On December 31, 2024, Zhuoyu Technology, formerly known as DJI Automotive, unveiled its end-to-end (E2E) smart driving solution designed for mid-tier computing power. This innovative system features hardware that includes seven cameras and a TDA4-VH chip delivering approximately 32 TOPS of computing power, supporting advanced functionalities like city navigation assistance.
This solution targets vehicles priced around RMB 100,000 (USD 14,000), aiming to make advanced smart driving more accessible. Zhuoyu said that automakers using this system can provide free upgrades via over-the-air (OTA) software updates, offering continuous improvements to end users.
Vehicles equipped with Zhuoyu’s system, such as models from Baojun and Chery’s iCar, are expected to feature this functionality as early as the first half of 2025. The company’s efforts align with an industry-wide push in 2024 toward scalable, cost-effective E2E solutions. Major players, including Huawei, Xpeng Motors, and Li Auto, have primarily relied on Nvidia’s Orin X chip platform, which offers a hefty 254 TOPS of computing power as a baseline. Zhuoyu, however, has taken a different path—developing E2E capabilities with only 32 TOPS, significantly lowering the barrier for urban smart driving.
As automakers ramp up the deployment of smart driving technologies in 2025, suppliers like Zhuoyu are refining their solutions to meet shifting market demands. Zhuoyu’s development strategy centers on two distinct chip platforms:
- Compute-intensive solution: Built on Qualcomm’s 8650 chip with 100 TOPS of computing power, this platform supports advanced features such as map-free navigation in urban environments. It is already deployed in FAW Hongqi’s E007 and E009 models.
- Mid-tier solution: Powered by the TDA4-VH chip, this system debuted in September 2023 with highway navigation assistance. It has since evolved to incorporate memory-based driving and E2E capabilities, broadening its functionality and appeal.
Zhuoyu’s advanced edge computing techniques further enhance the deployment of E2E models, enabling high performance even on platforms with limited computing power. According to a technical expert from Zhuoyu, both the mid-tier and high-tier solutions share a unified artificial intelligence model architecture. This design integrates core modules—perception, prediction, decision-making, and planning—while incorporating a safety inference framework for redundancy in trajectory outputs.
Shared architecture and training datasets ensure a consistent user experience across scenarios, whether on a 32 TOPS or 100 TOPS system. This unified approach not only simplifies development but also enhances scalability.
Zhuoyu’s binocular vision-enhanced E2E architecture represents a standout innovation. By incorporating geometric data from both image and point cloud inputs, the system allegedly achieves a 10–20 times improvement in training efficiency. This approach drastically reduces the reliance on extensive long-tail data, requiring only about 10% of the training data needed for conventional monocular or heterogeneous solutions to reach comparable performance levels.
This efficiency promises to reshape how smart driving solutions are developed and scaled, ensuring advanced capabilities can reach a broader range of vehicles and consumers, driving the industry closer to universal adoption of smarter, safer driving technologies.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This article was written by Li Anqi for 36Kr.