In just one week, there have been three accidents related to assisted driving systems exposed by the media, and most of them happened to new car manufacturers, which once again raised questions about the safety of ADAS (Advanced Driving Assistance System).

In just one week, there have been three accidents related to assisted driving systems exposed by the media, and they mainly happened to new car manufacturers, making the outside world once again raise doubts about the safety of ADAS (Advanced Driving Assistance System, Advanced Driving Assistance System ).

Assisted driving

html On August 10, a Xiaopeng P7 suddenly rear-ended on an elevated highway in Ningbo, Zhejiang, resulting in one person's death. According to reports, a Xiaopeng P7 in Ningbo, Zhejiang, suspected of using the intelligent driving assistance function on a viaduct, collided with a person checking for vehicle faults and a faulty vehicle in front of it. The accident resulted in the death of a passenger in the front car. The

driving recorder and surveillance video showed that a Xiaopeng P7 was driving in the leftmost lane of the viaduct section, and then hit a broken-down vehicle parked in the same lane. A person standing at the rear of the broken-down vehicle was unfortunately knocked away, while another person escaped because he was standing close to the guardrail. The information provided by the owner of the Xiaopeng car exposed on the Internet stated that the vehicle had turned on the LCC at that time, and he was distracted at the time.

The so-called LCC, Xpeng Motors manual explains it this way: it is a driving assistance function, not automatic driving in the complete sense. The driver still needs to keep his hands on the steering wheel when the system is activated so that he can take over the vehicle in the event of danger. It also specifically points out that the system cannot respond to unexpected dangerous situations and the driver has the responsibility to remain vigilant at all times.

In response, Xiaopeng Motors responded that it is fully cooperating with the investigation in accordance with the requirements of relevant departments. The results of the investigation in this case have not yet been announced.

A few days later, there were media reports that a user died in a traffic accident while driving a Xpeng car. According to reports, the accident occurred on July 27. At the time of the accident, the vehicle had no obvious braking marks, and the airbag did not deploy after the collision. The video captured by the driving recorder showed that the driver gradually deviated from the lane while driving, eventually causing the vehicle to hit the barrier. A section of the railing that was struck damaged the driver's brain. In addition, the vehicle showed no obvious signs of slowing down before hitting the guardrail.

Xpeng Motors responded: At present, the fact has been found out that the vehicle’s driving assistance function was not turned on at the time of the incident and it was in a manual driving state. Before the inferred collision point, the vehicle has been accelerating, and the opening of switch is about 27%. After the collision point, the brake pedal was depressed, and the maximum braking pressure was 37 Bar, which was considered moderate braking. The final report is subject to official output from the traffic police department.

then ideal auto . According to a video exposed on the Internet, a ideal ONE turned on driving assistance and drove on the highway at a speed of 77km/h. A construction vehicle with obvious text prompts appeared in front of it. Although the construction vehicle clearly had "slow down" written on it, neither the driver nor the ADAS system performed emergency braking or deceleration braking, which ultimately resulted in the vehicle rear-ending construction vehicle .

In response to this accident, Li Auto’s official response stated, “The user turned on the NOA function, but the user was always hands-off, that is, not holding the steering wheel. This accident exceeded the working range of the 2021 Li Auto ONE’s ADAS (stationary vehicle).” At the same time, it said, "The accident has been handled and the user recognized the cause of the accident."

observed the above three accidents. Although some cases have not yet been finalized, one of the key issues deserves special attention-that is, whether the driver has excessive trust in the vehicle's ADAS. According to logic, these accidents are likely to occur as follows: either the ADAS system is turned on, but the obstacle is not recognized, and the driver does not remain vigilant at all times; or the driver thinks that ADAS is turned on, but in fact it may not be, and he crashes into the obstacle. But no matter what the situation is, it is a scenario that will occur frequently in the current L2-level assisted driving stage.

We have always believed that in the stage where ADAS is somewhat useful but cannot be completely trusted, the driver is actually more tired than driving alone.Of course, some people can refute this point of view by citing many experiences in which they experienced less worry and no accidents after turning on ADAS on the highway. However, you must know that once the system or a person makes a mistake, repairing the car will be time-consuming and labor-intensive at best, and it will cost lives at worst. What users want is not "safety most of the time", but "absolute safety".

Another important question at the moment is whether car companies have exaggerated the effects when promoting relevant functions, causing the assisted driving system that should not be so trusted to gain "excessive" trust, which is also one of the causes of such frequent accidents.

Such cases also trigger our thinking about technology. Why does the current combination of mainstream cameras plus millimeter wave radar have such a low recognition rate for static objects? Is it because the technology cannot be achieved, or is it a problem with the alarm boundary?

Yin Wei, manager of the intelligent driving environment integration system of SAIC Motor, gave a possible answer on the Zhihu platform.

In his opinion, the conclusion is that this combination is fully capable of identifying static objects. But the problem is that "the test results cannot be fully utilized."

He pointed out that there are two issues that need to be paid attention to in system identification. One is false positive, which means that something that does not exist is considered to exist, which can be understood as a false detection. The second is false negative, which means that what exists is considered not to exist, which is also called missed detection.

This involves a concept of boundaries. If the alert protects low-value items, then people don't want it to be too sensitive, which will be very annoying. And if it's protecting high-value items, then one would want it to be more sensitive.

has been used in autonomous driving, which is currently divided into L2-level routes for assisted driving and L4-level routes for driverless driving. In L2-level assisted driving, the responsibility lies with the driver. Therefore, when designing system strategies, car companies will prefer not to "misdetect", that is, not to be too sensitive. Because this will cause the system to frequently decelerate and brake, making the driving experience extremely poor. Therefore, some object systems are clearly aware of it, but are blocked by it, in order to ensure the smoothest possible experience.

But L4 driverless driving is different. Its responsibility lies with the operator or OEM behind the car, so it needs to not "miss detection". The more sensitive the better, and every disturbance must be able to be detected and reacted to.

Under these different strategies, the system sensitivity and "error rate" are obviously different. What's more, even for systems of the same level, the boundary strategies set by different car companies are slightly different.

Understand this, and you will understand why we call for not relying on assisted driving systems at this stage.

(Source: The Paper News )

For more exciting information, please download the "Jimu News" client in the application market. Please do not reprint without authorization. Welcome to provide news clues, and you will be paid once accepted. The 24-hour reporting hotline is 027-86777777.