companies are already contemplating that self-driving vehicles may make the roads safer. PTOLEMUS Consulting Group forecasts in the Useage-Based Insurance (UBI) study that the penetration of autonomous vehicles by 2030 will reduce the number of accidents by 30% and lead to a significant reduction in insurance premiums for drivers.
PTOLEMUS research and marketing director Thomas Hallauer says the slow integration of autonomous features are already having an impact on safety. He says frontal collision avoidance and cruise systems could reduce losses by up to 50% and that “the crash-less vehicle will be on the road long before the driver-less one.”
One caveat is there has yet to be comprehensive testing for these vehicles operating in “real world conditions,” says Schoettle. They have not been extensively tested in accident-prone situations such as rush hour traffic in congested busy cities with lots of pedestrians and aggressive drivers, or in heavy downpours, high winds, and snowstorms. “Most of the miles have been in nice conditions in sunny parts of the country. There has been some testing but it hasn’t done well so far,” he says.
Another limitation is that computers can’t engage in human decision making and judgment calls, says Schoettle. Computers can’t make eye contact with other drivers or pedestrians to gauge their intentions. Self-driving vehicles can’t yet optimally perform at four-way stops, a person standing on side the road, another driver waving a vehicle into traffic, congested areas with a large number of pedestrians. “A person might be able to make eye contact and gauge a persons’ intentions but a self-driving vehicle can’t,” he says.
Alexander believes that self-driving vehicles by their nature will be safer because consumers and regulators will hold them to higher standards. Ongoing research will also help programmers and car manufacturers incrementally improve safety and teach self-driving vehicles to better respond to human drivers on the road. “These vehicles collect a lot of data and I think moving forward, ongoing testing and studies will continually help make them safer,” says Blanco.