The Xiaopeng P7 hit, under the help of driving.
Please close your eyes when it gets dark, yesterday, a Small Peng P7 in the state of assisted driving, directly hit a car lying on its side in the middle of the road, fortunately, there were no casualties in the accident.
It was dawn, there was a car crash last night, and now there is a call for the owner of the accident and Xiaopeng to speak.
Owner: After going on the national highway, I turned on automatic driving, more than ten kilometers later, I encountered an accident car in front of me that had overturned on the road, the vehicle did not warn, nor did it make any avoidance action, I stepped on the brakes and did not react, the speed of the car is still rising.
Xiaopeng: The owner should have enabled ACC+ LCC (Adaptive Cruise Plus Lane Keep), and the accident was caused by the owner not keeping an eye on the environment in front of the vehicle and taking over the vehicle in time. The owner has been actively contacted to arrange a scooter for the owner and communicate about the repair.
Traffic police department: After investigation, it was determined that the conclusion was in line with Xiao Peng's statement, and the speech was completed.
Judging from the existing statements, it is highly likely that the owner did not standardize the use of assisted driving, and just happened to encounter the scene that the current system can not be determined, and this accident occurred. In other words, losing money is impossible and unreasonable.
Moreover, from the owner's speech, it is seen that the owner uses "automatic driving", which also reveals to a certain extent that the owner is not familiar with the function of his vehicle, and the occurrence of the accident is partly "inevitable".
In addition, the owner uses the word "automatic driving" to illustrate the negative impact of Xiaopeng on the previous edge ball naming, and it is not enough to eliminate it.
In order to facilitate the understanding of this event, I would like to switch to the identity of "storyteller" to give you an example.
At the exit of your home community, there is a transparent glass wall, in order to prevent someone from accidentally hitting the glass, the property deliberately painted a red warning tape on the glass, and also sent a brochure to each owner, which wrote how to correctly enter and exit through the door next to the glass wall.
But at the same time, in order to select the title of "cleanest community", the residential property must wipe the glass clean as air, and promises that by 2026, it will be replaced by a glass wall that can be freely shuttled.
Under normal circumstances, everyone is safe, but one day someone walked and played with mobile phones, hit a glass wall, and got into an argument with the property.
You can say that the owner is not serious about walking, or you can accuse the property of not really starting from the user experience point of view, why should it confuse the public between air and glass, and increase the risk of the owner hitting the glass.
At this time, there will be knights who are familiar with keyboard operation skills to stand up and shout, and the gradual promotion of air glass is to play wandering, either directly to a cement wall, or to engage in air glass that can pass through itself, and also quote the scriptures, and the courtyard of the Valley Lord's family in the west is what it is.
But has the hero ever thought about how to directly build a cement wall and let those owners who walk seriously and enjoy the beauty of the glass wall under the constraints of the rules?
Let more than 90% of serious owners pay for the lack of seriousness of other owners, which is the truth of which door, self-restraint is not a sacrifice.
Therefore, there is no solution to this problem, love it, everyone is good at it.
After listening to the story, the details still have to be plated. Why is it that a stationary car, turning over and lying down the system is not recognized? Are car companies thinking of solutions? When will it be solved?
The temptation to lie on your side
It's essentially a "wolf is coming" story. At this stage, the assisted driving solution of most models is the rear fusion perception scheme of camera + millimeter wave radar, and this type of scheme does have "limited ability" for stationary side sleepers, and Tesla has also suffered losses.
This scheme can be understood as the system uses both a camera and a millimeter-wave radar to detect information, and the video screen of the camera is high in pixels, but the distance information is missing.
Millimeter-wave radar brings back a point cloud with distance, but extremely low pixels. Both of these pieces of information are entered for post-fusion processing. Post-fusion means that both sensors will output data to the same target, but the system will weight the output of the two sensors for different scenarios, and the weight determines who believes in this matter.
Due to the low resolution of the millimeter-wave radar, it can only tell the system what kind of speed state obstacles there are in the X meters ahead, but it does not know what the obstacles are.
For moving obstacles, it is easy to deal with, and the system in most cases directly chooses to believe in millimeter-wave radar. But for stationary obstacles, it is difficult to top.
Most of the information on the road is static and does not affect the driving route, such as green belts, road signs, gantries, overpasses and so on. This is like the story of "the wolf is coming", the invalid information is given too much, and the system chooses to filter directly.
But static information about the pavement environment. In addition to the above, there are many real enemies, such as stationary vehicles, pile barrels, children playing with 123 wooden people, and this rollover accident car.
Therefore, the current practice is to identify various objects on the road surface through machine learning, classify them, and then judge whether to believe the distance information returned by the millimeter-wave radar, and brake when encountering the obstacle of the brake, and the rest of the stationary information continues to be ignored.
Xiaopeng XPilot3.0 has been very familiar with all kinds of tails, the pile barrel has also been learned, if the car is a normal posture this time, or the owner of the side lying down can put the pile bucket in front of the road in time, then the system can probably recognize and remind the takeover.
But the side-lying vehicle is really embarrassing AI, which is to give a crab to a person who has never seen a crab, and he is of course helpless. It can't be difficult for humans to have a correct view of things outside the cognitive range, and the same is true for machines.
This is a disease, and it can be cured
This state of affairs is dangerous, but fortunately the car companies are already in action. Let's go back to the key to the problem, the camera resolution is high but there is no distance information, the millimeter wave radar has distance information but the resolution is low, then can there be a high pixel information with distance is not the end of the problem?
Tesla, Xiaopeng, Weilai, Ideal and other car companies have made plans along this line of thinking. Tesla's solution is to remove the millimeter-wave radar and use a monocular stereo vision algorithm to estimate the distance information of the road ahead, so as to plan a safe drivable area. Most car companies, such as Xiaopeng, Weilai and Ideal, have chosen lidar, a sensor that can output high-resolution point cloud information by itself.
The Xiaopeng P5 and Xiaopeng G9, as well as the replacement Xiaopeng P7, will be equipped with semi-solid-state lidar, which can output a precise obstacle orientation and drivable area through the laser point cloud algorithm, which can effectively prevent such collisions.
At present, Xiaopeng G9, Ideal L9, NIO ET 7, NIO ET5, NIO ES7 and other models are equipped with semi-solid-state lidar, and the farthest detection distance under 10% reflectivity is between 150m-200m, which means that at least 150 meters in advance can see the black stationary vehicle.
This time, the system doesn't have to care whether it's upright or lying on its side, slowing down and stopping or changing lanes.