Mountain View, California - Google's self-driving cars are mastering complex situations on public roads, from cars going the wrong way to bicycles darting in front of traffic, as it strives to win the high-profile race to achieve full vehicle automation.
The company, which has been developing autonomous cars since 2009, said its self-driving vehicles had logged 3.2 million kilometres on public roads, and it continues to log about 48 000km of test drives per week.
Google has been a leader in testing autonomous vehicles on public roads, prompting traditional carmakers to step up their self-driving plans to avoid being eclipsed by the tech leader, which has yet to disclose a business strategy for its car project.
It has focused on making cars fully autonomous, with no need for a driver, which could make driving safer and more efficient and open up transportation to the disabled and aged. Google said in 2015 such cars would be ready for production by 2020.
This approach stands in sharp contrast to many carmakers, including Toyota and Tesla, which are moving towards autonomous driving in incremental steps, currently still requiring drivers.
Tesla has said it logged more than 160 million kilometres since October 2015 from drivers using its partially autonomous Autopilot system.
Nevertheless, pointed out Dmitri Dolgov, head of Google's self-driving technology effort, kilometres on predictable highways were easier than navigating busy city streets; Google's cars had to cope with more complex situations than just following a car on the freeway, he said.
“You have to have a deeper understanding of what's on the road and to the side,” Dolgov said after a ride in a self-driving Lexus near Google’s Mountain View headquarters, during which the car reacted to individuals in wheelchairs, pushing strollers and a car making a U-turn.
“Every time you drive it's different.”
The Google car can detect whether a moving object is a child or a bicycle and anticipates that both can make fast, unpredictable movements. The car's driving system calculates the probability of such movements and uses the results to determining how the vehicle will react.
Those types of complicated social interactions are the last, most difficult element of autonomous vehicle technology, according to Dolgov.
“You get to 90 percent autonomy in 10 percent of the time and then spend 90 percent of your time on the last 10 percent,” he said.
The closest rival to Google's self-driven distance in California is component company Delphi, with 26,660 autonomous kilometres, according to a report filed in January with state regulators