QCraft completes several hundred million yuan in Series C+ financing. CEO Yu Qian: Mass production of end-to-end intelligent driving to begin early next year | Exclusive from 36Kr.

Text | Angel Li

Editor | Qin Li  

36Kr exclusively learned that recently, the intelligent driving company "QCraft" has completed a several-hundred-million-yuan Series C+ round of financing, which is invested by Zhulu Juhang Fund. It is reported that this round of financing is mainly used to increase the R & D investment in full-stack autonomous driving technology and promote the large-scale mass production and delivery of mid-to-high-level NOA solutions.

This is the second time this year that QCraft has received capital favor. In June, QCraft received a several-hundred-million-yuan Series C round of financing jointly invested by Zhongguancun Science City Company and Cuihu Fund.

Up to now, QCraft has completed 7 rounds of financing, and the investors include Lenovo Capital, IDG Capital, Genesis Capital, CICC Capital, Yunfeng Capital, China Merchants Venture Capital, Horizon Robotics, Meituan Dragonball, TCL, etc.

QCraft was founded in early 2019 and is a domestic startup focusing on RoboBus (driverless minibus) and intelligent driving for passenger vehicles. The core team members are from international technology companies such as Google Waymo, Tesla, NVIDIA, Facebook, etc., and the proportion of R & D personnel is as high as 80%.

Intelligent driving is undoubtedly an intelligent selling point that is fiercely competed for by current vehicle companies. Many vehicle companies, on the one hand, are researching and developing their own technical solutions, and on the other hand, they are also using the products of third-party intelligent driving companies to improve the basic intelligent driving capabilities of vehicle models and quickly popularize intelligent driving.

QCraft has also caught this intelligent trend. QCraft said that after the company's intelligent driving full-stack mass production function was designated for a leading new energy vehicle company's intelligent driving mass production project, it has been pushed to more than 400,000 users, and the user mileage has reached hundreds of millions of miles.

According to the data from the Gaogong Intelligent Vehicle Research Institute, from January 2023 to July 2024, in the ranking of domestic passenger vehicle NOA high-level intelligent driving system solutions, QCraft ranked first with a market share of 50.84%.

QCraft CEO Qian Yu said that in large-scale mass production practices, its intelligent driving technology and product experience can be continuously optimized. QCraft claims that currently, its high-speed NOA can achieve a safe takeover level of over 1,000 kilometers, the elevated section in congested road conditions can achieve a safe takeover level of over 200 kilometers, the lane change success rate is 99.5%, the lane change efficiency is 92.6%, and the urban LCC intersection pass rate is 97%.

In order to further expand the market, QCraft has created three sets of mid-to-high-level NOA intelligent driving solutions: Chengfeng Air, Chengfeng Pro, and Chengfeng Max, hoping to leverage vehicle companies and users with the ultimate cost performance and performance experience.

For example, Chengfeng Air, based on Horizon Journey 6E and equipped with 7V (7 cameras including 4 fisheye cameras), enables 100,000-yuan-level vehicles to be equipped with high-speed NOA functions; while Chengfeng Pro adopts 11V, targeting 100,000 - 150,000-yuan-level vehicle models, and has urban NOA functions such as urban memory driving on the basis of high-speed NOA;

Chengfeng Max is based on Journey 6M, adopts 1L (that is, one lidar, optional) and 11V, does not rely on high-precision maps, and brings the full-scenario urban NOA to 150,000-yuan-level vehicle models.

However, in April this year, QCraft's full-scenario urban NOA was mainly targeted at the 200,000-yuan market. Regarding the adjustment of the product's target market, QCraft CEO Qian Yu told 36Kr that thanks to the continuous optimization and maturity of the technology after the previous large-scale mass production and delivery, the cost of the solution can be further reduced.

Qian Yu predicts that in the future, high-speed NOA will be fully standard on vehicles around 100,000 yuan, and urban NOA experience will basically be standard on vehicles around 150,000 yuan. "A better cost-effective product solution is what we have always insisted on, and this will help us finally complete the business closed loop from L2+ intelligent driving to L4-level autonomous driving."

Regarding this round of financing, QCraft CEO Qian Yu accepted an exclusive interview with 36Kr.

Currently, end-to-end and other technologies have become the technical highlands that top intelligent driving players are competing for. Qian Yu also told 36Kr that end-to-end technology is an inevitable technical trend. Whether it is L2+ or L4, end-to-end is the only way.

For this reason, QCraft has done a lot of development work on end-to-end, and it is expected that the end-to-end capability will be demonstrated by the end of the year, and the end-to-end intelligent driving can be mass-produced in early next year.

Qian Yu believes that more efforts for end-to-end should be made in the cloud, not just in the vehicle.

He believes that the training of end-to-end is not simply solved by stacking cloud training cards, but depends on the training efficiency; the more critical challenge is the quality and coverage of the training data, as well as the more efficient data processing capability.

He believes that the intelligent driving model is ultimately gradually converging, and it is necessary to efficiently handle and make good use of the data, and promote the development in a data-centered development method. And the data-driven development paradigm that QCraft has accumulated previously will enable the company to continuously iterate products and provide a better intelligent driving experience.

The following is the interview record between 36Kr Auto and QCraft CEO Qian Yu, slightly adapted based on the original meaning:

36Kr Auto: After doing the project for the top major customers, QCraft's market share in intelligent driving is increasing. Can this indicate that QCraft has established a firm foothold in the intelligent driving field?

Qian Yu: Currently in the mid-to-high-level NOA solution, we have achieved the highest market share. We are still continuously iterating and upgrading. From Horizon J5 to J6, we are the industry benchmark. We are the first designated for the J6 series, and it is expected to be the first mass-produced and delivered on the J6 platform. Many potential customers highly recognize QCraft's large-scale delivery, stability, reliability, and product experience.

Autonomous driving is a long-term competitive track. As a startup company, we always maintain a prudent and optimistic attitude. On the one hand, the competition is fierce, and the growth of intelligent driving is very rapid. On the other hand, in terms of technology, everyone is constantly innovating, and the product experience is constantly improving. QCraft has made a good start, but in the long term, we need to continuously work hard on product experience, technology, and cost performance in order to maintain the lead.

The mass production scale is very important. Some suppliers without large-scale delivery are more difficult. Because autonomous driving is driven by data, the data scale is very important, and we have a very good position. At the same time, QCraft's ecological cooperation circle is very good. We collaborate with Horizon Robotics and NavInfo to win the battle of mass production together.

36Kr Auto: From the beginning of this year to now, the price range of QCraft's full-scenario NOA solution has changed, from targeting the 200,000-yuan market to targeting the 150,000-yuan market. Is this change to respond to the price war?

Qian Yu: This is not a simple price war issue. We hope to allow more consumers to experience better and safer products, so we are going down to a lower market. In fact, the entry of the full-scenario NOA into the 150,000-yuan market is the embodiment of our large-scale mass production and delivery, continuous optimization and maturity of technology, and also the result of the strong alliance between QCraft and a wide range of ecological partners. With the maturity of technology and the expansion of mass production scale and efficiency improvement, the cost of the solution is gradually reducing.

In the future, high-speed NOA will be fully standard on vehicles around 100,000 yuan, and urban NOA will basically be available on vehicles around 150,000 yuan. This trend is in line with our expectations for technology popularization and commercialization. But in terms of safety, whether it is medium or high configuration, QCraft's philosophy is top configuration.

36Kr Auto: The sensor configuration solutions and positions of different brand vehicle companies are not the same. If QCraft makes solutions for other vehicle companies, what work does the intelligent driving system need to re-do? What are QCraft's experiences in solution migration?

Qian Yu: QCraft's intelligent driving solution emphasizes platformization. Based on our virtual camera technology, we can adapt to different camera configurations from 6V to 11V, including installation positions, sensor models, etc., with a strong generalization ability. This is also due to the larger intelligent driving model, which has better generalization than the previous low-computing power platform. In terms of platform migration, QCraft has very mature and efficient experience.

For example, our rich experience on the Journey 5 platform enabled us to complete the full-stack function development and real vehicle deployment of Journey 6 within two weeks.

36Kr Auto: Currently, end-to-end and other technical solutions are very popular, and the progress of the top new energy vehicle companies is also very fast, and the resources are more abundant. For new technologies, what deployment plans or actual investments does QCraft have?

Qian Yu: Regarding end-to-end, it is difficult to say that vehicle companies must do better than third-party companies, because the accumulation of different vehicle companies is different. Although end-to-end is a newer R & D paradigm, it is not completely separated from the previous intelligent driving accumulation and the underlying data-driven capability. In these aspects, we have a very extensive accumulation, and we have also done a lot of development work on end-to-end. There will be an end-to-end capability demonstration at the end of this year, and it can be mass-produced in early next year.

More efforts for end-to-end are not in the vehicle, but it largely depends on the data processing capability, and the efforts are in the cloud. On the other hand, its use of vehicle computing power is actually more efficient than the previous rule-based approach. QCraft has a strong data-driven development paradigm, which allows us to maintain the lead in the iteration of new technologies.

36Kr Auto: Computing power resources are a big challenge for end-to-end. How can third-party intelligent driving companies make more efforts in the cloud? How to raise more computing power resources?

Qian Yu: End-to-end does indeed require a lot of cloud accumulation, but it is not simply solved by stacking cards, and the training efficiency is very important. Data is not the more the better. The scale is one aspect, but the quality and coverage of the data are also very critical. Also, the use of simulation is very important. QCraft's accumulation and reserves in these aspects can fully play a role.

36Kr Auto: What are the thresholds and challenges for different players to do end-to-end?

Qian Yu: I think the main threshold challenge is whether there is a large enough mass production scale and enough mass production data. Other technical challenges are secondary. Everyone always talks about computing power, but I think this is only a short-term situation. Because the model is constantly optimized and new algorithms appear, the use efficiency of computing power is constantly improving.

For example, from 2012 to 2015, when GPUs were not yet widely used in deep learning training, many Internet companies used CPU clusters for training, so the use of CPUs was huge, and almost no one could afford it. Network training also takes a long time, but now, a single machine with a small GPU capacity can complete the same training volume in a very short time.

So after the emergence of new technologies and the improvement of the efficiency of new training methods, the demand for computing power will gradually converge, and it is not necessary to concentrate all computing power together to train well.

36Kr Auto: Data is also a very important factor. Are vehicle companies willing to open data to intelligent driving companies now? To what extent?

Qian Yu: Generally speaking, if the opening of data is very important for the product development experience, it will definitely be opened. Vehicle companies have data, but if the role of data cannot be maximized, it is not helpful to improve the product experience. Therefore, vehicle companies and suppliers will find a model that can give full play to everyone's advantages to the overall best.

36Kr Auto: Does the data of different vehicle companies need to be isolated?

Qian Yu: Data must be isolated, and the data itself cannot be used across vehicle companies. But the experience can definitely be cross-applied, and the experience is our technical precipitation and accumulation.

I think that the domestic intelligent driving does not lack data, but more lacks the data processing capability and the ability to maximize the value of the data. This is the biggest threshold. After all, having data does not necessarily mean having knowledge.

This involves the tool chain and many underlying accumulations and know-how. Now the entire autonomous driving has transformed from model-driven to data-driven. Because the model is ultimately gradually converging, it is more about how to make good use of the data, and the development method centered around the data is advancing.

36Kr Auto: Some time ago, Tesla's driverless vehicle attracted a lot of attention. Do you think that the intelligent driving based on end-to-end has the opportunity to develop to L4 autonomous driving?

Qian Yu: End-to-end technology is an inevitable technical trend. But the product forms of L4 and L2+ are different. L4 requires a lot of safety backups, and the redundancy in safety and other aspects is inevitable to prevent some system errors. Whether it is a sensor or a computing platform, the entire safety mechanism design is required.

But can end-to-end solve all the problems of L4? I don't think this exists, but it will bring the entire autonomous driving capability to a very high level, so that the places that need redundancy supplementation will become fewer. Therefore, whether it is L2+ or L4, end-to-end is the only way.

36Kr Auto: After this round of financing, will there be any changes in QCraft's overall strategy?

Qian Yu: Our overall strategic goal is very clear and firm. In terms of mass production, we will gradually expand the scale, deepen the mass production advantages, and continuously improve the product experience through the continuous accumulation of technology, experience, and data to serve a wider range of consumers. A better cost-effective product solution is what we have always insisted on, and this will help us finally complete the business closed loop from L2+ intelligent driving to L4-level autonomous driving.