The Baidu Research and Development Center in Sunnyvale, California opened two months ago. It is Baidu’s third research center in the US and Baidu’s second office in Silicon Valley. Unlike the first two centers, this one is focused on its most competitive area of research: autonomous driving in the AI field.
On January 5, the Sunnyvale center opened to the media for the first time. Baidu also for the first time invited outsiders to participate in a road test of the Baidu Apollo 2.0 system.
Baidu’s Apollo 2.0 system will be released at the international consumer electronics exhibition on January 9.
L4 Level: “Driverless” is Better than “Manned”
“This is the first time we have invited outsiders to participate in the L4 road test. It will be an experience for you, and you will be like our test subjects. Please share your feelings,” a Baidu engineer said before this reporter got on the test.
The upcoming L4 level autonomous vehicles will be used for autonomous driving on simple urban roads.
An engineer said “simple city road” refers to conditions with few unexpected accidents. “In Silicon Valley, one-way roads can be considered simple urban roads,” he said.
According to the definition of the International Automobile Engineers’ Institute, vehicle automation is divided into L0-L5 levels. The higher the number, the more the car is automated and intelligent. L4 belongs to “highly autonomous driving” under certain scenarios. Except for some special situations, it usually requires no human intervention. The L3 level refers to advanced auxiliary driving functions such as automatic driving, road correction, automatic parking and adaptive cruise.
Journalists tested a Lincoln MK L4 smart car. During the 2 kilometer journey, the car automatically turned corners, changed lanes, identified signals, automatically followed cars and avoided vehicle collision.
The screen beside the driver showed the route planned by car’s “brain” according to road conditions. At the same time, detectors could recognize the environment and display it in real time. A yellow box referred to a pedestrian, and a green to other vehicles.
The Lincoln MK car is equipped with high-standard autopilot hardware. The car roof has a 64-line laser radar, there are two 16-line lasers on both sides of the roof and two cameras near and far. The front bumper is fitted with a millimeter-wave sensor for more accurate identification of front roads and obstacles. The entire vehicle’s brain – two supercomputing machines – are placed in a backup box. One computer is in charge of the center control and the other is in charge of visual recognition.
This sensing system can help the car to recognize pedestrians, vehicles and structures on the road. After being set, it can identify small objects such as dogs and cats.
The current market price of a 64-line laser is $80,000. In the process of testing, staff told journalists the development of production technology and precise spare parts would help to reduce the cost. And technology companies will also offer more feasible solutions to speed up the commercialization of automatic driving.
ThePaper.cn repeatedly tried Baidu’s different self-driving cars. In terms of passenger experience, the L4’s self-driving cars were more fluid in the brakes, deceleration, avoidance of obstacles and steering. Its driving was closer to human control. At the Asian Consumer Electronics Expo in Shanghai last summer, the self-driving cars that ThePaper.cn reporters tested performed less satisfying in parking and accelerating as the movements were too sudden and stiff.
An engineer from Baidu’s research center said the improvements were due to enhanced sensor precision and algorithms, enabling vehicles to observe and predict road conditions faster. At the same time, the vehicle’s deep learning ability will constantly imitate and learn human driving habits, making “driverless” better than “manned”.
The engineer said faster speeds put higher requirements for computing and sensing accuracy. Baidu;s L4 self-driving cars have performed well in tests at 90kph or even faster.
Level L3: Applications for Car Sharing
Since launching its Apollo open platform last year, Baidu has attracted more than 7,000 developers to settle in, more than 1,700 partners to download the Apollo code and more than 100 to apply for open data.
The car sharing company Pand Auto is one of the partners that joined the Apollo platform. In the afternoon, Pand demonstrated its automatic outbound and automatic parking of ag car with Baidu’s L3 level automatic driving capability.
“We think that in the future, people will not spend time looking for cars or parking cars when they use a shared car. It’s the right way to drive,” a Pand Auto spokesperson told ThePaper.cn.
He open an App and clicked “I want to use the car”. A self-driving car in the distance came up to the demonstrator. “After using the car, the driver can get off and click ‘I want to return the car’. The car will go to the parking space and stop automatically.”
During the demonstration, a cyclist rushed to the front of the car and the car slowed down until the bike passed. Its cars can also stop at a red light and pass at a green light when there are simulated signals.
All those involved in the test received a disclaimer from Baidu before getting on the car.
“Driverless car technology has yet to be proven and it is constantly developing,” the disclaimer said, “There are inherent risks in testing driverless cars. Before starting the test, you must ensure that the test subject understands the risks associated with the new technology and agrees to take the risk associated with participating in the test.”
The disclaimer noted that during the unmanned vehicle test, passengers may face daily driving risks inherent in environmental conditions, such as bad weather, traffic jams and other negligent drivers, as well as mechanical failure which could cause injury or death. In addition, self-driving car tests may have other technology-related risks, such as computer program errors, sensor errors, computer processing errors and GPS positioning error. All these risks can significantly increase the risk of accidents.
According to Baidu’s staff, signing such a document is a matter of local law in the US.
Self-Driving Tracks in Silicon Valley
During the trial, Baidu research center recruited volunteers on Twitter. A junior student of mechanical engineering at University of California, Berkeley was selected as the first regular passenger to experience the self-driving car.
On the afternoon of January 5, he arrived at the scene early and waited nearly an hour before he got on the bus. After the test, the college student said, “Baidu should be more confident about its own technology. The experience would be even better if the test passenger could sit in the driver’s seat.”
California is the world’s first region to allow driverless cars on the road for testing. As early as 2015, California opened real road tests to licensed institutions and also introduced a special administrative act.
Previously, the local government ruled that driverless cars would have to be equipped with a tester to be ready to take over the vehicle at any time, and the vehicles must be equipped with a steering wheels and brakes. In September 2016, the governor of California signed a new bill that would eliminate the need for drivers, which means “driverless” car test can be unmanned.
Thanks to the policy, the world’s leading self-driving companies have converged. In Silicon Valley, where technology companies huddle, it is not hard to stumble upon a “driverless car” on the road.
As of December 28, 2017, the number of companies licensed with automatic driving in California has reached 49, eight of which are Chinese companies. Among them are large technology companies such as Baidu; traditional automobile enterprises, such as SAIC and Chana; start-up technology companies such as Pony.ai, Xtecher and Roadstar.ai; as well as domestic new car companies, such as NIO and Faraday Future.
“Autonomous driving is a very hot area and it is not impossible for Chinese companies to be the first to achieve it,” said Wang Jingao, senior director of Baidu’s research center, “The Chinese government supports this technology, and China’s car companies are willing to take advantage of that opportunity to develop it. China is still the biggest car market.”