Photocoupler

For the landing of higher-level autopilots, the first thing to do is to choose the right scene. Compared to complex urban internal traffic, highways may be a more reliable destination. In this scenario, the current laser sensor is more difficult, so the camera needs to be seen very far, and the laser radar itself also faces the problem of capacity and cost, which may limit the pace of commercialization of autonomous driving. Let's take a look at the related content with the car electronics editor.

Autopilot "jumping" is not that simple

Autopiloting is not as simple as going over the corner to go straight to the L4.

According to the current autonomous driving technology, the jump from the L2 level to the L3 level is a qualitative change. However, Volvo, Ford, Toyota and other auto companies have recently said that the automatic driving can be directly upgraded to the L4 level automatic driving stage across the L3 level. .

The US Highway Safety Administration divides auto-driving cars into five levels: driving support (L1), partial automation (L2), conditional automation (L3), high automation (L4), and full automation (L5). At the L4 level, depending on the system requirements, the driver does not have to respond to all system requests, including road and environmental conditions, while the L5 level represents fully automated driving, which is ideal for unmanned driving.

"L3 is a process of man-machine driving, requiring the driver to let go, but he has to be ready to take over at any time, which means that the driver can only close one eye with one eye, and at the same time ask the driver to concentrate on the whole process. This is It is very contradictory." Why are car companies tend to skip L3, and Chen Chaozhuo, research director of China-Sweden Traffic Safety Center, explained this.

However, it is not so easy for autonomous driving to bend over and go straight to the L4. "Doing an L4 on a city road is a matter of 10 years or even longer. If it is in a section like Beijing, it is very long to do L4." At the 2017 NetEase Future Technology Summit held in Beijing on July 15th. The future co-founder of Tucson, Hao Jianan, said.

Why should I skip the L3 level?

This month, the new generation Audi A8 will be the world premiere in Barcelona, ​​Spain, and will be officially launched this year. It is reported that this is the world's first production model equipped with L3 level autopilot technology.

The so-called L3 level refers to conditional automatic driving, and the driver needs to provide a response when appropriate. "Whoever is at L3 level, who is looking for legal troubles." In Chen Chaozhuo's view, "L3 level autopilot is a legal and security nightmare."

"Because it is a man-machine drive, the driver's hand can spread the steering wheel, and once the accident occurs, the driver will take over when the car has no way to deal with it. Usually this period is very dangerous and will cause safety hazards." Chen Chaozhuo Say.

In his view, the development of autonomous driving does not have to be gradual, from L2, L3 to L4, L5. Volvo is now skipping L3. The L4 unmanned vehicle has been developed and released in September last year. Volvo also plans to use four years to verify the road in the world, including the United Kingdom, China, the United States, Sweden and other countries, and plans to mass production in 2021. .

Not only Volvo, but Ford has also announced that it will skip the L3 autopilot and focus directly on the development of L5-class autopilot technology that can be mass-produced. In this regard, Ford Asia Pacific Smart Mobile Strategy Director Lei Xiantong explained that the L3 level requires the driver to be ready to take over the vehicle at any time, which adds a lot of uncertainty to the driving process. Ford chose to make such a choice to completely eliminate human interference in the auto-driving behavior of cars.

"For consumers, the L3 level can't make consumers completely out of driving, so why do consumers pay a high price for this?" Zhu Xi, a professor at the Tongji University Automobile Institute, questioned.

L4 level faces technical nightmare

If L3 is a legal and security nightmare, does L4 and L5 mean a technical nightmare? Because full driverlessness is considered the mother of all artificial intelligence projects.

Chen Chaozhuo said that L4 level unmanned driving faces many challenges, such as human-computer interaction, time-effect mode, automatic driving and other vehicle system interaction, and environmental awareness.

"L2 environment awareness rate is 80%, 90% is no problem, because the driver is the main body; once entering L3 or above, the environment perception must be close to 100%, 99.9% of the recognition rate will not work." Chen Chaozhuo stressed.

In addition, Chen Chaozhuo believes that whether it is a laser radar, a millimeter radar or a camera, it is difficult to fully realize the obstacle recognition after the L4 and L5 levels, and it is difficult for artificial intelligence to judge the intention of the person ahead.

"For example, driving to the urban and rural junctions of the Fifth Ring Road and the Sixth Ring Road in Beijing, it is difficult to determine which old lady wants to 'touch porcelain' in the automatic driving of the L4 level. The front takeaway brother is going to turn left or right. Judging personal intentions, I think that L4 can't do it now," Chen Chaozhuo said.

Even more serious is the fact that fully automated driving faces the challenge of information security. "Unmanned vehicles will be agents in the future. If they are controlled by hackers, they may not be a problem of property loss, but a loss of life," said Tian Daxin, assistant dean of the School of Transportation Technology and Engineering at Beijing University of Aeronautics and Astronautics.

In August last year, only 100 cars were hacked in Texas, and these cars only need to be able to start and drive through the computer. In the opinion of industry experts, network information security is a never-ending battle, and car manufacturers must have sufficient technical ability, patience and hope to establish a security barrier.

"The high-level driverless technology does not have a well-recognized mature solution today, so if we solve this problem, we will stand on the foot of the entire supply chain." Hao Jianan said.

In Tian Daxin's view, 2030 may be the more reasonable node for L4 autopilot. In Peng Jun, the founder of Pony.ai, Xiaoma Zhixing, L4 autopilot is going to run on all the streets in Beijing, and it may be too early in 2030.

Choose the right scene to land

In order to promote the automatic driving technology as soon as possible, Baidu previously announced an "Apollo" plan. The program will provide an open, complete and secure software platform for automotive and autonomous driving partners to help them quickly build their own complete autonomous driving system in conjunction with vehicle and hardware systems.

At CES Asia 2017, Baidu not only statically displayed the software and hardware in the Apollo program, but also brought the autopilot test car in cooperation with the Great Wall. In these test vehicles, Baidu relied on a camera as a sensor to achieve the L3 level of automatic driving. This means that car companies can significantly improve vehicle safety and increase selling points with only a small increase in costs.

However, it is not an easy task for Baidu Apollo to succeed. In the industry analysis, the premise of success is that on the one hand, it needs to gain leadership in algorithms, deep learning, artificial intelligence and other technical aspects, on the other hand, it needs to be recognized and supported by component manufacturers and OEMs.

The success of the Apollo program remains to be seen, but it is an industry consensus to promote automatic driving as soon as possible.

“Autonomous driving is not a helper, not a helper, but a replacement. People are not involved in the whole driving process. L4 may be completely unmanned in a specific scene. L5 is all scenes. Driving.” In Hao Jianan’s view, for the landing of higher-level autopilots, the first thing to do is to choose the right scene. Compared to complex urban internal traffic, highways may be a more reliable destination.

Hao Jia-nan said that relatively high-speed scenes may have to consider longer distances. In this scenario, the current laser sensor is more difficult, so the camera needs to be seen very far, and the laser radar itself also faces the problem of capacity and cost, which may also limit the pace of commercialization of autonomous driving.

"Different sensors have their own characteristics. For business scenarios, they will choose to be acceptable within the cost range, and then based on this limit, constantly tap the potential of the sensor itself, and then achieve the goal." Hao Jianan added.

Chen Chaozhuo, an engineer, finally emphasized that functional safety is essential for a complete transition from human driving to machine driving, that is, the functional safety of each component cannot be defeated. In addition, the automatic driving system must have a redundant design. When one sensor fails, the other should be connected immediately to ensure personal safety.

The above is about the car electronics - automatic driving "jumping" is not so simple related introduction, if you want to know more information, please pay more attention, electronic engineering will provide you with more complete, more detailed, updated information.