The cost of using DJI's replacement Magotan has risen by 7,000, but Zhijia is better than Tesla?

In April, the Beijing Auto Show will be unveiled at the new generation of Magotan, and its intelligent strength has caught up with new energy vehicles.

In order to meet the needs of consumers in the domestic market, the intelligent driving solution provided by DJI has been adopted at the intelligent level. The total hardware cost of this solution is about 7,000 yuan, and the solution used should be the 7V+100TOPS perception hardware solution in the DJI intelligent driving system (from hardware to software, it is provided by DJI full-stack).

The functions that this system can meet include high-speed NOA, navigation function of non-map urban areas, etc. In other words, the ability of this configuration can basically realize the perception hardware capabilities of new energy vehicles without lidar.

So, what will be the effect of this configuration after it is put on the car, and how easy is it to use?

7,000 dollars for something, than

tesla

OK?

This plan will be successively equipped with about 150,000 level cars this year, and DJI's hardware cost to the main engine factory is 7,000 yuan. Since DJI's hardware cost to the OEM is 7,000 yuan, the price that the OEM gives to users will definitely not be a figure that is difficult for users to accept when they buy a model of 150,000 yuan.

To make a comparison, now new energy vehicles, including brands such as Wenjie and Xiaopeng, will cost at least tens of thousands of yuan to open the right to use the intelligent driving function. The price of this configuration will most likely not be as high as that of new energy manufacturers.

The hardware configuration given by DJI includes 3 models. 7V+32TOPS basic version,Then it's 7V+100TOPS upgraded version,The difference is that it has a Qualcomm chip with better computing power;High configuration10V+100TOPS is just an increase in the number of high-definition cameras,That's all,And this will also come out with a version with lidar。

It does not extend the configuration of the low-configuration and high-configuration schemes, mainly saying that the perception scheme suitable for the use of 150,000-200,000 level models in the middle configuration. 7V+100TOPS solution, capable work:

City Pilot + City Memory Pilot and High-speed NOA to meet the needs of urban commuting and intelligent driving in high-speed sections;

Memory parking and cross-floor parking;

Functions such as Driving Assist, Parking Assist and Active Safety.

The above functions are realized, and the hardware behind the support is only 7 cameras + Qualcomm Snapdragon 8650 chip. Seven cameras, including a pair of 8-million-pixel front-looking inertial navigation binocles, a 3-million-pixel rear-looking monocular, and four 3-million-pixel surround-view cameras.

There is no concept of column configuration, and it is more intuitive to put a comparison.

It's all pure visual solutions,DJI's set is only one less camera than Tesla'sHW3.0,But the former's camera pixels are comprehensively crushed TeslaHW3.0The camera in the system1.2 million pixels,The difference in chip computing power44TOPS;And the functions realized,It's more abundant than HW3.0,Mainly more urban NOA functions。 The optimization of the algorithm at the technical level includes the road topology perception model based on Transformer, the 4D scene based on binocular + omnidirectional fisheye, and the prediction and decision-making joint large model based on Transformer.

This algorithm is relatively close to the end-to-end logic of Tesla's FSD V12, and the difficulty of HW3.0 to upgrade HW4.0 is very great, and there is basically no possibility. Even if the new FSD V12 is OTA on HW3.0, the camera's low pixel will affect its use.

From this point of view, the features and advantages of DJI's latest intelligent driving solution include low computing power threshold, simple and efficient hardware configuration, and 7V+100TOPS can achieve the effect of surpassing Tesla HW3.0, at least in terms of configuration parameters and algorithm logic.

Cost-effective solution, can it be used?

At present, this set of 7000 yuan configuration scheme has not been used in mass production cars, if the replacement of Magotan uses this set of schemes, it is likely to be the first product of this scheme. However, some models have been equipped with 7V+32TOPS scheme configuration, which can basically be speculated on the ease of use of the perception scheme on the replacement of Magotan.

The camera perception ability is worth affirming.

In the two schemes of 7V+32TOPS and 100TOPS, the hardware of the camera has not changed in quantity and quality, so the most important thing in the forward direction is still two 8 million pixel cameras, and these two cameras play a very key role in the entire intelligent driving system.

According to the test results of the models that have been installed, these two cameras perform well in lane line recognition and object recognition. For example, in the normal driving lane, under the premise of continuous ice cream cones, automatic braking is achieved, and there is no avoidance and detour operation under this working condition, which may be related to the computing power of 32TOPS, if it is increased to 100TOPS, there may be a coherent detour operation.

In addition, operations such as unclear lane markings and misaligned intersection steering are very smooth, and you can also follow the steering of the vehicle in front.

In addition, during the test, there is a complex section of the on-ramp intersection, which needs to change lanes many times to enter the navigation section, which takes one and a half minutes and goes through 7 lane changes; In this route, the advantages of high-definition cameras continue to be transmitted, but it also shows that under the continuous and complex lane change path, the lack of computing power leads to a takeover.

Let's talk about the shortcomings of DJI's solution (7V+32TOPS) in intelligent driving:

The distance and speed of static and dynamic following in the city are relatively conservative, and the distance between the car in front and the car in front is kept at about 2 meters after braking, and the pores during dynamic driving are also larger, which is easy to be incorporated by the side lanes and reduces driving efficiency;

When taking a turn on or off a ramp on a highway, keep the speed at 20-25km/h and stay away from the vehicle in front;

The ability to change lanes in continuous corners is not good, and the diversion line is not measured at one time, and the perception of the algorithm diversion line is insufficient.

So, what can be improved after the computing power is upgraded to 100TOPS?

The increase in computing power will definitely bring about changes, and it will be more obvious, under the premise of perceiving the same hardware, the increase in computing power will make computing, decision-making and execution faster. What may be improved is that the reaction time of following the car will become faster, and the driving pressure diversion line on the ramp may be caused by the lack of computing power and the slow response.

Overall, this set of intelligent 7V+100TOPS driving scheme, it is really irresponsible to say that there is not much difference with the products installed with lidar by the new forces, after all, the gap in hardware is difficult to make up for in software, unless there is a deep learning network like Tesla that can reach a similar level.

Then, after the replacement Magotan uses the DJI system, it will definitely have a very obvious advantage in intelligent driving among models of the same level. If it is pulled out compared with the new forces' models without lidar, it is slightly stronger, and it can support the NOA function of the no-map city, which is better than the intelligent driving function of the entry-level version of most new forces.

DJI's intelligent driving solution is to complete 90% of other people's intelligent driving functions at the lowest cost. It is an exaggeration to say that a 150,000 car is not inferior to a 300,000 car, but in the 15-200,000 track, this will be one of the most cost-effective solutions, and this is DJI's cheapest solution, not the best technical level.