While there's been a lot of hype about the number of miles Google's autonomous vehicles have traveled without accident, it seems the cars aren't as good as humans yet. According to data Google recently shared with the California DMV, engineers assumed control of the self-driving car 341 times between September 2014 and November 2015.
Given that the fleet covered 423,000 miles in that time, this means there was about one intervention roughly every 1,240 miles. Most of these interventions, 272 of them, were the result of hardware failures. Wired spoke with Google and got to hear that they do not find this terribly troubling, as hardware reliability isn't the key concern at this phase in the development cycle:
Of the 341 instances where engineers took the wheel, 272 stemmed from the “overall stability of the autonomous driving system”—things like communication and system failures. In other words, something broke. Chris Urmson, the project’s technical lead, doesn’t find this terribly troubling, because hardware is not where Google is focusing its energy right now. The team’s more concerned with how the car makes decisions, and will make the software and hardware more robust before going to market.
Where it gets more interesting are the remainig 69 takeovers, these were related to safe operation of the vehicle. Basically, if the engineer in the car isn't fully confident that the AI will do the right thing, he or she will take control of the vehicle. Data from these interventions gets analyzed at Google's HQ, all of the car's data gets uploaded to a computer that makes a simulation of what the car would have done without human intervention.
It turns out that in 13 of those 69 incidents, the self-driving car would have led to a crash. Google often mentions that since 2009, its autonomous vehicles have driven 1.3 million miles, there have been 17 crashes but the self-driving car has never been at fault.
It now turns out that the crash-free operation of Google's self-driving car is the result of numerous interventions by the engineers. Without these interventions, the Google cars would have had about one accident every 74,000 miles, which is significantly worse than the US average of one crash per 238,000 miles.
Without a doubt, self-driving cars are the future, but at present they're still not as good as human drivers.