Self-driving cars being tested in real-world traffic have a higher crash rate than conventional vehicles, according to a new study by the University of Michigan Transportation Research Institute.
However, the study also concluded that data are still limited, so it cannot rule out the possibility that actual accident rates for self-driving cars could ultimately be determined to be lower than those for traditional vehicles.
"Once these cars accumulate more miles, we'll get a more complete picture of crash rates," said Brandon Schoettle, a project manager at the research institute. "So far, they appear to be more likely to be in accidents, but remember those crashes were not their fault."
The study, conducted by Schoettle and Michael Sivak, analyzed records from three self-driving vehicle programs: Google, Delphi and Audi. Those programs, along with others from the likes of Tesla, and BMW, are required to report any accidents that happen while driving in California.
The crash reports, along with other public data released by Google, Delphi and Audi, were then compared to the safety records for all conventional vehicles in the U.S. in 2013.
Since 2012, the self-driving cars tested by Google, Delphi and Audi have logged more than a million miles on public roads, and have been involved in 11 accidents. By comparison, researchers say there were 5.7 million police-reported crashes involving all vehicles in the U.S. in 2013.
After crunching the data, Schoettle and Sivak concluded there's an average of 9.1 crashes involving self-driving vehicles per million miles traveled. That's more than double the rate of 4.1 crashes per one million miles involving conventional vehicles.
Because the National Highway Transportation Safety Administration estimates that 60 percent of property damage-only crashes and 24 percent of injury-related crashes go unreported in the U.S. every year, the researchers made a statistical adjustment to the conventional vehicle crash data to adjust for underreporting.
"The good news here is that these self-driving cars have avoided more serious accidents like a head-on collision," Schoettle said. "So they may be more likely to be in an accident, but the accidents have not been as bad as others."
What's more, not once was the self-driving vehicle the cause of the accident. In eight of the 11 crashes, the self-driving cars were either stopped or going less than 5 miles per hour when they were hit by another vehicle, according to the researchers.
Chris Urmson, head of Google's self-driving car program, has talked at length about the challenges autonomous-drive vehicles face by sharing the road with conventional cars and trucks.
While computers in the Google prototype can steer through traffic based on what's happening in front of them or in the lanes on either side, they are unable to avoid being rear-ended by a car approaching at a stop sign or intersection.
In September, Urmson declined to predict when self-driving cars would be out on roads in larger numbers.
"We don't really have a set timeline because first and foremost this is really about safety, and we don't want to rush that," he said at the time. "It has been amazing to see how the technology has developed over the last few years."
Delphi spokeswoman Amy Messano said that although the company has not yet seen the results of UMTRI's study, "Delphi's self-driving prototype has not been involved in any accidents while testing its autonomous-drive technology."
Google says its self-driving cars have caused exactly zero collisions in the 1.2 million miles of autonomous driving they've done since the project started in 2009.
"We publish the details of all crashes we've been involved in on our website each month, and there's a clear theme of human error and inattention," a Google spokesman told CNBC. "The researchers themselves concede in their report that the actual crash rates of for self-driving vehicles could be lower than for conventional vehicles."
Audi did not immediately respond to a request for comment.
Questions? Comments? BehindTheWheel@cnbc.com.