With so many companies developing driverless cars, it can be easy to forget about Baidu.
But the Chinese internet giant, often referred to as the “Google of China”, is making big moves to get its autonomous vehicles on the road.
Baidu is planning to use its driverless tech for a public, shared shuttle service in 2018. But Baidu could also produce self-driving cars for private use after 2018, said Jing Wang, a senior vice president who heads Baidu’s business unit developing autonomous cars.
But Baidu isn’t only focusing on China. On Thursday, Baidu was issued a license to test its autonomous vehicles in California. This doesn’t come as a huge surprise since Baidu had set up a team in Sunnyvale, California earlier this year.
Two key advantages
Wang told Business Insider that Baidu has channeled its efforts into two key technologies to advance its driverless car efforts.
The first is a focus on a branch of artificial intelligence called deep learning to give its cars optimal image sensing and recognition.
Deep learning allows cars to learn on their own so they can deduce what’s around them and make faster, reactive decisions as to how to proceed. Chinese tech company Inspur has created a high-performance, image recognition server by partnering with NVIDIA, the largest vision computing company in the world, that Baidu now uses in its driverless cars.
Baidu’s strategic partnership with Inspur has paid off.
“According to the KITTI Vision Benchmark Suite, the world’s largest traffic scenario dataset, our vehicular recognition has achieved an accuracy rate of 90.13%. This is one of the highest in the world,” Wang said.
Using deep learning software for highly sophisticated image recognition is a strategy others are starting to adopt, as well. Start-up Drive.ai is touting its deep learning software as setting it apart from the competition, and Toyota is investing heavily in using AI for its driverless cars as well.
Baidu is also partners with NVIDIA to develop a cloud platform that will help develop Level 3 autonomous vehicle control, which is when a driver should be present but the car can handle most functions on its own, and autonomous parking capabilities. That platform will be made available to other automakers.
The second key technology Baidu is advancing to aid its driverless car efforts is its HD maps.
Mapping routes is an important part of getting driverless cars road ready. The cars compare what is outlined on the map to what they’re “seeing” with sensors, LiDAR, and radar in order to detect and avoid obstacles.
Audi, Daimler, and BMW acquired Here, the former mapping division of Nokia, to make real-time maps to aid their driverless car efforts, and Google is also mapping its own routes, to name just a few of the efforts.
However, Wang claims that Baidu is the first Chinese company to produce high-resolution, 3D maps of routes.
“Baidu’s high-precision mapping data contains hundreds of properties, and it is precise down to the centimeter,” Wang explained. “The traditional unit for precision in map data is a meter.”
Baidu’s partnership with NVIDIA will also help advance the company’s HD map efforts.
The bigger picture
Baidu got its driverless car efforts out into the public eye in December when a BMW 3-Series modified with Baidu’s driverless tech successfully completed an 18.6-mile route in full autonomous mode.
“It carried out multiple complex driving maneuvers, including slowing down when approaching other vehicles, switching lanes, overtaking vehicles, entering and exiting ramps, and turning around,” Wang said.
Wang added that the cars have also completed other tests in difficult conditions like rain and fog, in addition to driving at night.
Baidu plans to ramp up its testing efforts in China. “We are currently testing in Beijing and Shanghai and will expand to other cities. Our plan is to test in 10 cities within three years,” Wang said.
Baidu cars will also begin testing on public roads in California “very soon” now that it received an Autonomous Vehicle Testing Permit for the state, Wang said in a statement to Business Insider.