Author / Yu Lei
In 2021, lidar seems to have solved the mass production problem at once, and once only saw the expensive configuration of the Robotaxi, it began to be applied by many quasi-production models, hoping to create a higher level of intelligent driving system. This year is also known as the first year of lidar mass production.
Although most of these models have been released or listed, the delivery time has basically been pushed to 2022, and many new cars under the brands such as Weilai, Xiaopeng and Zhiji are also included.
However, looking closely at the autonomous driving solutions of these models, there are more Nvidia Orin chips that are still in the "futures" stage, and there are new models such as the Feifan R7 that have not yet been listed, whether they can avoid "skipping tickets" and achieve large-scale mass production delivery within the year, at this stage can not be conclusive.
Xiaopeng P5: A compromise on mass production time
Xiaopeng launched the P5 in September last year, but due to the impact of production capacity and millimeter-wave radar shortages, the delivery time of most car owners has been postponed to 2022. Among them, the 550 and 600P models are equipped with lidar, which can be upgraded to the XPILOT 3.5 version that supports urban NGP, but it will not be upgraded until the first half of this year through OTA push.
The perception system of XPILOT 3.5 has a total of 32 sensors, including 2 lidar, 13 high-definition cameras, 5 millimeter-wave radar, and 12 ultrasonic radars. The lidar comes from the Livox HAP developed by DJI's Lanwo Technology, which is arranged in the position of the traditional fog lights on both sides of the bumper.
Cost reduction is the premise that Livox HAP can be commercialized, the traditional rotary lidar is expensive, mainly because the cost of optical modules including scanning systems is too high, and the change of semi-solid or solid-state mechanical radar scanning methods and algorithm optimization can effectively reduce costs.
Livox HAP adopts a double wedge-shaped prism scheme with thick side and thin side, which changes the laser scanning direction by adjusting the speed of the two prisms, and can achieve a larger FOV without the need for multiple module stitching. With fewer transceiver units, higher point cloud density can be achieved.
According to DJI, Livox HAP has reduced the laser transmitter and receiver to 6, but with an integration time of 0.1 seconds, it can reach the equivalent level of 144 lines in the middle ROI area, and the detection distance for low-reflectance objects can reach 150m, the lateral field of view angle is 120°, and the angular resolution is 0.16° ×0.2°.
Judging from the parameters of Livox HAP, and its scanning characteristics, the region of interest is scanned at twice the frequency of other regions, and it can also have reliable detection capabilities in the high-speed driving stage. It should be known that the Valeo 4-line lidar used in the Audi A8 L3 automatic driving system was stipulated to work only below 60km/h due to the limitation of detection capability.
In addition, XPILOT 3.5 is equipped with a high-precision map, satellite navigation system GNSS + inertial navigation IMU. Another Tesla known for its automatic driving, because it mainly relies on visual solutions and electronic maps, it is inevitable to face some driving situations based on feelings, and high-precision maps can help the automatic driving system to perceive complex information on the road surface in advance, such as slope, curvature, heading, etc., combined with intelligent path planning, and make correct decisions in advance.
The computing platform of XPILOT 3.5 is NVIDIA Xavier chip, and it is a monolithic solution. Released in early 2018, Xavier consists of an 8-core ARM 64-bit CPU and a 512-core Volta architecture GPU with 30Tops hashrate, 30W power consumption, and 15 times more energy efficiency than the previous generation architecture.
Compared with the models delivered within this year, the Xavier chip power used by the P5 is indeed not enough to see, and Xiaopeng also replaced the NVIDIA Orin chip in the next model that supports XPILOT 4.0, and confirmed that the current hardware of the P5 cannot be upgraded to XPILOT 4.0.
However, NVIDIA Orin is still a futures, has long planned to list and deliver last year's Xiaopeng P5, in fact, there are not many choices, after all, it is impossible to wait for Orin mass production.
In terms of battery life, Xiaopeng P5 provides 3 battery pack options, NEDC endurance 460-600km: 600P equipped with 71.4kWh ternary lithium battery pack, NEDC cruising range of 600km; 550p, 550E, 550G equipped with 66.2kWh ternary lithium battery pack, NEDC cruising range of 550km; 460E, 460G equipped with 55.9kWh lithium iron phosphate battery pack, NEDC cruising range of 460km.
Xiaopeng G9: The real display of XPILOT
The G9 is positioned as a medium-sized SUV, which is priced between 300,000 and 400,000 yuan, according to Xiaopeng's previous prospectus on the Hong Kong Stock Exchange. The model was released at last year's Guangzhou Auto Show and is expected to be delivered in the third quarter of this year, but the model will still be delivered first with hardware that supports the XPILOT 4.0 system, and the software needs to wait until the first half of 2023 through OTA push.
XPILOT 4.0 is Xiaopeng's next-generation intelligent driving system, with point-to-point all-scenario driving assistance capabilities. In terms of perception hardware, the G9 is equipped with 2 lidars, 12 cameras, 5 millimeter-wave radars, and 12 ultrasonic radars, which is one less camera than the P5.
The G9's lidar layout is similar to that of the P5, but the transmissive semi-solid type provided by DJI has been replaced by the MEMS micro-galvanometer type of the Sagitar Juchuang M1, and the detection distance has been increased from 150m to 200m, and the lateral field of view angle is 120°.
Among Xiaopeng's existing products, the Xiaopeng G9 is the only model that can carry XPILOT 4.0, thanks to its first use of the central supercomputing + area control high fusion X-EEA 3.0 electronic and electrical architecture, Gigabit Ethernet backbone communication architecture, and NVIDIA Orin dual-chip solution computing platform, the computing power of up to 508 TOPS.
However, although NVIDIA CEO Huang Jenxun personally made a decision and announced that the mass production time of Orin chips will be advanced by one year from 2023, there has not been much news about orin's mass production since then. This year, a number of models equipped with the chip, including the Xiaopeng G9, are planned to be mass-produced, and whether large-scale supply can be achieved will determine whether these models can be delivered on schedule.
In addition, Xiaopeng G9 also adopts the 800V high-voltage SiC platform and XPower 3.0 power system for the first time, which can supplement the endurance of up to 200 kilometers in 5 minutes of charging in the future, and the maximum efficiency of the electric drive system can reach more than 95%.
NIO ET7/ET5: Advanced hardware solution
Nio's ET7/ET5 is positioned as a medium-sized sedan and a medium-sized sedan respectively. According to the plan, ET7 will be delivered in March this year and ET5 will be delivered in September this year. Both models are equipped with the exact same NAD autonomous driving technology: the Aquila Supersensing System and the Adam Supercomputing Platform.
The Aquila supersensitive system has a total of 33 sensors, namely 7 8-megapixel high-definition cameras, 4 3-megapixel surround view cameras, 1 ultra-long-range high-precision lidar, 5 millimeter-wave radar, 12 ultrasonic radars, 2 high-precision positioning units, 1 vehicle-road collaborative perception and 1 enhanced main driver perception, which can generate 8GB of image data per second.
Among them, the 8 million pixel camera can detect vehicles more than 680 meters away, pedestrians more than 220 meters away, and cone barrels more than 100 meters away, and the farthest detection distance is 3 times that of the 1.2 million pixel camera; the lidar is developed in cooperation with Innovusion invested by NIO Capital, which is equivalent to 300 lines, with a 120 ° horizontal viewing angle and the highest resolution of 0.06 ° x 0.06 °, and the farthest detection distance can reach 500 meters.
The NIO ET7 is an earlier model released between the two, initially advertising 11 8-megapixel cameras, and then reducing 4 surround view cameras to 3 million pixels, which has also been used by ET5. In addition to 360° panoramic images, surround view cameras are also used for short-range blindness correction for autonomous driving, and high resolution is really unnecessary.
Even so, the hardware specifications of NIO ET7 and ET5 are still the highest in the current mass production scheme, but this route of pushing all cameras to high resolution has also been controversial before. On the one hand, the perceived demand of the backward camera is lower than the forward direction, and the lateral direction is lower than the backward direction, and some insiders say that they can select the type according to the demand. On the other hand, the existing computing power, image transmission bandwidth, the time consuming caused by the algorithm processing of the large map, whether it matches the high-resolution camera of the whole vehicle well, and whether it will bring potential risks.
But according to Weilai's route, at least the hardware that matches the perception is ready. The Adam supercomputing platform is equipped with 4 NVIDIA Orin chips, and the total computing power reaches 1016TOPS.
Among the 4 chips, there are 2 main control chips for daily driving calculations, which can use a computing power of 508 TOPS, 1 is a redundant chip for backup, and 1 is a chip for automatic driving algorithm training.
In addition, the Adam supercomputing platform also supports ultra-high-bandwidth image interfaces, and isps can process 6.4 billion pixels per second, while having an ultra-high-bandwidth data network of 32GB per second.
As for the software, in the case of complete hardware, whether to put it in place in one step or continuously upgrade through multiple iterations, there is a flexible choice.
Zhiji L7: With large-scale mass production as the core goal
Zhiji Automobile is a high-end electric brand jointly built by SAIC Motor, Zhangjiang Hi-Tech and Alibaba Group, and the first model of Zhiji L7 Angel Wheel Edition has a pre-price of 408,800 yuan, positioned as a pure electric medium and large sedan, and will be delivered in the first half of this year according to the plan.
Similar to the previous model, intelligent driving is also the main feature of the Zhiji L7, but its IPD intelligent driving center has two kinds of matched lidar and unmatched lidar. The earlier models were not equipped with lidar, a vision-based perception scheme, and the sensors included 15 high-definition cameras (2 forward-looking, 5 enhanced-sensing, 4 surround-view, and 1 cockpit), 5 millimeter-wave radars, and 12 ultrasonic radars.
The scheme selection also takes more into account the current application scenarios and mass production capabilities, only the forward two cameras resolution of 5 million, and through the 28 °, 120 ° collocation, to achieve some car companies rely on 28 °, 60 °, 120 ° three cameras to do different distance detection tasks.
For 4 side-view cameras and 1 rear-view camera, the 2 million resolution cameras with sufficient coverage to cover the required detection range are selected, and the 2 million resolution cameras with lower computing power and bandwidth requirements have the farthest detection distance of 75 meters and 140 meters, respectively.
Although the early solution of the Zhiji L7 autonomous driving did not have lidar, the five millimeter-wave radars all used the latest fifth-generation product, which is often referred to as 4D millimeter-wave radar. Compared with the previous generation of products, in addition to judging the distance, speed, and direction angle of the object, it is also possible to detect the height of the object.
In addition, 4D millimeter wave radar is also known as imaging radar because it can provide high-quality point clouds. Not only can it effectively analyze the contours, categories, and behaviors of targets, identify tiny targets such as mineral water bottles and tire fragments, and incomplete targets such as obstructed pedestrians or bicycles, but also maintain robustness in all weather and visibility conditions. This also makes it change from a traditional auxiliary perception device to a core perception component.
Moreover, 4D millimeter wave radar focuses on point cloud quality, rather than blindly pursuing the number of point clouds, which is also more conducive to reducing the generation of redundant information and landing on the current computing platform. The Zhiji L7 vision scheme uses 1-2 NVIDIA Xavier chips with a computing power of 30-60TOPS. The solution to increase lidar is replaced by 2-4 NVIDIA Orin chips, with a computing power of 500-1000TOPS.
In terms of power, the Zhiji L7 is equipped with a dual motor system in front and rear, with a comprehensive power of 400 kW, a comprehensive torque of 700 N·m, and an acceleration time of 3.9 seconds per 100 km/h. Among them, the maximum power of the front axle motor is 175kW, and the maximum power of the rear axle motor is 250kW. At the same time, it will also be equipped with a 93kWh ternary lithium battery pack, energy density of 300Wh/kg, and a cruising range of up to 615km.
Sharon Mech Dragon:
4 lidar schemes questioned by "stacking"
Sharon Great Wall Motor's new high-end brand, the first model Mech Dragon released at the Guangzhou Auto Show, the global limited edition model priced at 488,000 yuan, will be delivered this year.
The Captain-Pilot intelligent driving system of The Dragon has reached the highest number of lidars in mass production models, with a total of 38 intelligent sensing components such as 4 lidars, 7 8-megapixel cameras, 4 surround view cameras, 5 millimeter-wave radars, 12 ultrasonic radars, and 1 independent high-precision positioning unit.
Among them, the lidar is a 96-line hybrid solid-state lidar from Huawei, with a single radar viewing angle of up to 120 °*25 °.
Of the four lidars, 3 are arranged at the front end, which is the center and sides of the front bumper, which can achieve forward and side perception, which is also the most common layout scheme. However, the last one, the lidar installed in the center of the exhaust decoration of the tail, caused a relatively large controversy.
Automatic driving for backward perception of the demand is relatively low, relying on a lower resolution camera + millimeter wave radar, basically can meet the demand, and then add 1 lidar not only limited effect, but also will produce a large number of computing power burden. In addition to lidar, the mech dragon has 1 800-pixel camera and 2 angular millimeter wave radars in the rear, and the hardware level is already very high.
In addition, on both sides of the lidar of the tail of the mecha dragon, there are two raised and serious exhaust decorations, which may also cause some obstruction, which is also another part of the reason for the controversy.
In terms of computing, Mech Dragon adopts the 610 dual redundancy scheme of Huawei's MDC intelligent driving computing platform, with an NPU hash rate of 400TOPS and a CPU hash rate of 440K DMIPS.
In terms of power, the mecha dragon adopts the front and rear permanent magnet motor drive scheme, with a maximum power of 405kW, a maximum torque of 750N·m, and an acceleration time of 0-100km/h of 3.7 seconds. The new car is equipped with a 115kWh battery pack using Dayu battery technology, CLTC has a range of 802km, supports 800V/480kW overcharge, and can supplement 401km of battery life in 10 minutes.
Feifan R7: An application test for the overhead sensor layout
The prototype of the Feifan R7 is the ES333 of SAIC R Motors, and after the brand name change, the model was also named after the new Chinese at last year's Guangzhou Auto Show. According to the plan, the Feifan R7 will be listed in the second half of this year and delivered.
The PP-CEM full-stack self-developed high-level intelligent driving solution equipped with the Feifan R7 has a total of 33 perception hardware, including a 1550nm Luminar lidar, 12 ultra-high-definition cameras, 2 ZF 4D mmWave radar, 6 mmwave radar, and 12 ultrasonic radars.
The Feifan R7, Zhiji L7, and NIO ET7/ET5 are all equipped with sensors on the roof. Among them, Feifan R7 is lidar, Zhiji is the camera, and Weilai is the camera and lidar.
These sensors are sensitive to pollution, and traditional front-facing cameras are therefore commonly installed inside the windshield that the wiper can sweep. If these sensors do not have independent and effective cleaning devices, it may affect the actual performance.
According to early accounts, the R7 computing platform will use NVIDIA Orin chips with a computing power of up to 500TOPS -1000+TOPS, which is a 2-4 piece solution.
In terms of power, Feifan R7 will provide single-motor and dual-motor versions, equipped with ternary lithium battery packs with a capacity of 77kWh and 90kWh, and equipped with L0-level thermal runaway management system, which is officially said to be 0 thermal runaway.
BMW iX: The smartest BMW in history
BMW iX has previously completed the listing, a total of xDrive40, xDrive50, M60 three models, the price range of 746,900 to 996,900 yuan. Among them, xDrive 40 will start delivery in January this year, and M60 will deliver in the middle of the year.
The iX, which was called the next generation of pure electric vehicles in the iNEXT concept car stage, is also equipped with an intelligent driving system that is higher than bmw's current production model capabilities.
In terms of perception, BMW iX has also added lidar, including 1 lidar, 6 cameras, 4 surround cameras, 5 millimeter wave radar, and 12 ultrasonic radars.
It is reported that bmw iX lidar is Innoviz's InnovizOne vehicle-grade solid-state lidar, which has an angular resolution of 0.1° x 0.1°, a maximum field of view of 125° x 25°, a scanning frame rate of 5-25FPS, and a detection range of 250 meters.
BMW did not announce the specific computing platform of iX, but according to the products previously displayed, it is speculated that it may be equipped with High PAD (hPAD), which uses 2 Mobileye EyeQ5H chips and a single-chip computing power of 24 TOPS.
It is rumored that the two chips, one is using Mobileye's algorithm, while the other is an open version of BMW's self-developed algorithm, and can be redundant with each other.
PoleStar 3:
Waymo's exclusive global L4 partner
Last June, Waymo, the star self-driving company in the United States, announced that it had become the exclusive L4 partner of Volvo Car Group in the world. Through the cooperation between the two parties, Waymo Driver technology will be implemented on a new pure electric vehicle platform.
The platform was identified as the Volvo SPA2, and the Polestar 3, which is speculated to be its first model, is also close to mass production. According to a document on Polestar International's official website, the SOP of polestar 3 in China is expected to be 45,000 units per year in the 17th week of 2022, that is, in late April.
But so far, The Pole Star 3 has not had much news exposed, and the official has not announced a specific mass production plan, including the eye-catching automatic driving program. Only rumors suggest that polestar 3 will use the same model of Luminar lidar as the Feifan R7, which can emit a 1550 nm beam, a maximum field of view of 120 ° x 25 °, an angular resolution of 0.1 ° x 0.1 °, and a detection distance of 250 meters.
The autopilot chip is rumored to come from NVIDIA, and the comprehensive computing power can reach 254TOPS. According to this parameter, the Orin monolithic scheme is used.
In terms of power, Polestar 3 will use an electric drive system with a maximum power of up to 410 kW, with an acceleration time of less than 4 seconds per 100 km/ h, equipped with a battery pack of 150-200 kWh, and a WLTP range of more than 800 km.
If 2021 is the first year of lidar mass production, then 2022 is the first year of lidar on the car. As one of the first products to explore higher-order automatic driving capabilities, the routes of each company are also very different, there is the pursuit of exaggerated computing power and perception schemes of Weilai, there are also Xiaopeng and Zhiji, who are mainly stable, but it is precisely because of this that the performance of these models after landing is more worth looking forward to.
However, with the approaching of the mass production time, the Orin chip to be used in some models has not been updated the news exposure, and although the Feifan R7 and Polar Star 3 are also within the scope of production this year, they have not yet been listed and released. Exactly how it will be processed by the end of the year remains to be verified.
About Auto Byte
Auto Byte launches the Automotive Technology Vertical Media for The Heart of Machines, focusing on cutting-edge research and technology applications in the fields of autonomous driving, new energy, chips, software, automobile manufacturing and intelligent transportation, and helping professional practitioners and relevant users in the automotive field understand technology development and industry trends through technology to gain insight into products, companies and industries.
Welcome to follow the stars and click Likes and Likes and Are Watching in the bottom right corner.