"Radish Run" Test In the event of an accident, behind the pilot of unmanned commercialization of autonomous driving: perception technology and decision -making technology urgently need to improve computing power

1. What is the good for the development of autonomous driving technology?

Jason, an industry insider engaged in autonomous driving technology, believes that the pilot policy of all -unmanned commercialization issued by Wuhan and Chongqing cities this time is a good exploration and attempt. "Technology research and development requires a large number of tests. Only by open places for us can technology develop faster."

Bian Jun, the founder of Shanghai Panshi Information Technology and Sasetech, the car safety public welfare community, also believes that the open part of the area will help technological progress. "Some companies are indeed very technical, and they need to let go to further. Discover problems, but you need to grasp the risks that may face after letting go. "

↑ Autonomous driving system experience. Picture according to IC Photo

2. Can driverless commercial operations popularize large -scale?

Jason said that from a technical perspective, several domestic heads in China can indeed be unmanned, but because the current unmanned sensors are expensive, the structure is precise The popularization of technological development will stay for another time.

"Now the driverless technology can meet some small scenes, such as the shuttle bus in a certain park, or some bus lines, the scene is relatively simple, the route is relatively fixed, the scope area is relatively small, but the cross -regional large -scale area The current technology of commercial operations cannot be supported. "

Moreover, there are relatively large restrictions on driving speed and operating range at present, and there are also high risk of accidents. "When no one in this car is guaranteed, 90%of accidents may directly evolve into accidents."

Jason believes that from the perspective of commercial operations, it is still in the exploration stage. Whether it is the operating service provided by the enterprise or the actual ride experience of passengers, there is still a long way to go.

Bian Jun also said that the current driverless technology has not yet matured, and there are many marginal scenes that the current perception system cannot solve, such as identification and accurate decision -making of some complex road conditions, and the response to extreme weather. Essence Therefore, for the time being, you can only try to run in a small area, accumulate some data, and gradually let go.

"For example, there is a pit in the middle of the road, and the rain filled it after the rain. At this time, your perception system may not be able to identify it as a pit, and it may be considered as a normal pavement reflection, or the water is shallow, so you just think that the water is very shallow. Pressing directly will cause safety risks. "

3. What is the technical logic of autonomous driving technology?

Jason stated that according to the "Automation of Automotive Driving" in national recommendation standards, domestic autonomous driving is currently divided into six levels of L0 to L5. Auxiliary driving still needs the driver to control the vehicle. The technology is just the auxiliary driver's judgment of driving decision -making. "But many consumers are actually unable to distinguish between these (levels) and attribute all technologies to autonomous driving. This cognitive blur will also indirectly lead to many accidents."

Jason told the Red Star News reporter that the driverless vehicle of Wuhan's pilot is L4 level. The bottom demand in this level is to "remove people". The perceived information and the set destinations to calculate the optimal driving route and driving strategy.

4. Incidents related to autonomous driving technology occur frequently, how to ensure safety from technology?

Jason said that unmanned driving above L4 is mainly relied on redundant to ensure safety. "All components on the L4 level autonomous driving body are multiple certifications. Even if a camera or chip is broken, the other hardware will replace it instead These hardware completes the task. "

However, there is no need for too much redundant devices in autonomous driving below L3. The main role to ensure safety is the driver. In the national recommendation criteria of "Automation of Automotive Driving", the driver's role is clearly defined.

↑ 2021 Shanghai Auto Show, Baidu Autonomous Driving Platform Apollo. Picture according to IC Photo

Jason said that whether it is the driverless driving at the L4 level or the intelligent auxiliary driving of the L3 level and below, the types of sensors needed for the surrounding environment are the same. Basically, they are lidar and cameras. have difference. Among them, L4 driverless must use laser radar, which is higher in terms of accuracy and safety, while L3 and below only need cameras and millimeter wave radar.

Bian Jun told the Red Star News reporter that the ultimate solution of autonomous driving is the problem of security. Among them, perceptual technology and decision -making technology are the focus of two urgent solutions. That is, whether it can solve the various scenarios encountered by the vehicle, and whether the scenes that cannot be resolved can take risk avoidance measures in advance. "In fact, we have been solving the problem of 99.99 last 9. This aspect requires the improvement of computing power, and on the one hand, there must be accumulation of test mileage."

5. What is the biggest difficulty in the current development of autonomous driving technology?

Jason believes that for the driverless driving of L4 and above, the biggest difficulty of technology research and development is that it is impossible to understand all driving scenarios. "There are often many unexpected scenes in daily driving. It is not solved by it. It just said that L4 mainly relies on redundant to ensure safety, but this also means that there are no details during driving. Do it. "

For the intelligent auxiliary driving technology of L3 and below, the biggest problem lies in how to coordinate the interaction between people and cars, that is, how to be safe and allowed to steal some "laziness".In this regard, Jason stated that DMS (driver monitoring system) is currently more commonly used solutions. This system can monitor the driver's state through the camera's recognition of people, steering wheel induction including voice interaction, etc."But at present, there are still many shortcomings in this system. On the one hand, if the sensitivity is too high, it is easy to make an error judgment.If you are sensitive, it is easy to supervise and supervise it in time when there is an accident. And at present, it is more to see the driver's response. If the driver does not take over the vehicle, the accident cannot be avoided. "

—End-

- END -