Great Wall Motors made its NOA debut in urban areas, and focused on its intelligent driving system, SEE integrated intelligent driving model.
The Great Wall official explained that the SEE integrated intelligent driving model realizes the integration of perception and decision-making path integration, which is different from the current mainstream modular intelligent driving, but tends to be similar to Tesla's FSD V12 end-to-end intelligent driving that was previously opened in the United States.
Previously, the Great Wall declared a new Wei brand Blue Mountain car with lidar, and the official introduction is that the Great Wall's intelligent driving also relies on Coffee Pilot Ultra, and the hardware configuration contains 1 lidar. However, in this test, no lidar model was used, but pure end-to-end intelligent driving.
0 takeover in the whole process, is it really easy to use end-to-end?
Briefly talking about the results of the urban NOA test, the test site is in Baoding, Hebei Province, the base camp of Great Wall Motors, the urban road conditions are more complicated, and there are more pedestrians and shuttle vehicles. The length of the test route is 11.5 kilometers, including suburban roads, viaducts, and urban roads, and the whole site has achieved zero takeover.
According to the official data of the Great Wall, the pass rate of no lane or lane line blurred driving in this test was 100%, and 33 intersections were successfully passed, 36 different types of traffic lights were identified, 3 and 6 left and right turns were made, 7 pedestrians were avoided, 12 vehicles were avoided, and obstacles were avoided 4 times.
Judging from the data, the urban NOA test was very successful, but from the actual measurement of nearly an hour, there are some flaws in the middle. When the vehicle just drove out of the Great Wall Park and came to the public road, the first right turn encountered, when two cyclists were identified on the left front side, the vehicle did not brake in advance, but braked urgently at a very close distance from the pedestrians;
In addition, when a taxi occupies the lane in front of it and there is no car in the adjacent lane, the test vehicle does not choose to detour, but waits for the taxi to get off and then follows the way.
Judging by the average level of current intelligent driving, the Great Wall's urban NOA solution is different from most of the previous intelligent driving that purely relies on lidar in that it is indeed more humane.
For example, in the same lane change scenario, some modular smart driving models will not perform lane change when they encounter a solid line, and if they cannot change lanes, they will wait for the opportunity to change lanes before they are realized. When encountering obstacles, they will not look for opportunities to merge and detour, but will wait until it is absolutely safe to do so, and then choose to change lanes.
The biggest feature of end-to-end is the ability to learn from end-to-end, which is also a kind of AI large model, similar to ChatGPT, that is, through uninterrupted learning, the above problems may be optimized in the next intelligent driving, without the need to restrict intelligent driving through code.
Lower cost, end-to-end will become mainstream?
Tesla took the lead in promoting in the United States, Great Wall China completed its debut, and on the afternoon of April 15, Jiyue Automobile, which is also based on pure vision, also held a live broadcast of AI large model intelligent driving. In other words, end-to-end intelligent driving is on the rise.
For Great Wall, Momo Zhixing provides a modular intelligent driving solution, based on the Orin-X high-computing domain control platform, 1 lidar, 3 millimeter-wave radars, 11 high-definition cameras and 12 ultrasonic radars. In March, it was also rumored that Great Wall would cooperate with Yuanrong Qixing, which is the leader in end-to-end model onboarding, and will become China's first end-to-end model on-board artificial intelligence company in 2023.
At the 100-person meeting of electric vehicles in March, intelligent driving became a hot topic, in addition to technological development, cost is also the focus of car companies. We all know that intelligent driving to high-precision mapping and hardware is the foundation of cost reduction. DJI said that 7,000 yuan can solve the urban NOA, and trams and fuel vehicles can be used, and the industry's first batch of AI+ binocular solutions, 4,000 yuan can realize the navigation assistance function. The goal of Yuanrong Qixing is also to create intelligent driving technology that everyone can afford.
In terms of modular intelligent driving, Yuanrong Qixing has lowered the cost of the entire intelligent driving solution to $2,000, and the hardware cost of LiDAR + camera has dropped to $7,000. What's even more economical is the end-to-end that even lidar is saved.
It is not difficult to see that the Great Wall has a modular intelligent driving solution under the premise of adopting an end-to-end intelligent driving solution, it is certain that the technical optimization of modular intelligent driving is relatively more difficult, and the Great Wall is not in a leading position in the industry at present. Through end-to-end, cost-saving, and good results, and at the same time, the world's strongest end-to-end Tesla FSD has not yet entered China, and Great Wall has the opportunity to catch up with the gap from this road.
The next intelligent driving is cost, so there are a number of end-to-end intelligent driving system car companies based on pure vision, which have gradually attracted attention. Of course, it takes time to achieve it in one fell swoop, and it takes time to achieve it overnight in the face of China's complex road conditions.
Even so, there is a tendency for end-to-end to become mainstream. It is not difficult to see from the structure of new energy products that the gap between driving prices is one is the battery and the other is intelligent driving. At present, there is not much difference in the cost of liquid batteries, depending on the choice of ternary lithium and lithium iron phosphate, as well as the choice of battery capacity. Intelligent driving has become the only core point that can open up the value of new energy vehicles under the same conditions.
For example, the second-listed Hongmeng Zhixing Zhijie S7, its lowest configuration model is equipped with an intelligent driving system with a pure vision solution, while the higher-priced model uses a 192-line single lidar solution. It can be predicted that if the two solutions are mixed, the price of the whole vehicle may be higher, but fortunately, the cost of different threads of lidar is quite different, for example, the price of about 270,000 AVATR 12 uses 3 92-thread lidar, and the Zhijie S7 with a similar price only uses 1, and the actual effect is not much different. However, if AVATR 12 only uses 1 92-thread lidar, although the cost will be greatly reduced, the intelligent driving effect will lag behind the Zhijie S7.
The number and performance of lidar directly affect the cost of intelligent driving system, and the biggest advantage of end-to-end is reflected in comparison. This is also the reason why the cost of Huawei's ADS 2.0 high-end intelligent driving is high, and why Tesla was able to reduce the price for a while. Moreover, the maintenance cost of modular intelligent driving relying on lidar is also high, because it needs to constantly improve intelligent driving rules and tests, and end-to-end, only big data feeding is required.
Previously, Tesla announced that it would launch robotaxi in August, and the important role is that the big data it collects can effectively train FSD end-to-end intelligent driving, so as to lay the foundation for the use of civilian models. At present, Tesla's intelligent driving mileage has reached 1 billion kilometers, but it still fails to feed a relatively safe and intelligent end-to-end intelligent driving, so how many kilometers can feed a reliable end-to-end, it is not known at present, it can only be said that including the Great Wall, Jiyue and other end-to-end intelligent driving, the cost is saved, but there are still difficulties in data.
Epilogue:
Will there be a more powerful fusion scheme in the future, such as end-to-end advanced + lidar guarantee, there is a possibility, but it seems that there is little need at this stage, after all, whether the two sets of solutions can be integrated and adapted, whether the cost can be enough, and what the actual effect will be, still needs to be verified, but irreversibly, intelligent driving is also like the battery industry, must be achieved through various ways to reduce costs, and the era of intelligent driving to raise the price of smart cars is almost over.