GOOGLE says the computer brain inside its self-driving car is getting smarter fast, with the number of times human test drivers had to take over for the autonomous vehicle decreasing sharply in recent months.
In its latest mandatory autonomous-car report to the California Department of Motor Vehicles, and its first on the subject of “disengagements of autonomous mode”, Google reported its cars handed over control to the driver seven times less frequently in the fourth quarter of 2015 than in a similar period in 2014, USA Today reports.
“What’s encouraging is that the numbers are going down,” Chris Urmson, Google self-driving car project lead, said. He noted that of 13 incidents in which Google test drivers suddenly took control of the car, five happened over the course of 595,457.kilometres in 2015 — that’s down from eight within 85,295km of driving in 2014.
The disengagements, which Google reported in a blog post, were grouped into two categories: failure of the autonomous system due to unfamiliar road parameters, which alerted the driver who then took over control (272 incidents), and unanticipated human actions taken before the computer could react (69).
Urmson says that by using modeling software, Google’s self-driving car engineers were able to play out those 69 incidents in which the driver suddenly took over to see if there would have been any contact had the human not intervened. Contact would have occurred in 13 of the 69 events, either with another motorist or in some cases traffic cones.
Of the 272 instances in which the car’s computer alerted the driver that he or she needed to take the wheel, the gap between these events decreased in recent months to once every 8529 autonomous kilometres, as compared to once every 1263km a little over a year ago.
Urmson said that the Google car’s software currently is “set conservatively” for testing purposes, meaning it is designed to err on the side of being easily triggered so engineers can refine and upgrade the system.
Over the past six years, the company’s fleet of sensor-packed Lexus SUVs and prototype pod-cars have logged 2.09 million kilometres on public roads in Mountain View, California, and in Austin, Texas.
Self-driving car tech is the new buzz word for traditional automakers. Ford and Kia recently announced at the Consumer Electronics Show that they are beefing up their autonomous car plans, while Volvo, Audi, Tesla and Mercedes-Benz are quickly growing the number of driver-assist features that edge close to autonomous driving but require drivers to stay engaged in the process.
In all these instances, the automobiles will retain their steering wheels and pedals. In contrast, Google has made clear that its cars will be two-passenger machines will lack those features.
“We have always believed that fully self-driving is the right path,” Urmson says, noting that the more a car can handle driving chores the less alert its driver would be to the possibility to taking back the wheel.
“It is challenging to maintain vigilance if you’re untrained” in terms of taking back control from an autonomous car.
But Urmson doesn’t believe Google’s car will be alone on the roads one day, but rather just “one part of many solutions to a changing transportation landscape.”
If there is one thing that drives Urmson and his team, it is in fact statistics on deaths caused by automobiles. The news there isn’t good. US traffic deaths jumped 14 percent in the first half of 2015, according to the National Safety Council. Around 14,000 people lost their lives in that period.
“That’s the equivalent of a 737 full of passengers falling out of the sky every weekday, yet we seem to accept (that) as the price of our mobility,” Urmson wrote in his blog post.
“Self-driving cars have the potential to reduce those numbers, because they eliminate the driver inattention that leads to thousands of collisions, injuries and deaths.”