laitimes

Shen Shaojie, the person in charge of DJI Vehicle: 7,000 yuan is not a figure for intelligent driving, hoping to achieve equal rights in science and technology

author:Intelligent driving network
Shen Shaojie, the person in charge of DJI Vehicle: 7,000 yuan is not a figure for intelligent driving, hoping to achieve equal rights in science and technology

On March 17, DJI Automotive Shen Shaojie shared some thoughts on DJI's efforts to move towards the era of intelligent driving standards at the forum of the China Electric Vehicle 100 Association. The following is the content of Shen Shaojie's speech, which has been edited and edited:

DJI's corporate mission is to provide a safe and easy travel experience for everyone, what is "everyone", I think if it can be standard for all models, it is everyone, and then I would like to share a little bit of our thinking in this regard.

First of all, you can look at some data, everyone here should be happy to see this data, with the joint efforts of all of us, intelligent driving-related functions have risen from 14th in 2021 to 5th in 2023 in the decision-making priority of consumers buying cars, in any case, this is a very exciting thing. Of course, this is the result. We can look at moving up from 5th place, or it doesn't have to be to 1st, what else do we need to do if we want to move towards standard?

Take a look at a few data: as the absolute mainstream of the automobile market must be from 80,000 to 300,000 mainstream models, expand these mainstream models to see the configuration rate of intelligent driving-related systems inside, in the past so many years, from 2021 to 2023, there are such data, with the joint efforts of everyone, the configuration of more than 300,000 new energy models for high-end intelligent driving has been very high, and the high-end + low-end has reached a state of close to 100%. But in fact, if we look back, cheaper cars, cars below 200,000 level, the configuration rate of high-end intelligent driving is close to 0, and all the configurations related to intelligent driving are L2.

New energy vehicles are a little better, if you look at fuel vehicles in the future, no matter what price of fuel vehicles, the configuration rate of high-end intelligent driving is close to 0. Although electric vehicles are the overall trend, for consumers who buy fuel vehicles for various reasons, is it possible for us to give them some opportunities to feel the benefits of intelligent cars?

If the standard configuration is realized, the function should reach the stage where consumers feel that it is easy to use and love to use.

For a long time, in the low-level stage, most of the assisted driving functions, whether in the structured road of driving or in parking, can only solve relatively simple traffic scenarios, but under the actual traffic conditions in China, we usually encounter more of the latter kind, whether it is a moderately unruly and more challenging scene, or in the future, this completely unruly more challenging scene, if a high-level system can solve the problems in these scenarios, so that everyone feels usable, Easy to use, this is the watershed of intelligent driving towards the standard.

The previous talk is about "what is the task", and the key to achieving the task is technical maturity and easy function. Let's take a look at the little homework we can do to get the job done.

The basic version of DJI's current platform is called 7V+32TOPS, for those who are not familiar with the solution, I can try to explain:

The so-called "7V" is a pair of 8 million pixel inertial navigation binocular with a front view, plus a 3 million pixel monocular with a rear view, plus 4 3 million pixel surround view, a total of 7 cameras; "32TOPS" is based on Ti (Texas Instruments)'s TDA4-VH 32TOPS machine learning-related computing power, we are also the world premiere of TDA4-VH.

Such a configuration seems to us to be a very clean and beautiful configuration, 7 cameras plus medium computing power domain control, if you want to configure millimeter-wave radar can be configured 0 or 1, ultrasonic radar can be matched or not, the whole system does not depend on high-precision maps, and we have also successfully achieved no division of power, oil and electricity, even in the oil truck on the various actuator limitations, the system can also have a relatively good performance.

Although the whole system is a little "sparrow is small" feeling, but after a lot of efforts by our engineers, it can achieve a series of functions such as high-speed piloting, urban memory piloting, memory parking, etc., before someone said that DJI car is best at "carved sesame", this sentence I think half of the evaluation is right, in fact, what we really do is not carved on sesame seeds, but carved forest on watermelon, after all, sesame seeds are too small.

This system represents our engineering capabilities to make high-level functionality easy and usable with very limited resources.

Compared with the all-in-one machine with a lower cost but not too much lower, plus four surround views, plus ultrasonic, millimeter wave and other accessories, adding up to three or four thousand, the overall cost of the entire 7V+32TOPS is about 5000, for such a price difference, a qualitative change has been achieved in the entire performance. Especially under the treatment of extreme scenarios that must be possessed by high-level intelligent driving, such as static obstacle braking and handling of some construction road sections, our system can be rated as a leapfrog or surprise by the industry, and to a certain extent, it can reach a technical point comparable to more than 200,000 or even 300,000 vehicles. So we think this system is still relatively beautiful. In addition to the beauty, the whole system can also be done on the car fast, based on the existing models of the entire adaptation workload of about 3 months, the new model landing 6-9 months or so, I believe this represents the overall technology maturity, the cost is about 6000 yuan, we boasted of Haikou last year, this year is realized.

At present, there are models, including the Cloud Lingxi Edition, iCAR03, and Yueye Plus are equipped with 7V+32TOPS, and we believe that this configuration is the starting point for achieving high-end driving configurations, and more than 20 models will be equipped with DJI's in-vehicle systems in 2024.

Although this system provides a new cost-effective option for high-end intelligent driving, from our point of view, we feel that it is still a bit imperfect, why? because it cannot achieve any point-to-point city navigation, and in the absence of a map, this computing power cannot be done.

So we think about what kind of increment needs to be done in this system in order to make the city pilot, which represents the ceiling of high-level driving and is also a compulsory course?

Upgrade to 100TOPS, we are currently using Qualcomm SA8650 chip, and the corresponding localization chip configuration is also underway, the whole system can have the ability to lead the city, the upgrade is to change the upgrade of a domain control, the configuration of the entire computing power sensor is still 7V, the domain control is also based on the pin to pin replacement scheme, in other words, the current 32TOPS domain control model, in the future, if necessary, the model function can be directly upgraded by replacing the domain control, which is 7V+100TOPS scheme. At the same time, in order to make the city pilot have faster landing ability and better adaptability to various road changes, we also customized a standard definition map embedded in the domain control with a head map dealer, which is not high-definition, or the standard definition map used by mobile phone navigation companies, why should the map be embedded in the domain control, first of all, there is no need to dig the corresponding map information from the car machine, and the engineering uncertainty brought by the landing adaptation of the whole model will be greatly reduced. This is also the action brought about by a series of thoughts such as how to get on the bus quickly, as I just said.

Is this configuration a pie? How much is it? Direct price, 7000 yuan, the hardware cost of the whole system is 7000 yuan, when can you get on the car? Before the Beijing Auto Show in April, our 7V+100TOPS program will be open to test drive experience, and everyone is welcome to visit the site.

To sum up, the overall configuration of DJI vehicles in the L2+ stage is probably such a matrix, with 7 camera configurations, 32TOPS and 100TOPS, the only difference is whether it has city pilot, and it will be open to experience in April. As the flagship version of the city pilot, it is based on the same computing power but has a configuration of 10 cameras, and there is an additional telephoto camera in front, which is mainly used to deal with the detection of traffic lights at very wide intersections, to ensure that the traffic at large intersections is smoother, and two cameras are made next to it to make the decision-making space larger, which can be used as a flagship configuration, 10 cameras, 100TOPS computing power, and the number of cameras is 7 or 10, and the corresponding models will be mass-produced within this year.

I've just talked about something more recent, but I'm actually very honored to use this platform to introduce something a little more innovative.

As I mentioned earlier, the installation rate of high-end intelligent driving fuel vehicles is close to 0, and we are trying to break through the limit of 0, and the SAIC Volkswagen Tiguan L Pro, which will be mass-produced and launched this year, will be equipped with DJI's on-board system, which should be able to bring a new high-end intelligent driving experience to the majority of fuel vehicle consumers. I don't know if this is an exaggeration, but it can be said that the Tiguan L Pro, which is equipped with DJI's high-end intelligent driving system, may be regarded as one of the smartest gasoline cars, and it should also be regarded as our efforts to promote the integration of gasoline and electricity.

At present, most of the solutions are based on pure vision, and even pure to the point that even millimeter-wave radar and ultrasonic radar do not have pure vision, here we want to sing a little bit of the opposite, at this year's Beijing Auto Show, we will release the eye-catching system, in fact, it is the inertial navigation trinocular and then integrate a full-chain self-developed lidar, this lidar has more design ingenious places: first of all, it may not be the front fusion, but the front front fusion, because in the design stage, we think about how to combine the advantages of vision and lidar, how to use the performance of calibration synchronization and other aspects, combined to achieve the overall optimal effect. At the same time, this system is installed in the windshield, and the cleaning problem can also be solved.

In the process of lidar design, we have done a lot of thinking, how to make trade-offs on the performance of lidar, will not waste performance, but also can overcome the performance shortcomings of all aspects of vision, the whole, and for some places to reduce the performance can also bring performance optimization to the cost, the overall selling price cost should not exceed the current number of separate lidar, but it is an assembly, we are more confident that the performance of the whole system far exceeds a single lidar, and even many solutions of separate assembly. The entire plan will also be officially displayed at this year's Beijing Auto Show.

Finally, people who do intelligent driving know L3 and L4, and we are also conducting research and development on the domain control of heterogeneous processors and processors that are expected to be functionally safe in the future.

I would like to take this opportunity to share the evolution trend of different prices and different configurations related to our plan in the next few years.

In 2023, the low configuration of most models is an all-in-one solution, and the high-configuration, relatively expensive models are actually based on the lidar solution of large computing power models, and we first opened up a 7V+32TOPS solution on the cloud Lingxi version to achieve the function of high-end intelligent driving, which is a little bit of a door. In 2024, as just mentioned, our plan based on 100TOPS urban navigation capabilities will be configured on 8-150,000 and 15-250,000 high-end models, and at the same time, there will be 10V flagship solutions on more expensive models, and the green area is the range of solutions we can provide.

In 2025, we expect that through a large-scale cost reduction strategy, the 7V solution should have some hope to gradually replace the all-in-one machine in cheaper models. Other functions will also be decentralized.

By 2026, the eye-lifting system and L3-related solutions will go up, there will be a certain hope, we write a little bit of cowardice when writing PPT, I think this eye-lifting system can be mass-produced in 2025, and we are also very much looking forward to working with our car partners to promote the implementation of this very distinctive integration solution.

Read on