DJI Vehicle, with a valuation of about 10 billion yuan, is about to usher in two heavyweight partners - BYD and FAW.
According to 36Kr, DJI has received a clear investment intention from BYD and FAW to jointly develop autonomous driving solutions.
On July 21, 202, BYD has obtained the first domestic autonomous driving road test license, and domestic and foreign car companies such as Wenjie, Xiaopeng, and Mercedes-Benz have also successively obtained automatic timing road test licenses.
The era of autonomous driving is coming, and car companies need partners to join to enhance their strength!
With low cost and strong effect, DJI has become a sweet spot
After years of exploration, car companies and autonomous driving companies have launched a variety of solutions, among which the most secure solution, in addition to high-computing power chips, also requires a large number of cameras, millimeter-wave radars, ultrasonic radars, and a certain number of lidars.
At the 2024 International Consumer Electronics Show, DJI showcased its urban pilot assistance system. The demonstration vehicle is based on a single Qualcomm Ride (SA8650P) platform autonomous driving system-on-chip, 12 cameras, 12 ultrasonic radars, and 1 millimeter-wave radar, which realizes high-end assisted driving at the NOA level in urban areas, focusing on low cost and high profit.
In 2023, Baojun Automobile, a subsidiary of SAIC-GM-Wuling, announced that the Baojun Yunduo 460 will be equipped with the "Lingxi Intelligent Driving 2.0" system jointly created with DJI, which realizes functions such as 0~130km/h full-speed intelligent driving assistance cross-layer memory parking and tracking reversing through 8 million binocular cameras, 4 surround view cameras and 1 rear monocular camera.
According to the information disclosed by DJI Automotive, a minimum of 7~9 cameras with 32TOPS computing power autonomous driving chips can be used to achieve regional memory driving, while L2+ urban pilot driving only needs to upgrade the autonomous driving chip to 80TOPS. In order to be able to adapt to the needs of the autonomous driving era, the chip Thor developed by NVIDIA has a computing power of up to 2000TOPS, which is 25 times that of DJI.
In addition, although some car companies can use low-computing chips to achieve high-end assisted driving, they need the blessing of high-precision maps, and DJI's autonomous driving solution completely gets rid of the dependence on autonomous driving.
The 100,000 yuan ~ 150,000 yuan Baojun Cloud 460 Lingxi Edition, with the blessing of DJI's on-board technology, can achieve high-end assisted driving. BYD and FAW have higher sales of products, and the technology can be further decentralized by evenly sharing R&D and production costs. It is even said that in the future, A0-level vehicles such as Seagulls and Dolphins may achieve high-level assisted driving, and even achieve automatic driving in fixed road sections through memory mode.
36Kr reported that BYD had been in contact with Huawei's automotive business, but the latter believed that there was not much room for cooperation and politely rejected BYD. Xiaotong speculates that it may be that Huawei thinks that the cost cannot be compressed to a range acceptable to BYD, after all, at this stage, Qin PLUS, some versions of the destroyer 05 have supported L2 level assisted driving, and it is inevitable to upgrade to urban pilot assisted driving in the future, but the car at this price is obviously difficult to popularize multiple radars.
Huawei is already a leading LiDAR company in China, so it naturally hopes to promote a more secure converged vision model and not pursue extreme cost compression. However, throughout the automotive industry, low-cost fusion vision without lidar and high-precision maps, or even pure vision solutions that cancel millimeter-wave radar and ultrasonic radar, are gradually becoming the mainstream.
Is security really not as cost-important as it gets?
Pure vision autonomous driving solutions generally refer to the term "binocular vision", which refers to two cameras arranged by human eyes to capture the scene ahead from two angles and use it to build a model of the environment.
The explanation given by a car company is that people can recognize environmental information by relying on two eyes, and cars can also achieve autonomous driving by relying on two cameras. For safety reasons, car companies generally distribute multiple cameras around the car to judge the surrounding environment. DJI's low-cost solution requires 7~9 cameras, which is also to ensure safety.
Car cameras usually have low pixels, only the binocular vision camera is 8 million pixels, and other cameras have lower pixels, some may even be less than 1 million pixels, and the cost of a single camera is about 100 yuan. According to the information given by Debang Securities, the cost of a single traditional millimeter-wave radar is generally 250 yuan ~ 400 yuan, and the cost of a single lidar is as high as 6000 yuan ~ 7000 yuan.
The price of a lidar is enough for a car to be equipped with a set of cameras and autonomous driving chips. The cost of millimeter-wave radar is certainly lower, but under the cumulative sales volume, no matter how small the cost is, it is not a small number. Taking BYD Qin PLUS as an example, 455,800 units will be sold in 2023, assuming that each car has an extra millimeter-wave radar of 300 yuan, the cumulative cost in 2023 will increase by 137 million yuan.
Similarly, high-precision maps can provide centimeter-level accuracy of environmental data, but they require a lot of manpower and material resources to map, and some areas may need to be repeatedly mapped, so the increased cost is too high. Because of this, in recent years, car companies have been promoting intelligent driving without pictures to get rid of the dependence on high-precision maps.
We can understand the cost savings of car companies, but the abandonment of lidar and high-precision maps means that higher requirements are put forward for autonomous driving algorithms. Human judgment of the environment not only depends on the eyes, but also hearing, touch, etc., the human brain is also different from the car machine system, computer neural network can not achieve the real sense of the simulation of the human brain.
Of course, in addition to the cost issue, LiDAR has another disadvantage, that is, it relies heavily on high-precision maps. Vehicles need to collect information through lidar and locate it in combination with high-precision maps to improve accuracy. However, China covers an area of 9.6 million square kilometers, and it is impossible for any company to achieve national high-precision mapping, and high-precision maps cannot become a part of integrated vision in most scenarios.
Lidar wants to get rid of the dependence on high-precision maps unless the chip computing power or computer neural network is at a very high level.
For various reasons, the cost of lidar has gradually decreased, but the sound has not improved.
Car companies are still exploring, and the future has not yet been determined
It is uncertain when autonomous driving will be available, but Level 3 autonomous driving has already been tested on the road and is expected to be commercially available by 2030. As for which plan can become mainstream in the future, it is not possible to give a conclusion for the time being, and car companies still need to continue to explore.
Xiaotong believes that high-end luxury cars are expensive and can be equipped with high-computing power chips + lidar to ensure safety. The mid-to-low-end models can take the hybrid vision route through cameras and millimeter-wave radars to ensure safety while reducing costs as much as possible.
The pure vision solution has a low cost and questionable security, especially in scenarios where the brightness is too low, the light source is complex, and the camera is not as reliable as the radar.
Companies such as Huawei and Hesai Technology are working to reduce the cost of lidar, and their expected goals are less than $200 or even $100. With the continuous development of neural networks, it is also possible that lidar will get rid of its dependence on high-precision maps in the future.
DJI's low-cost solution is suitable for ordinary consumers, and the cloud Lingxi version of more than 100,000 yuan can achieve autonomous driving. With this shareholder wind, DJI's automotive business may not be without the possibility of becoming the next Huawei car BU.