2023 Shanghai Auto Show Autonomous Driving Inventory

At this year's Shanghai Auto Show, Jiuzhang noticed that the most mentioned words were "multi-domain fusion", "emphasis on perception, light on maps", "urban NOA", and "BEV+Transformer".

Here is a brief description of some of the phenomena Jiuzhang observed at the auto show:

1. Everyone is talking about "emphasis on perception and light on maps", but in fact, most of them are still using high-precision maps and light on high-precision maps, and cannot completely eliminate high-precision maps. Currently, only Huawei has launched a solution without high-precision maps.

2. Some manufacturers that previously focused on 1550 lidar, such as Tudatong and Leishen, launched the 905 lidar.

3. Domestic 4D millimeter wave radar gradually enters the testing period.

4. Algorithm companies and chip manufacturers are actively “embracing” large models.

5. Going overseas has become a trend.

6. Suppliers introduce optional/detachable/upgradeable solutions to meet the different needs of customers.

The following is the collected auto show information:

(The information collected is not comprehensive, you are welcome to add additional information in the comment area)

1. Chip and domain control

1. Chip

Company name: Horizon

Technology: BPU intelligent computing architecture

Highlights:

BPU Nash is designed for large-parameter Transformers and large-scale interactive games. At the same time, it is optimized for cutting-edge algorithms to win the best algorithm efficiency; at the same time, it uses AI-assisted design, which can greatly improve the programmability of the architecture.

Compiler: From rule-based methods to Policy Network, and further adding Value Network and Monte-Carlo Tree Search under the reinforcement learning framework to improve the compiler capabilities.

Hardware computing architecture:

Programming paradigms will be provided in the future: DSL, HPL

Promote cooperation ecology: On the one hand, it supports the full-stack self-research of OEMs, and on the other hand, it helps them build a set of tools for collaborative integration through TogethertheROS.Auto, which can enable OEMs and software suppliers to collaborate better.

An IP licensing model like BPU has been finalized by one customer.

Auto show official announced partners: Nezha Automobile, ZF, Flextronics, Didi Autonomous Driving Freight KargoBot, BYD, Continental, Deep Blue Automobile

Company Name: Black Sesame

Products: Wudang series smart car cross-domain computing platform, C1200, Huashan series A1000 chips.

Highlights: On the eve of the auto show, Black Sesame released the first C1200 chip in the Wudang series, 7nm process, CPU core A78AE (150KDMIPS), car-grade GPU core G78AE; supports ASIL-D level Safety Island and National Secret

Security modules for Level 2 and EVITA full. A single chip can meet cross-domain computing scenarios including CMS (electronic rearview mirror) systems, parking and parking integration, vehicle computing, infotainment systems, smart headlights, cabin sensing systems and other cross-domain computing scenarios.

Samples of the C1200 chip are expected to be available in 2023.

Official partner of the auto show: Hechuang Automobile (Hechuang V09H-VIP3.0 adopts a parking integrated domain controller for the first time, which is equipped with the Black Sesame Huashan No. 2® A1000 chip)

Company name: Xinchi Technology

Products: Second-generation central computing architecture SCCA2.0, cockpit chip X9SP, L2+ single-chip mass production solution V9P

Highlights:

SCCA2.0 uses X9 and V9 processors as the open computing core, and uses the intelligent vehicle control unit (Vehicle HPC) composed of G9 processor and E3 MCU as the integrated controller of the chassis domain + power domain. The total CPU computing power reaches 300KDMIPS.

X9SP is a full-scene cockpit processor. Compared with the previous generation X9HP, the CPU performance is increased by 2 times and the GPU performance is increased by 1.6 times. X9SP and X9HP maintain hardware Pin-To-Pin compatibility and software compatibility. They can be smoothly upgraded from X9HP to X9SP in one month, and can achieve rapid mass production of models in about nine months.

V9P is a new generation of automotive processor specially designed for the integrated ADAS domain controller.

l Cooperated with Arm China to customize the high-performance "Zhouyi" X1 NPU, with an overall computing power of up to 20TOPS

l8-core Arm Cortex-A55 CPU, 70K DMIPS

lImagination PowerVR 3D GPU, 200GFLOPS

lCar-grade ISP module, up to 1Gpixel/s image processing capability, supports 8-megapixel camera input

Auto show official announced partners: Desay SV, Neusoft Reach, Unity China

Company name: Ambarella

Product: Ambarella announced the launch of the AI ​​vision system-on-chip (SoC) CV72AQ based on Ambarella's new generation CVflow®3.0 AI architecture.

Highlights: Through the CVflowTM architecture, the AI ​​performance of CV72AQ is 6 times higher than that of the previous generation product CV22AQ under the same power consumption, and it can efficiently run the latest deep learning algorithm based on Transformer neural network.

2. Domain controller

Company name: Neusoft Reichi

Product: Autonomous driving domain controller X-Box4.0

Highlights: The X-Box 4.0 integrated domain controller is based on Horizon Journey 5 and Xinchi Technology X9U series chips, and can support the access of 8M cameras, 4D point cloud millimeter wave radar and lidar, and DSI3 ultrasonic radar.

This product has been designated for mass production of a domestic mainstream model and will be mass-produced and launched in the second half of 2023.

Auto show official announcement partners: ZF, Juefei Technology, Ishi Intelligent, Yixing Technology

Company name: MINIEYE

Products: Full range of intelligent driving domain controller products

Product introduction:

Intelligent pilot assistance iPilot domain control series products:

iPilot 2, iPilot 3, and iPilot 3 plus respectively adopt the Journey 3, Journey 5, and two Journey 5 chip solutions to support travel and parking.

iPilot 3 plus can run up to 256TOPS of computing power, support up to 11 high-definition cameras, 5 millimeter-wave radars, and 3 lidar connections, and can implement functional applications such as urban NOA and autonomous parking AVP.

The three products iPilot 1, iPilot 1 eco, and iPilot 1 plus are based on Texas Instruments chips and support functions such as integrated parking and parking and high-speed NOA.

Highlights: MINIEYE displayed two mass-produced delivery models at the auto show, one equipped with MINIEYE's integrated travel and parking solution based on self-developed domain control, and the other equipped with its self-developed smart cockpit solution. In addition, MINIEYE revealed that among the nearly 100 models that have been designated, more than 10 are export models. The main one that has opened up foreign markets is the iCabin smart cockpit solution.

Auto show official announced partner: Nezha Automobile

Company name: Zhongke Chuangda

Product: Intelligent driving domain control solution RazorDCX Pantanal

Product introduction: Based on Qualcomm's new generation Snapdragon Ride platform, it can support both basic configuration 7V5R12USS and high-end configuration 11V5R12USS access, and reserves multiple CAN/CANFD interfaces, and can provide 8-channel automotive-grade Ethernet interfaces, compliant with ASIL- D standard.

RazorDCX Pantanal

Company name: Hongjing Zhijia

Products: High-end intelligent driving system solution based on BEV’s new generation sensing technology, Shanzhengcheng 3 parking integrated domain controller, Shanzhengcheng 3 smart camera module

Highlights:

The high-end intelligent driving system solution uses a single SoC domain controller, with flexible and configurable computing power. The maximum computing power can reach 500TOPs and 250K DMIPs. It is based on the 7nm process and supports 4D imaging radar deployment.

The software algorithms matched by this solution include high-precision active sensor perception algorithms, BEV Fusion (timing fusion) to achieve more accurate environment perception, and real-time SLAM algorithms combined with lightweight maps to replace RTK and high-precision maps.

Shanzhengcheng 3 Xingbo integrated domain controller supports two solutions from Infineon and Xinchi, and also supports European regulations to facilitate the creation of overseas models.

The Shanzhengcheng 3 smart camera module can support an 8-megapixel high-definition camera with a chip computing power of 5TOPS.

Auto show official announced partner: Magneti Marelli

Company name: Heduo

Products: Autonomous driving domain controller HoloArk, smart front-view camera HoloIFC

Highlights:

HoloArk 1.0~2.0 version solution will cover 18TOPS and 288TOPS level computing power

HoloIFC 1.0, based on the Horizon Journey 2 chip, has a resolution of 2 million pixels and meets the C-NCAP 5-star safety standard.

HoloIFC 2.0, based on the Horizon Journey 3 chip, has a resolution of 8 million pixels and meets the international NCAP test requirements. It will enter overseas markets simultaneously with the development of China's independent brands at the end of 2023.

Auto show official announcement partner: Zebra Smart

2. Perception

1. Lidar

Company name: Tudatong

Products: Image-level ultra-long-range main view radar Falcon platform, new generation Robin platform, new perception software platform OmniVidi

Product introduction:

The main view lidar Falcon platform has a 1550nm laser wavelength, a standard detection distance of 250 meters at 10% reflectivity, and an angular resolution of 0.05° (H) × 0.05° (V).

Medium and short-range wide-angle lidar Lingque-W has a field of view of 120° (H) × 70° (V), a detection range of 70 meters under 10% standard reflectivity, and a resolution of 0.1° (H) × 0.4° (V).

Long-range forward-looking lidar Lingque-E, with a detection range of 180m under 10% standard reflectivity, 120° field of view, resolution 0.1° (H) × 0.2° (V), weighs about 450 grams, and consumes less than 9W, noise less than 20 dB(A).

The OmniVidi software platform consists of two major components: perception middle platform and tool chain. The sensing center outputs the raw data of lidar as sensing results through model algorithms. The tool chain is based on powerful lidar data visualization, data annotation, model training and other tools, forming a closed loop of data and algorithms.