The self-driving car race just got another contestant. Baidu, A.K.A. the Google of China, announced Wednesday that its robocar, a retrofitted BMW 3 Series, autonomously drove an 18.6 mile loop from Baidu's headquarters in China through city roads and highways, and back home to Baidu HQ:
The robotic Beamer hit top speeds of 62 m.p.h. It was proficient at making turns. It slowed down when it detected cars ahead. It could change lanes, merge into oncoming traffic from ramps, and take exit ramps properly.
“Fully autonomous driving under mixed road conditions is universally challenging, with complexity further heightened by Beijing’s road conditions and unpredictable driver behavior,” said Wang Jing, the general manager of Baidu’s newly established Autonomous Driving Business Unit, in a statement.
That business unit will start off by developing fully autonomous vehicles that will be used as public shuttles, according to the Wall Street Journal. It doesn't have set plans for making its self-driving vehicles available to consumers.
The company has been working on autonomous vehicles since 2013. It's competing against Google—the clear leader—Apple and Tesla, among others, who have also set up their own autonomous car operations. Baidu wants to build fully autonomous vehicles, instead of adding incremental autonomous features, like Tesla did with its autonomous highway-driving feature, for instance.
The Chinese search giant's robocar is powered by software dubbed Baidu AutoBrain, which includes what the company's calling highly automated driving maps, positioning systems, and smart decision-making and control. The car makes use of Baidu's own mapping service, which it's had for 10 years. Last year, the company invested $10 million in a Finnish mapping startup. As we pointed out before, maps are a crucial part of winning the robocar race.
Google arguably has the best maps, though lately Apple and other automakers have been making strides toward building better mapping technologies.
"There’s a lot of data on a map you don’t think about when you pull up a map on the street. You don’t know lane by lane what’s going on,” Nvidia’s Danny Shapiro told Fusion in August. “A car has to have very precise software and hardware to understand all that.”
Baidu, it seems, is making sure its cars are able to do all that. Like other self-driving cars, Baidu's is powered by deep learning, a breed of artificial intelligence that's helping tech companies make massive improvements in image recognition.
Andrew Ng, whom Baidu hired last year to head up its AI operations stateside and in China, is one of the people working on the project. Before joining Baidu, Ng set up the Google Brain project, which was the search giant's foray into deep learning. He's among the most respected researchers in the deep-learning community.
As The Guardian reported in June, Ng and his team enjoy one very big advantage over robocar developers in the West: flexible legislation. The company is one of only a few to have been given permission to test autonomous cars out.
But the challenge is greater, too. Beijing is a bit more complex than Mountain View.
Daniela Hernandez is a senior writer at Fusion. She likes science, robots, pugs, and coffee.