According to recent reports, Waymo is leading the push for autonomous vehicles by a large margin but that lead may not be enough to ensure safety in all conditions. Of course, that’s not for a lack of effort on the part of the Alphabet-owned company. The company has logged nearly twice as many public road test miles as its closest U.S. competitor across 25 cities and in a wide variety of driving conditions. Setting aside its nearly five billion virtual test miles that are still undertaken 24-hours a day for 10,000 virtual test vehicles, five million real-world tests have effectively covered the gamut. Those include everything from the sun-scorched roads of Arizona to rain and snow in San Francisco and Michigan. With that said, there may be a good reason to believe that all of those miles don’t really amount to much when it comes to comparing the viability of an A.I. driver over a human driver.
To begin with, human drivers encounter conditions that cause a fatal crash one time for every 100 million miles traveled by a vehicle. So, even at five million miles of real-world tests, Waymo has only hit around 5-percent of the miles needed to make that kind of a comparison. Potentially even worse, human drivers very rarely, if ever, simply stop driving while behind the wheel of a vehicle. Waymo’s latest figures on A.I. driver disengagement – the number of times a human driver needed to take control – fell in at around 0.18 per 100,000 miles. That essentially means that for every 556,000 miles driven autonomously, the system encountered something that caused it to disengage. In those instances, whether virtual or real-world, it would fall to a human driver to take over. Again, that’s much better than competitors, such as General Motors Cruise’s 0.8 disengagements over the same distance. However, any disengagement at all is potentially fatal if a car is completely autonomous, with either nobody in the driver’s seat or no way to manually control the vehicle included in the design.
Those are all potent factors to consider when looking at some automated services that are planned to undergo testing in other portions of the world. Even more so when considering that some of these are intended to show how A.I. drivers can provide transport for those who cannot drive, for whatever reason, including children. So it makes sense that Waymo’s own tests are planned for an area that doesn’t often experience inclement weather conditions and are generally limited. With any luck, the company will have learned valuable lessons and tweaked the algorithms enough to avoid potential problems, or at very least to circumvent the worst-case scenarios for passengers.