In March last year, Baidu and Geely established Jidu Automobile, Baidu, which has rich experience data in the field of automatic driving, has not had a platform that can exert technology, and Geely, which has the latest vast architecture, is the object that Baidu can favor, and their wave of union, like two giants, move forward together, and the Jidu they have born has the potential to monopolize in the field of automatic driving and complete the unification.
On April 18, Jidu officially released the design details of the front part of Jidu's first concept car, as well as its ideas and concepts for the vehicle architecture and automatic driving system. Set degree of the concept of the car front part, two lidar is symmetrically arranged on the front cover, some of the past those old cars similar to the design of the flip light, and the set degree to confirm that these two lidar can be lifted and lowered, and the flip light is really the same magic, but the cost of two semi-solid-state lidar can be really not low, plus other sensing equipment, the concentration to face the instability factors are obviously a lot.
Another new idea outside the mainstream of autonomous driving?
Jidu's announced automatic driving system this time can be described as a new automatic driving idea, our common solution to achieve high-end automatic driving is nothing more than these two: one is the pure visual automatic driving system used by Tesla - FSD, the radical Tesla has cancelled the millimeter wave radar of all Tesla models sold in the United States in the second half of last year, and completely uses the pure visual solution of the camera to complete the automatic driving, toyota has also announced that it will take this approach.
It is the domestic new force Wei Xiaoli and other widely used vision + lidar + high-precision map of the automatic driving system, in addition to the addition of the vehicle-road collaboration system, especially with the appearance of the ideal L9, Wei Xiaoli has fully entered the era of lidar, which is also a widely used automatic driving solution by domestic and foreign car companies.
The system used by Jidu is Tesla + Wei Xiaoli, and it is not only to integrate these two schemes together, but to use the two schemes to exist independently on the car, to achieve mutual independence and mutual backup under the same conditions, and to play a role in improving the safety of automatic driving. This can be regarded as a new solution in the field of automatic driving, it is not like installing a Windows system on an Apple computer and switching at any time, but two sets of systems operating at the same time, handed over to the same processor for data processing, and the complexity has gone up to an order of magnitude.
There is a question here, if the information collected by the dual system is contradictory, then who is listening?
The hardware is all "old actor", how is it different from the streaming scheme?
First of all, in terms of hardware, from the information that has been released, the lidar carried on the Jidu model is the AT128 of Hesai, and friends who often read our articles are certainly not unfamiliar with this name, it is in the ideal L9, Lotus Eletre, and the lidar that will be mounted on the Xiaomi model in the future, there are 128 independent lasers inside, and 128 lasers can be fired at the same time in the vertical direction for scanning. Through the chipized 128-channel solid-state electronic scanning, and the point cloud in the horizontal and vertical direction of the full field of view angle without stitching even distribution, can provide camera-like image-level ultra-high resolution and structured data, 200 meters of telemetry capability, the dot frequency of more than 1.53 million points per second. In particular, the set degree is also arranged for two, which can cover the front horizontal range of 180 ° and the degree of redundancy is higher.
Lifting lidar can ensure the beauty of the overall model, but also to ensure the safety of the lidar's own equipment, after all, after parking the lock car, it can be retracted, without being placed on the roof of the car as the wind and sun. At the same time, for users who pursue a sense of ceremony, this lifting design is very ceremonial, others celebrate the goddess lifting, our lidar lifting, although the car price difference is a lot, but lidar is not cheap.
Secondly, although there is no mention of camera information, but since it is the use of pure visual solutions, and the CEO of Jidu has announced that the new car will be mass-produced in July next year, now at this stage, the number of cameras must be indispensable, and the pixels are expected to be 8 million pixels, and at this stage, 2 million pixels are used, and the domestic new power models absolutely cannot look up, lidar has been used so expensive, and the money for the camera must not be saved.
Next, when it comes to the core component of autonomous driving, the chip, Jidu will use another of our old friends, a single Nvidia Orin chip with a hashrate power of 254TOPS. This is also our article often appear in the old acquaintance, the earliest equipped with Orin chip of Weilai ET7, has been opened on March 28 this year delivery, because ET7 has 33 perception hardware devices, the amount of data that needs to be processed is extremely large, the vehicle NIO Adam platform integrates Nvidia's 4 Orin chips, the computing power exceeds 1000TOPS, it is expected that the number of hardware perception devices of the Jidu model may exceed the WEILAI ET7, especially the dual system has brought a huge amount of data, The number of chips needed may be higher, but if there is a better algorithm platform, the data is categorized, and more chips are not necessarily piled up.
As we said before, NVIDIA has deeply penetrated into the automatic driving system of Chinese car companies, and more than 25 OEMs around the world have reached cooperation with NVIDIA on Orin, including many new car-making forces such as Weilai, Xiaopeng, Ideal, WM, SAIC Zhiji, R Automobile, etc.; BYD, Mercedes-Benz and other traditional OEMs have also joined this camp; In addition, Desay SV, Baidu Apollo, Wenyuan Zhixing, Yuanrong Qixing and other autonomous driving companies will also develop their own ADAS solutions based on Orin.
Stronger than FSD and NGP?
Although we have not seen more use cases for this self-driving system, based on the currently known information, we can already compare it with the current major autonomous driving systems, as well as the advantages and limitations.
All along, everyone criticizes Tesla FSD point, is that it completely relies on vision, relying on its own algorithm, although the cost of this program is very low, which also leads to a lot of limitations, such as light and shade changes, lack of light, no objects in the database, FSD may be invalid, so it is very uneasy, especially for Chinese who are willing to leave everything to the backhand, this scheme can not be trusted by most people. This system of jidu can be said to be able to meet most of the needs for automatic driving, the double redundancy brought by the dual system can go further from perfection, especially this system also comes from Baidu, Baidu has done automatic driving experiments for several years, which is definitely one of the best in China and even in the world, starting early, and the amount of data and technical reserves is very rich. Compared with Tess, this system is more "reliable", and correspondingly, it is more expensive.
The domestic new forces to build cars Wei Xiaoli is based on vision + lidar + high-precision map of the automatic driving program, which is also based on vision, other sensors as an aid, but we also see that they are compared to Tess, perception hardware equipment is rich a lot, which is also the most mainstream way on the market, vision as the dominant, lidar, millimeter wave radar and other equipment at the same time to perceive, can leave more redundancy. The difference between them is the degree of integration, which lies in their perception systems, maps, and vehicle-road coordination, which work simultaneously within a set of systems, although each performs its own duties, but the collection is uniformly handed over to the algorithm platform to weigh this information, so it needs a large computing chip supplemented by calculation. The set of this dual system, it seems to be more complex than a single system, the two sets of automatic driving programs are backed up and complemented by each other, the two sets of programs can be efficient self-driving, but also efficient collaboration, which is intended to show that the two systems are working together all the time, but because the two systems are collecting and processing data separately, the amount of data calculated by the integration of the single system will be reduced.
Just a few days before Jidu's release, Intel's self-driving giant Mobileye demonstrated a full-scenario test of robotaxi' self-driving dual-system Robotaxi equipped with its True Redundancy sensing system in Israel.
The dual-system solution adopted by Mobileye and Jidu can not only greatly reduce the amount of data required to verify the perception system, but also ensure that the vehicle can continue to drive safely in the event of a stand-alone system failure.
Mobileye's environmental sensing solution, which includes two separate autonomous driving subsystems: a self-driving car based on pure cameras and an autonomous vehicle system based on radar and lidar, has basically the same logic and set size scheme. However, the two subsystems back up each other, not complement each other. The two systems are still primary and secondary, the camera subsystem is the pillar of the self-driving car, with the camera as the main sensor, while the radar and lidar subsystems are mainly responsible for emergency avoidance and play a role in safety redundancy.
That is, in the automatic driving to ensure the operation of a system, another set of systems is also running, but does not intervene in the control of the vehicle, in a standby state, when necessary, the two systems are connected in parallel, collaborative control of the vehicle, which can not only reduce the processing of a certain amount of data, but also can immediately intervene when needed to strengthen redundancy.
This may not be easy to understand, but use the example of a vehicle drive system:
1. Tesla's automatic driving is like a single front drive or a single rear drive model, the normal road surface is no problem, and the wet and slippery road surface may lie down.
2. Wei Xiaoli's scheme is like a full-time four-wheel drive, the vehicle is in four-wheel drive at all times, the central differential or clutch plate has been in a working state, assigning power to the front and rear axles, and the fuel consumption is high.
3. Mobileye's new scheme is like a timely four-wheel drive system, such as the new Audi Q5L quadtro ultra four-wheel drive system, when there is no need for four-wheel drive intervention, the vehicle front-drive form, bringing better fuel economy, and the sensor scans the road surface all the time, the four-wheel drive system is always on standby; when the four-wheel drive intervention is required, the electronic control multi-plate clutch is pressed tightly, and the power is distributed to the rear axle to form a four-wheel drive.
4. The dual system of concentration can be compared to the electronic control four-wheel drive system, the front and rear axles are not hard connected, the dual motors control the front and rear axles respectively, maintaining the full-time four-wheel drive, but the front and rear axes are disconnected from each other, and the front and rear motors are controlled by the computer when needed.
To realize the great unification, all families must be vigilant
This dual-system automatic driving program of the set degree, for the various challenges to be faced, has more and stronger adaptability, more redundant space, but there are many problems in the current situation, for the control of supply, for the cost control of this system, are very severe, here will not be expanded, see here, you must know that these equipment are expensive.
Jidu automobile as standing on the shoulders of Baidu Apollo, the giant, and Baidu's relevant senior person in charge has also said that Baidu Apollo's L4/L5 level automatic driving system is to unify the sky, and this display of Jidu has really made me have such worries, integrating the advantages of the current automatic driving, its own early entry, solid technical precipitation, plus large capital behind, it is not difficult to see that Jidu's technology really has the possibility of dominating the world, I believe that when more technical content about the set degree is disclosed, the current group of new forces to build cars may really be a bit panicked.