Chinese language Internet search provider Baidu Inc. announced that its autonomous car has successfully completed rigorous, fully autonomous tests on one route with mixed roads under a variety of environmental conditions.
The Baidu autonomous car is the first in China to have demonstrated full autonomy under mixed road conditions, marking a milestone in China’s autonomous driving effort, the company says. The road tests were carried out under complex road conditions, and the Baidu vehicle, a modified BMW 3 Series, completed the tests by executing a comprehensive set of driving actions and accurately responding to the driving environment, according to Baidu.
“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, senior vice president of Baidu and general manager of Baidu’s newly established Autonomous Driving Business Unit.
The 30-kilometer test drive route began at Baidu’s Beijing headquarters near Zhongguancun Science Park in Haidian District, extending to the G7 highway, Fifth Ring Road, Olympic Park, looping back and ending at Baidu’s headquarters.
Baidu says the car demonstrated full autonomy on the entirety of the route and successfully executed driving actions, including making right turns, left turns and U-turns, decelerating when detecting vehicles ahead, changing lanes, passing other cars and merging into traffic from on-ramps and exiting from off-ramps. The car speed peaked at 100 km per hour during the test runs.
Baidu’s Institute of Deep Learning, one of the three labs under Baidu Research, has led the autonomous driving project since 2013. Baidu AutoBrain, the core of Baidu’s autonomous driving technology, includes highly automated driving (HAD) maps, positioning, detection and smart decision-making and control.
Baidu’s self-developed HAD maps record 3-D road data to within a few centimeters of accuracy of vehicle positioning. The car can detect, recognize and follow other vehicles with high accuracy; recognize road lanes; and accurately gauge distance and velocity, according to the company.
Baidu says its autonomous car approach is to advance incrementally through different environments, rather than through different levels of driving autonomy. For example, a bus that runs the same fixed route will be able to fit a very reliable model for that route, training computer vision and other deep learning systems through repetition, and further expanding the number of routes mapped by HAD mapping. Within five to 10 years, the majority of China’s roadways could be mapped with HAD mapping, according to Baidu.
Baidu’s autonomous driving initiative is currently the only research project in China that has passed the IOS 26262 functional safety standard.