Xiaomi SU7 car accident is still continuing to ferment. In addition to focusing on whether the door can’t be opened and the front of the car catches fire, a lot of attention is focused on whether the active safety functions such as emergency braking (AEB) of Xiaomi SU7 are triggered in the accident.
On the evening of April 1,The official once again issued a statement that Xiaomi SU7 Standard Edition has two sub-functions of forward anti-collision assistance, including collision warning (FCW) and emergency braking (AEB), which are targeted at vehicles, pedestrians and two-wheelers, and the working speed of AEB function is between 8-135 km/h. This function is similar to that of AEB with the same configuration in the industry. At present, it does not respond to obstacles such as cones, water horses, stones and animals.
In short, Xiaomi believes that the current AEB function does not support the identification of roadblocks in accidents, but it does not support the same configuration models in the industry.
In this regard, an intelligent driving industry person told reporters that construction roads will generally be marked on high-definition maps, and guiding NOA to divert will remind them to take over. If the intelligent driving system gives a certain weight to the real-time updated HD map, there is a certain probability to avoid the accident.
According to official information, the company just completed the full-scale promotion of "Xiaomi End-to-end Full Scene Intelligent Driving (HAD)" in February this year, indicating that "end-to-end user driving data training has been introduced, and on the basis of" no map ",the trajectory is more anthropomorphic, traffic is more efficient and driving is safer", and it is clear that HAD covers highways, urban expressways and so on.
The price of "no graph"
When discussing the boundary between intelligent driving ability and responsibility, we have to admit that even in the case of artificial driving, Xiaomi SU7 has a certain probability of distress in this German-Shanghai expressway accident.
According to Xiaomi’s notification, before the accident, the vehicle was in the state of NOA intelligent assisted driving and continued to drive at a speed of 116 km/h. Due to construction repairs, the road section was closed with roadblocks and diverted to the reverse lane.
At present, the official police report has not been made public. However, because the speed is too fast, and Deshang Expressway is located near Jiuhuashan Mountain, and the lighting situation at night is not good, whether the direct cause of the accident is caused by the vehicle hitting the roadblock at the high-speed NOA or the driver’s improper steering+braking after taking over is actually inconclusive (although Xiaomi may be to blame in both cases).
So, why can’t high-speed NOA recognize lane changes?
A person in the smart driving industry who did not want to be named told the reporter that this may be related to the fact that the smart driving system does not use high-precision maps, or the decision-making weight of maps in the smart driving system is too low.
"If the intelligent driving system uses a high-precision map, at least it will prompt the construction ahead, and then NOA’s route planning will be different," the person said, but pointed out that when dealing with the diversion, whether to continue NOA according to the new road or slow down and ask for takeover, each scheme may be different.
However, the interviewee also told reporters that road construction is divided into temporary construction and planned construction, and the planned construction will generally be uniformly marked on the road network, prompting the smart driver to change the route. But the temporary construction information is "not necessarily" uploaded.
Considering the risk of landslide caused by the geographical location of the German-Shanghai Expressway, the possibility of temporary construction is also great-however, the interviewee revealed to the reporter that according to his understanding, this construction is not temporary construction.
The reporter looked up the official subscription number of Xiaomi Automobile and found that at the early stage of Xiaomi SU7′ s release, the company publicized that it adopted the "light map" urban NOA scheme. However, in October last year, Xiaomi announced that it had pushed version 1.40 smart driving in large quantities, and made it clear that the NOA function of the city allowed vehicles to avoid construction.
However, on March 19 this year, the subscription number issued a document, saying that in February this year, the end-to-end full-scene intelligent driving function was fully pushed, and it was repeatedly emphasized that the intelligent driving system was "no map".
"Xiaomi’s plan is biased, and there is definitely no high-precision map at the bottom. Another system design is problematic." The aforementioned person pointed out to the reporter: "In fact, Cheluyun is a way to achieve automatic driving with the lowest collaborative cost."
Of course, car dealers flaunt "no map" not simply to reduce costs. According to the reporter’s understanding, the pure vision scheme will contradict the HD map that is not updated in time to a certain extent. In the choice of reference weight of intelligent driving system, in order to avoid the contradiction of reference data and make decision-making difficult, we will also give up high-precision maps on our own initiative.
The boundary of "pure vision"
Then, since we chose "no map" and emphasized perception, what about Xiaomi’s perception ability?
Back to the Xiaomi accident vehicle, from the sensor configuration point of view, the intelligent driving sensor configuration of the SU7 standard version Pilot Pro is two forward-looking, three upward-looking, four around-looking, a rear-looking camera and a forward millimeter-wave radar (of course, there are 12 ultrasonic radars, but the ultrasonic radar has insufficient performance and almost no weight in high-speed intelligent driving decision), commonly known as "9+1" configuration.
And this configuration is almost the same asThe plan is exactly the same. The HW 4.0 system is equipped with four side-looking cameras, three front-looking cameras at the top and a rear-looking camera at the tail, including a forward millimeter-wave radar at the front of the car, which is "7+1" configuration.
The intelligent driving decision-making system, which relies almost entirely on the camera, undoubtedly belongs to the so-called "pure vision scheme".
So, how reliable is the pure vision scheme? Can you handle unexpected emergencies?
Last year, the media conducted an in-depth report.Many casualties caused by opening FSD were exposed and analyzed.
In the report, FSD caused many "stupid" accidents, such as directly rushing out of the T-junction and chasing a police car with flashing warning lights. And one of the accidents is a typical warning to the pure vision scheme.
In this accident, the driver who started the automatic driving almost ran straight into the truck chassis that rolled over and heeled in three lanes at a medium speed of about 40-50 km/h around 3 am, causing the driver’s death. It is worth noting that, from the video, the rollover truck is huge in scale, close to three meters in height, while the lighting conditions on the road are acceptable, the speed is not fast, and the human eye can quickly capture and respond in time.
However, in the face of this simple emergency treatment from the perspective of natural persons, Tesla FSD, which was advertised as surpassing human driving ability, caused an accident.
The media quoted automobile engineering experts in colleges and universities as saying that the above-mentioned incidents may have occurred because Tesla could not identify the chassis characteristics of rollover trucks and mistook them for roads or backgrounds, so no emergency treatment was carried out.
This also brings a soul torture to the pure visual route: pure visual intelligent driving not only needs to know cars, but also needs to know things, not only pedestrians, pets, or even suddenly dumped goods-more unexpected irregular objects may become the culprit of traffic accidents.
Judging from Xiaomi’s reply, AEB system really can’t give feedback on the diversion obstacles such as cones, water horses, stones and animals. However, Xiaomi officially insists that AEB, a "friend", can’t do it either.
But the problem is that the standard version of Xiaomi SU7The number of equipment is almost "stingy" compared with that of friends at the same price, and the decision-making also depends entirely on pure visual ability, so it is obviously not convincing to compete with AEB system alone.
According to the data of knowing the car emperor, in the same price model, such as the ideal L6 intelligent driving sensing configuration is almost the same as that of Xiaomi SU7-but more include007, Tucki P7,Each configuration version of Han EV and Seal is equipped with at least 5 millimeter-wave radars and 1.; At least three or more millimeter-wave radars are equipped on the models of Wujie M5/M7 and Tucki P7+.
In other words, compared with "friends", the security redundancy of Xiaomi SU7 Standard Edition is far from enough.
"The training of a qualified driver, in addition to road traffic rules and driving skills training, relies more on the accumulation of intangible life experience to gain the ability to cope with emergencies. This is precisely the gift of life to mankind, and it is alsoAt present, it is far from the realm. We may be able to achieve pure vision in the future, but we should not train intelligence through accident accumulation. "The person from the aforementioned listed company told reporters:" Lei Jun (Xiaomi) may not be washed this time. "
关于作者