If you add 10,000 yuan to make 898's Qin plus multiple noa, will you buy it?
What the? Don't buy? Still too expensive?
This can include both the basic L2 and the high-speed urban NOA.
You must know that BYD's original basic L2 configurations such as lane centering assist, departure warning, and active braking add up to 10,000 yuan, and Denza's high-speed NOA is sold for 150 thousand.
Does it still seem expensive when you look at it this way? Sometimes I find out why I am working hard and whether I am working.
Okay, good, no kidding.
In fact, this is the case, recently, DJI received an olive branch from BYD and FAW.
In the future, BYD's Qin Plus, Song Plus, Seagull, Dolphin, and even FAW's Hongqi may be able to use DJI's urban NOA.
Those who don't know about smart driving here may ask:
When did DJI start to do smart driving? Why should it be better than what car companies do themselves?
In fact, let's compare Huawei to understand, this DJI Automotive and Huawei Automotive Division (Car Bu) were both established in 2019, and its qualifications are not shallower than Huawei.
In terms of technology and R&D strength, DJI Automotive is also fully capable of wrestling with Huawei's car.
On the other hand, traditional car companies such as BYD and FAW are still some distance from the first echelon of intelligent driving.
So in the second half of smart cars, BYD urgently needs to find someone to cooperate.
However, DJI's intelligent driving capabilities have not been verified by many products, so why don't BYD choose the more mature Huawei intelligent driving?
Not to mention, BYD had really talked to Huawei's car bu about cooperation before, but was politely rejected by Huawei.
This is not because Huawei does not give face, the main thing is that Huawei has not done high-end intelligent driving of models within 200,000 so far, and if it wants to land its system on BYD, the cost will be too high.
Compared with Huawei, DJI's idea of intelligent driving has been to pursue cost performance, reduce costs, and popularize as soon as possible from the very beginning, so it coincides with BYD.
For example, at this year's CES, DJI showed its own "Chengxing platform" It has neither lidar nor high-precision maps, and the chip used is Qualcomm Snapdragon RIDE (SA8650P), which only provides 100 tops of computing power, which is indeed a very "beggarly version".
However, according to the official test**, this system not only includes overtaking and detouring in high-speed urban conditions, automatic obstacle avoidance, but also memory parking function, which everyone basically has.
If you change Huawei to do it, not to mention the necessary lidar, the intelligent driving chip is estimated to use its own MDC 810, with a computing power of 400 tops, according to the terminal**, a complete set of ADS 20 probably takes 180 thousand.
On DJI's side, according to the Lingxi Zhijia that has been mounted on the Baojun cloud, a total of 7 cameras, 1 millimeter-wave radar, and 12 ultrasonic radars are used, and the price difference between the ordinary version is only 10,000.
You must know that these 10,000 also include ordinary L2 level intelligent driving, so if you only talk about high-level intelligent driving, the estimated cost is about 5,000.
But what we are most concerned about is the problem of use, will it be unsafe for you to be low?
After all, even the current mainstream intelligent driving scheme is not perfect, and the beggar version of NOA with missing arms and legs is probably more difficult to use.
In fact, judging from the video of the owner, the Baojun cloud can also be very smooth in one shot on the elevated highway, but in terms of safety, it is not clear who is superior or inferior to other intelligent driving.
The controversy among everyone is that in order to reduce costs, DJI chooses the "pure vision" solution and cancels the lidar, will it be unreliable?
You must know that the current intelligent driving reduction plan of car companies is basically related to this lidar, but whether intelligent driving is safe is not only determined by a lidar, but also by various algorithms and logic, which involves more content.
To explain this problem, we have to start with the cost war of intelligent driving.
Back in 2004, when the U.S. Department of Defense launched the DARPA Driverless Challenge in order to build driverless military vehicles, and although the participating teams showed their talents, none of them could complete the competition.
That's when a sound company called Velodyne turned things around when it started putting lidar in its cars.
In another edition of DARPA in 2007, seven teams made it, and six of them used its own lidar, so Velodyne became an instant hit.
So why is lidar so useful?
In fact, all sensors have imperfections, for example, the camera can only capture 2D images, and the recognition ability is very limited at night, in rain and fog, backlight, etc.
On the other hand, ultrasonic radar has a very short detection range and has almost no idea about the shape of the object, so it can only be used as a reversing radar.
Although millimeter-wave radar can detect the speed and distance of the target, it is also difficult to distinguish the shape of the object, and it is easy to "lie about the military situation" in a complex environment
So, in the end, the fusion solution also has to add lidar, which scans the world around it by emitting a laser to measure the distance in each direction with the same accuracy as real-world 3D printing.
But the problem is that it's more expensive.
By 2017, Velodyne's 16-, 32-, and 64-line lidars were priced at $8,000, $40,000, and $80,000, respectively.
If you want to be more intelligent, you have to have more "lines" here. If the wiring harness is doubled, the cost will not go down, and the development of perception technology will basically be stuck here.
So, in order to avoid lidar, Musk made a decision that went against his ancestors: only use cameras.
According to him, people can drive cars with just their eyes, so why does AI need all kinds of sensors?
So at Tesla AI Day 2022, Tesla introduced the Occupancy Network, which iterated the original "real-time strategy" like BEV (bird's-eye view) into the grass block world of "Minecraft", which can basically achieve the effect of lidar.
However, it was originally thought that pure vision could save a lot of money, but it turned out that it was becoming more and more expensive at the practical level.
The reason is that Tesla also needs to train and maintain this purely visual perception neural network.
First of all, in order to meet the computing power required by the algorithm, it is necessary to cut the chip, Tesla has been developing its own chips since 2014, and released the FSD self-developed chip in 2019, with a computing power of up to 600 tops.
And in 2021, Tesla also built its own chip D1 specifically for training to build the Dojo Exapod supercomputer.
So how much does it cost to make this thing, Musk said: it will take at least billions of dollars a year.
But at the same time, in 2020, there was a turnaround on the side of the multi-sensor fusion solution, and due to the localization of lidar, the best lidar on the market was instantly "wiped out".
This year, Huawei's car BU announced that their goal was to "reduce the cost of lidar to $200 or even $100." ”
In the past, Velodyne used mechanical lidar, which was expensive and easy to break.
Now, domestic manufacturers such as Huawei and Hesai use semi-solid-state lidar, and through the rotating mirror technology, the effect of past rotation can be achieved.
What's more, there is a two-dimensional rotating mirror, 1 line can also be used as 100 lines, such as Huawei's equivalent 96-line lidar, which completely reduces the cost of lidar.
Therefore, the cost of the current vehicle-mounted lidar is about a thousand dollars.
So much so that some car companies say, "Less than 4, please don't talk" seems to be wealthy, but in fact it is a ride on the localization of lidar...
Seeing this, in fact, we are clear about the difference between the routes of DJI and Huawei.
Huawei specializes in multi-sensor fusion solutions, and with LiDAR, it is easier to solve the perception problem of intelligent driving, so it can continuously challenge the limits of intelligent driving capabilities.
DJI's purely visual route, which solves practical problems with a lower cost of getting on the car, can be popularized faster, but the problem is that it is more difficult, and it is not yet known what kind of effect it can achieve in the future.
Even Tesla, a pioneer of pure vision, has encountered many problems at this stage.
However, since DJI can be equipped with it on real cars and has been recognized by many car owners, it can at least show that its basic functions are mature. Brother Neck is also very curious about what height it can reach.
After all, no one will choose to add 20,000 or 30,000 yuan to drive intelligently when buying a Qin, right?