The DJI motor business, which has been in preparation for 6 years, has finally been handed over. The process is not tortuous, there have been rumors of car manufacturing, and the cooperation with the world's largest car company Volkswagen Group has been disclosed early, but the first model delivered by DJI vehicle is the domestic Internet celebrity tram brand Baojun KiWi.
The intelligent driving system jointly developed by DJI Automotive and Baojun KiWi is called Lingxi Intelligence, and the main sensor configurations are: a set of forward-looking binocular cameras, 4 circumferential fisheye cameras, 1 forward millimeter wave radar, and 12 ultrasonic radars. Some industry insiders assessed to 36Kr that the cost of this set of hardware ranges from 3,000 yuan to 5,000 yuan, and it is almost impossible to buy a lidar.
But it provides a full range of functions, there are currently adaptive cruise, light lane change, with the car start and stop, large curvature cornering, auxiliary parking and a series of functions, almost in addition to high-speed navigation, the rest of the mainstream intelligent driving functions have been equipped. In addition, from the information revealed by DJI, there are still a series of functions waiting for OTAs.
DJI Vehicle and Baojun jointly launched the Lingxi system, pictured: live shooting
36Kr participated in the nearly two hours of real car experience, Lingxi intelligent product completion is also not bad, with the car, deal with emergency congestion, and the car parallel when the active avoidance, too large curvature bending, traffic jam under the start and stop, etc. are very smooth, in terms of parking assistance, parking space recognition rate, narrow parking success rate has also reached the practical level.
Baojun KiWi is an upgraded version of Wuling MINI EV. When the head car manufacturing company sold tens of thousands of vehicles per month to support the market value of tens of billions of DOLLARs, Wuling launched an electric vehicle brand MINI EV of less than 50,000 yuan, and the monthly sales rushed to 50,000 units in a short period of time.
In order to control the cost, the starting model of this car was not even equipped with airbags and air conditioners, so the market did not have much expectation for the intelligent driving performance of this car. After all, in the propaganda of car companies, intelligent driving technology has long been bound to expensive lidar, large computing chips and so on.
But from DJI car seems to give the industry another answer: low-cost car and hardware configuration, can still provide a fully functional intelligent driving system.
Passed the mass production barrier
"Do you have any experience in mass production?"
Since the beginning of this year, almost all L4 level autonomous driving companies have begun to enter the mass production preload business, Xiaoma Zhixing, Wenyuan Zhixing, Qingzhou Zhixing, Yuanrong Qixing, etc., but when they sit at the negotiating table of car companies, they have to face the above soul question.
Especially for large car companies, it is difficult to hand over a model to a start-up technology company with no mass production and engineering experience. The concerns of car companies are justified, a backward modified Robotaxi autonomous vehicle can smoothly travel through the city, because of the use of expensive sensors, unrestricted computing chips, but in the mass production car, hardware costs, chip computing power and so on are challenges.
A car company employee told 36Kr that the computing power of mass production cars is often only a few TOPS, and the algorithm of the automatic driving technology company needs to do a lot of optimization to run smoothly, "otherwise it will soon fill the memory when it runs up", and it is not uncommon for it to spend half a year on this optimization work.
Therefore, DJI is lucky to be backed by DJI's hardware and software capabilities and smoothly cut into the supply chain of Volkswagen and Wuling.
For example, the Lingxi system cooperated by DJI and Baojun this time uses the industry's rare binocular visual perception technology, and its special feature is that it can use the point cloud algorithm of binocular fusion to cloud the pixels and obtain a geometric information including depth information to achieve the detection of obstacles in front.
The advantage of this is that when the visual recognition of the type of obstacle is still relatively limited, the type of obstacle can be detected first, and then the obstacle avoidance treatment can be made.
The application of this technology is not common in the industry, but it is no stranger to DJI. DJI drones use binocular cameras for obstacle avoidance, and they have a muscular memory of binocular stereo vision technology.
Usually, the visual sees the image and gets more two-dimensional planar information, while obtaining three-dimensional information through point clouds is the specialty of lidar sensors. But DJI, based on its own algorithmic capabilities, has allowed low-cost binocular cameras to play the function of "lidar".
Of course, in addition to the algorithmic ability, this technology will also require engineering ability. As an executive at DJI Automotive said, "We have more than 10 million intelligent systems with mass production capabilities, and we have experience in the rigid connection design, consistent design, and high-precision calibration of the production line of the binocular camera, and my hardware can carry the algorithms developed by myself." ”
Lighting lane change is also an unexpected function given by the Lingxi system. The premise of the realization of the light lane change is to complete the detection of vehicles in the surrounding lanes at high speed, and the industry's common practice is to use angle millimeter wave radar, but the DJI Lingxi system obviously does not have this hardware, only a fisheye camera.
Fisheye cameras have short detection distances and are mostly used for low-speed parking, but DJI has tied the algorithm and hardware to make them available for both parking and lane target detection. Such technical ideas are also the reasons why DJI can use low-cost hardware to make more intelligent driving functions.
Proposing new technical directions and implementing applications, as well as using new product designs to achieve more functions at a limited cost, are all manifestations of DJI's engineering capabilities. This is also why when peers are still begging for car companies to designate projects, DJI has already come up with mass production products.
But it is clearly not an insurmountable barrier.
Competing for the front load, the commercial challenge is greater than the technical challenge
When the industry competes to compete for mass production models, there are actual delivery cases, which is certainly a good entry point. But self-driving companies can also achieve mass production through capital or alliances.
Just like the unicorn company Momenta, it has long received capital injection from SAIC, and is currently cooperating with SAIC's Zhiji Automobile to gain experience in mass production engineering. Wenyuan Zhixing also has a strategic cooperation with bosch, a leading automobile supplier, hoping to use the latter's automotive engineering experience to get on the bus.
DJI's mass production engineering capabilities have a first-mover advantage, but this is also more of an entry condition for the industry, and players who enter the field must pass this level.
In addition, it should not be ignored that DJI And Wuling have established a high degree of tacit cooperation, and Wuling is willing to be the product manager, allowing DJI to deeply integrate software and hardware to ensure the smooth realization of technical capabilities. But then, when facing the entire automotive industry, DJI's on-board software and hardware integration capabilities may also be hidden challenges.
Intelligent driving is regarded as the core competitiveness of the automotive industry in the future. Today's new car camp, Weilai, Xiaopeng Automobile, Ideal Car, and later Xiaomi Automobile, Extreme Krypton Automobile, Jidu Automobile, etc., are pursuing self-research in intelligent driving technology, and the old car companies Great Wall Motors and BYD are also recruiting.
This in itself is compressing the market capacity of third-party smart driving technology companies. And those car companies that have not developed their own and still purchase supplier solutions have also begun to strengthen their control.
SAIC, BYD, etc., as early as the case of industry cooperation, required the supply of flash to decompose the software and hardware, that is, the algorithm and sensor are independently purchased.
At this time, can the advantages of autonomous driving technology be better exerted? An employee of an independent car company that purchased the algorithm told 36Kr that their supplier was a well-known technology company in the industry, but when they asked for the use of the specified camera, the technology company's algorithm was obviously difficult to adapt, but instead required the car company to replace the windshield.
Of course, this is a technical challenge, and the commercial challenge behind it is that DJI's rival Huawei in the same city is precisely because of the pursuit of strong binding in software and hardware, which has triggered the vigilance of SAIC Chairman Chen Hong that "Huawei is the soul, and the car company is the body".
In the competition for smart electric vehicles, especially the car players in the main track, there is no doubt that they will strengthen their control over research and development and engineering resources. The efficiency of independent control is also obvious, taking the high-end intelligent driving function high-speed pilot assistance as an example, in addition to the self-developed Tesla, Weilai, Xiaopeng Automobile and Ideal Car, other car companies have not yet been launched.
In this context, intelligent driving technology suppliers should want to open up a larger market space, and they need to establish a set of benign and sustainable development cooperation mechanisms with car companies. And this time window for exploration is obviously not long.