Three major automatic driving schemes

Automatic driving is regarded by the industry as the biggest challenge to the development of the next generation of vehicle technology, and all of them are currently developing corresponding technologies for automatic driving situations, which include the ability of environmental sensing, core computing, and how cars interact with the environment and determine their own decisions. Figure 丨 Tesla's autopilot and in order to solve the vehicle's computational power demand of complex traffic environment, the choice of chip scheme is very important to the manufacturer. At present, support to L2 computing processing ADAS Scheme although a lot, but L3 above, the choice is really very limited. Limited by the law, in fact, the road can only play the most L3, and drivers must be ready to intervene. In other words, so far, automatic driving for humans, but only as a supplement, rather than a real replacement for human driving, which is the fastest in the next two years to enter the mainstream market automatic driving level. However, although L3 is only one level from L2, the computational needs involved are vastly different. L2 and L3 on the vehicle's automatic control ability, although not far away, but L3 allows automatic driving under limited conditions, and it needs to be able to judge the vast majority of the car's surrounding lanes, and at the same time replace the driver to make the various decisions in progress, and the decision-making process requires extremely large computational power behind it. Figure 丨 Mobileye before L2, because the main focus is on the visual computing part, so all the solutions are moving in this direction, for example, Mobileye, before being acquired by Intel, is mainly the supply of visual computing-oriented programs, and ate most of the relevant program market. But as the auto industry goes to a higher level of automatic driving, Mobileye's lack of strategic reasoning has become the biggest bottleneck limiting its future development, and the demand for Tesla, in fact, is also because of this reason to abandon the mobileye, rather than a few accidents on the decision to break up. Considering that 2020, even some research institutions think that by 2025, due to regulatory and related programme development constraints, L4 even L5 level of complete automatic driving ability to go on the road is still impossible task, so in the next few years, the L3 level of the automatic driving program will become the home including the depot, IC Design companies and programme companies are competing for the target market, and DT June believes that the upcoming CES 2018 is an important occasion to observe the trends of each program company. According to the message, including Intel, Qualcomm, and even NVIDIA, TI, will be announced at CES 2018 the latest layout of its autopilot program, the focus will be on the earliest can be on the road L3 level of automatic driving. Nvidia is fully committed to driving the development of self-driving technology, will likely from the emphasis on sustainable upgradeable design to start NVIDIA join the auto driving market competition against MobileyE creates a lot of pressure, and the direction of its design is different, and it is one of the reasons Tesla later chose NVIDIA to abandon Mobileye. Nvidia can be said to be the industry's first automatic driving "decision" of the company, but the decision to rely on is not the GPU, but the CPU, so from the earliest Drive CX to the latest Xavier platform, NVIDIA used the ARM architecture core, Drive PX2 on the use of 6 core processors, 2 of which are customized by NVIDIA depth Denvor core, 4 are standard cortex-a57. Xavier is using 8 deeply customized ARM cores. With the core of the computing power that has been greatly enhanced by NVIDIA, the driving scheme can be used in a very short time to determine the visual processing data that the GPU part computes and make decisions. Xavier is an automatic driving program, announced by NVIDIA at CES 2017, which not only greatly strengthens the computing power, but also uses a better semiconductor technology to control power consumption below 30W, and has made considerable progress in 250W power consumption of the previous generation of Drive PX2. Although NVIDIA may not be dominant in terms of power performance, Xavier is already the most mature and available L4-level automatic driving program in the market, and most competitors are currently only at the L2 level of the ADAS advanced driving system, so at present the depot wants to develop more L4-level products to quickly To come up with a plan for a road test, NVIDIA can be said to be the only option. Figure 丨 Nvidia's Xavier and its support for the L5 level of the fully autonomous driving program Pegasus platform, although the surface of up to 500W power consumption figures scare a lot of people in the industry, but DT June believes that NVIDIA launched this program is not targeted at the mass production market, but mainly for advanced development, After all, the current world of automatic driving regulations are still in the early stages of development, the industry also believes that before 2025, I am afraid L5 level of automatic driving cars are still difficult to go. Since the main trend of automatic driving in the next few years will be mainly L3, whether the Xavier platform that supports L4 level or not will not be introduced. But in fact, the design of the peripheral sensing element for automatic driving is actually quite mature, the key is the legal liability problem involved in the regulation and decision, and the development of high-resolution map data corresponding to the automatic driving. In other words, with the current hardware technology to design a complete L5 auto driving car is not too much of a problem, the key in the software environment part there is a great upgrade space. In other words, the depot picks upUsing the NVIDIA based Xavier platform, can be introduced in the hardware functions fully meet the L4 level of cars, but according to the regulations of different periods, as well as the maturity of the software environment, and gradually upgrade the software, its automatic driving ability from L3 to L4 or even L5. And as the depot development platform is expected to have very good consistency, as well as long-term platform support capabilities, the general Automotive Semiconductor program support during the period of at least 5 years take-off, long can reach more than 10 years. Although the current NVIDIA programme price is high, but if the programme can be designed from 2018 to the production of L4 level above the automatic driving car, the future depot only need to upgrade on the same platform, complete software environment, high-end, scalable automotive products positioning, Not only can the high cost of the scheme itself be diluted effectively, but also the market layout of the depot and the long-term technology development will be positively helped. This is also the wishful thinking of NVIDIA, in all the program manufacturers are also not able to push the L3 level of automatic driving scheme is now preemptive L5 market, to the automotive industry to pay attention to consistency, stability of the habit of the future halfway to the possibility of conversion platform is relatively low, to ensure that NVIDIA In the future the automatic driving scheme of the market and profit space. Of course, at present NVIDIA's L5 scheme has a large demand for power consumption, it is not impossible to produce a mass production vehicle directly, but the demand for the power management of the vehicle is difficult to solve, but with the development of semiconductor technology and chip design technology, this problem should be solved well. Also, DT-June believes that the current NVIDIA hardware for the autopilot solution is relatively mature, and that the information about the technology program at CES 2018 may not be too much, and the focus should be on other peripheral or collaborative progress. Mobileye will ADAS market advantage and rely on Intel to consolidate the status of automatic driving program Mobileye at L3 level and above the automatic driving program, after all, the relevant program will not be available until 2018, but its ADAS, that is, the L2 level below the products, can be said to firmly attract The eyes of the majority of mainstream manufacturers, the market now has more than 70% of the ADAS program are from the hands of the Mobileye. Figure 丨 Mobileye EyeQ3 and Mobileye is currently in the market to sell the EYEQ3 scheme to observe, although mobileye positioning it as an automatic driving program, and Tesla autopilot system, but in fact, its decision-making part of the performance is weak, this can be from the architectural To see the clue: Using 4 MIPS In 2006 has been launched in the 34K core, the time pulse is only about 500MHz, decision sexCan only reach about 1% of Drive PX2, basically can barely meet the demand of L2 automatic driving, higher level will need new EYEQ4 program to support. But because Mobileye in the past long hard ADAS market, the related visual sensor technology is quite mature, if the use of Mobileye scheme, basically do not worry about automatic driving needs of the visual identification processing technology will be acclimatized to the phenomenon, basically most of the scheme can be and Mobileye In conjunction with, even the upcoming EyeQ4, the time of its re collocation and verification can also be minimized. On the other hand, the next few years pure ADAS cars, that is, L2 level below the automatic driving car will still be the absolute mainstream of the market, L3 car positioning will be due to the relatively high-end and resulting shipments difficult to compare. In the next few years, most L2 cars will continue to use Mobileye EyeQ3, mainly because of its low cost, and has been widely validated by the market, the highest reliability. Figure 丨 Mobileye EyeQ4 Architecture of course, in order to ensure the L3 level of the self-driving car market, Mobileye will also be officially launched in 2018 to support L3 EyeQ4 program, its basic structure and EyeQ3 is quite similar to the CPU core of MIPS with The combination of vector acceleration Unit, but the structure of a newer and better version, the overall computational efficiency is also increased by nearly 10 times times EyeQ4, power consumption only slightly increased by 0.5W. That is, power consumption is still the most emphasized and obvious application advantage of EyeQ4, but even so, overall computing power is still far less than NVIDIA's scheme. Although the design itself has been finalized, the computing power of the chip does not have much room to change, but Intel has a baseband, CPU and FPGA Computing and network connectivity program, in the future may through external solutions to improve overall computing efficiency, that is, EyeQ4 may rely on Intel's technical assistance, By achieving a higher level of autopilot support without significantly increasing the system's power consumption, the attempt to dominate NVIDIA's dominance over L3 level markets has been thwarted. It is unclear, however, how Intel will assist EyeQ4 in expanding the L3-level self-driving car market, which is also believed to be the most critical information for Intel's auto-driving market Strategy layout at CES 2018. Qualcomm aiming vehicle networking technology into the market, but also with NXP Bluebox platform to enter the automatic driving market it is well known that Qualcomm has always been the industry leader in chip networking capabilities, and has been the most active in future networking technologies, despite the recent changes in customer relations and the controversy over patent charges, Lead to the future of the companyis overshadowed by the development of but Qualcomm's own technical advantages are still, and for the automatic driving field, from the acquisition of NXP set of visual recognition technology has a very good accumulation, including Baidu, FAW-Bus factory or driving program companies have adopted NXP technology to design related programs. However, the pure automatic driving scheme is not the current Qualcomm's concern, but in the automotive networking capacity, Qualcomm hopes to be able to speed up the process of V2V (car to Car), V2X (car to all things), regardless of the future of self-driving vehicles, the network is better able to adopt Qualcomm technology. Qualcomm believes that the purely artificial intelligent automatic driving scheme in the application has its limitations, after all, through the sensor to perceive the changes in the surrounding environment, can only be targeted at a small range of circumstances to make decisions, and can not improve the overall traffic network efficiency. Only the car itself can only be half of the automatic driving, there is no way to fundamentally improve the overall transport bottlenecks. In addition, the vehicle's own automatic driving software is written through the logic of the person, so people will make mistakes, automatic driving may also make a similar mistake, and visual processing is still designed by the people of the line, signal recognition object, these are not aimed at machine recognition optimization, In fact, the automatic driving system is a very heavy burden of calculation. Therefore, if we start with the basic construction, directly from the network update and location of the road, signal status, the vehicle's driving system at the same time to consider the information from the cloud collection, as well as the car itself collected near-distance sensor data, so as to allow automatic driving vehicles to carry out the best route selection and longer line driving decision making judgments, Will better optimize the overall driving efficiency, and more than simply rely on the car itself intelligent driving logic more reasonable. Of course, in order to achieve the high throughput ideal automatic driving car networking environment, the infrastructure must reach a reasonably complete level, and the integration of the traffic information is meaningful, and the automatic driving level before the L3 is basically not necessary to use V2X. Since the infrastructure of the network is not available in a short period of time, the automatic driving scheme from NXP does not seem to have any practical effect, so Qualcomm will not have to play in the field of automatic driving. Qualcomm has always been more focused on the long-term layout, although the various driving program manufacturers focus on the introduction of local intelligent driving software and hardware solutions, Qualcomm appears to be a little bit behind, but these plans will need to be networked in the future, and networking options, at present, Intel, MediaTek, exhibition, Qualcomm are optional options, Intel should be paired with its own solution, MediaTek, the exhibition is relatively backward technology, while Huawei, Apple may also develop the relevant programs, but also for their own use, then the most mature networking solution is naturally not. Of course, Qualcomm will also strengthen the ecology of the Bluebox from NXP, and through the Snapdragon platform of artificial intelligence processing capabilities to enhance the performance of decision-making, and emphasize its network as the core, integrated cloud and terminalLarge data computing capacity, the formation of a competitive advantage of the scheme, not only to make the car itself, but to make the whole city of traffic become intelligent, I believe that this is Qualcomm to face the future of automatic driving technology development trend, the heart of the real calculation.