China Intelligent Driving Tier 1 — DJI Automotive
SAIC Motor’s chairman made an interesting remark back in 2021 that cooperating with Huawei on intelligent driving would make Huawei the soul and SAIC the body. It was indeed the aspirations of many auto OEMs to build intelligent driving capabilities in-house.
Despite the “soul theory” circulating ever since, most of the OEMs still have to outsource intelligent driving or at least certain modules to suppliers. Furthermore, tech companies and startups are becoming the new generation Tier 1s for intelligent driving systems. This article is the first of a new series to introduce these local Tier 1s.
DJI, the world’s largest drone maker from Shenzhen, China, has been working on autonomous driving technologies for several years. The team was very low-key until the recent newsbreak that over 20 vehicle models will hit the road this year with DJI’s intelligent driving systems.
City Navigation on Autopilot (NOA), more or less an equivalent of Tesla’s FSD, has been a luxury feature limited to certain vehicle models priced above RMB 250,000. However, affordable cars cost below RMB 200,000 account for over 50% of the passenger car market in China. While other suppliers compete fiercely in the mid to high-end segments, DJI identified another route to Rome by targeting the mass market that is blank when it comes to intelligent driving.
DJI has built several low-cost solutions with single entry-level computing chips and minimum amount of affordable sensors, which can be easily adopted by cars with price tag ranged from RMB 80,000 to 200,000. This is as opposed to the rivals that have developed similar functions on the more expensive Nvidia Orin processors with a fancy suite of sensors.
Chengxing Basic Configuration (7V + 32 TOPS)
DJI named its intelligent driving solutions as Chengxing that literally means powering mobility. The basic version was first implemented by Baojun Yunduo model from SAIC-GM-Wuling (SGMW), a General Motors China joint venture, which went on sale in September 2023.
Chengxing basic version is built on the TI TDA4VH of 32 TOPS with 7 cameras only. It features an IMU and 8MP dual-camera stereovision integrated system for front sensing, four 3MP fisheye cameras for surrounding view and a 3MP/8MP camera for rearview.
Without HD map, the system enables active safety features like AEB, FCW, LDW, LDP, BSD, DOW etc., driving assist, parking assist, home zone parking (cross floors), highway NOA and city memory NOA. The solution cost is claimed to be about RMB 5,000.
Besides Yunduo, two new models have been launched so far of this year carrying Chengxing Basic version, which are Chery iCar 03 and SGMW Baojun Yueye Plus. All the models are priced below RMB 150,000, recognized as DJI’s initiative to democratize top-end intelligent driving features to the vast consumer base.
Chengxing Upgraded Configuration (7V/10V + 100 TOPS)
DJI introduced a new version of Chengxing in March that copied the exact same sensor configurations onto Qualcomm SA8650P with 100 TOPS of performance, enabling city NOA not limited to memorized routes as in the basic version.
The upgraded compute platform will be able to host a series of key technologies such as transformer based BEV (Bird’s Eye-View) and PnP (Prediction and Planning), dual-camera based 3D occupancy networks, road topology model, etc. While it is commonly believed that the above-mentioned technology frameworks would require at least one Nvidia OrinX of 200 TOPS, DJI team made it work through extreme hardware performance squeezing and algorithm optimization.
A typical sensor configuration for city NOA includes at least 11 cameras, multiple mmWave radars and ultrasonic radars and at least one LiDAR. However, Chengxing upgraded version achieves it through high reuse of the 7 cameras, which can be extended to 10 for improved experience in certain road scenarios. DJI addressed one of their design principles as strong expandability, making it easier to plug in additional sensors per OEM diversified needs.
The “7V + 100 TOPS” solution is stated as RMB 7,000, poised to equip passenger cars in the RMB 150,000 segment. It will be implemented on vehicle models in Q3 of this year, among which could be two of Volkswagen models for the Chinese market that are SAIC VW Tiguan L Pro and FAW VW Magotan.
Speaking of the two VW models that are both gasoline-powered, this is another of DJI’s initiatives to accelerate the adoption of intelligent driving systems on mainstream models of various powertrain solutions.
Chengxing Advanced (LiDAR-Stereo One-Box Solution)
DJI demonstrated a LiDAR-stereo one-box solution behind windshield during the very recent Beijing Auto Show. It is part of DJI’s advanced intelligent driving solution aiming at high-end vehicles in the RMB 250,000 price range for late 2025 or 2026. Together with a heterogeneous redundancy domain controller currently in development, this version is expected to achieve L3 autonomous driving.
We may have had the impression of DJI taking pure vision approach from camp Tesla until now. The automotive division’s representative Shen explained to media that the idea wasn’t to create another LiDAR to compete with many others. During the product definition phase, the team had in mind the ultimate goal as maximizing the overall intelligent driving system performance, resulting in certain trade-offs on LiDAR parameters. The one-box solution is a step ahead of data-level multi-sensor fusion as the cameras and LiDAR are pre-calibrated and synchronized.
With additional benefits such as flexible behind windshield integration and data processing module centralized to the domain controller, DJI expects the cost to be comparable to a standalone LiDAR product on the market.
When asked about how DJI differentiated its offering with high price performance ratio, Shen owns it to engineering capabilities and the corporate culture addressing engineering capabilities. The team simply followed a direction shaped by the company’s mission, advantages and resources at hands.
After a closer look into DJI history, it does make sense as the perception technologies involving combinations of IMU, stereovision and LiDAR have had presence on drones. Obviously, the core technologies required to navigate surroundings and self-control maneuvers are transferrable between drones and cars. Meanwhile, it is not unprecedented as Skydio, another drone-maker in the US, was started with a lot of Tesla’s early Autopilot software engineers.
