In order to deal with different levels of automated driving workloads, NVIDIA will use a variety of products and solutions based on the Ampere architecture built specifically for edge applications. For the fully automated driving of the L5 level of unmanned cabs, NVIDIA will make products with computing power of ?100TOPS? per watt.
The article 丨 AutoR Smart Driving? No1
The NVIDIA Startup Accelerator Program has become an important part of NVIDIA's development in the AI space.
The NVIDIA GTC China Conference was held online on December 15, and at the GTC conference, NVIDIA released a faster AI chip, the world's first "smart distribution city" with JDL Jingdong Logistics, and the world's first end-to-end network solution with 400Gb/s speeds, NVIDIA's Mellanox 400, and the world's first end-to-end network solution, NVIDIA's Mellanox 400, which has been used for over a decade. Mellanox?400G?InfiniBand.
As an important part of this GTC, NVIDIA conducted video presentations during the conference of 12 startups that were selected from more than 100 companies that signed up for the NVIDIA Startup Showcase.
The selected companies cover areas ranging from session AI, smart healthcare/retail, consumer internet/industry applications, deep learning applications/accelerated data science, autonomous machines/IOT/industrial manufacturing, and self-driving cars.
In the field of self-driving cars Treadmill and Macroview are the first to be recognized by NVIDIA as well as the industry, in which NVIDIA will provide support including technology, marketing and financing.
As an autonomous driving startup with a full-stack unmanned transportation solution for mining, Treadmill SmartDrive realized driverless mining trucks operating in multiple formations at the beginning of 2019, and this year realized unmanned night-shift operations with unattended cabs to start large-scale engineering deployment.
In order to be able to quickly implement the actual technology, Treadstone has developed a driverless transportation solution for open-pit mines called "Mining Valley" with the help of NVIDIA?Jetson?TX2i and NVIDIA?Jetson?AGX?Xavier computing platforms.
The Mining Valley consists of the vehicle system RuiChu, the ground system YuJiang, and the cloud control system TianShu, and the system takes the open-pit mine as the main application scenario, solving the perceptual pain points caused by the working environment of the mining area, such as the high dust and the ambiguity of the road boundaries.
In addition, the self-developed in-vehicle hardware computing platform M-box has completed the third major version of the iteration, integrated with the 5G communication module, passed the China Academy of Weights and Measures for the environment, reliability and EMC, including a number of automotive testing certification, is the first batch of integrated 5G?+?C-V2X communication, support for high-performance parallel computation, high security decision-making and control, and passed the vehicle certification.
As a startup company, Treadmill has signed more than 100 million yuan of commercial contracts in 2019, including a mining truck driverless project with Baosteel Group's Baiyun'ebo Iron Ore Mine, which completed its second-stage acceptance in October this year.
The driverless project contract with NDT South Open Pit Coal Mine is the first driverless transportation project for coal mines in China to be openly tendered, and the delivery and acceptance was completed in June this year.
The contract for 200 driverless wide-body vehicles with Zhonghuanxie, the largest coal mine EPC in China, has been delivered with three operational formations, which have realized 24-hour continuous operation in Zhonghuanxie's Erdos Yongshun coal mine.
It is reported that the conversion rate of the POC project to commercial orders is over 80%, with follow-up orders exceeding 300 million yuan.
Another autonomous driving startup, Macroview Intelligent Driving, was founded in 2018, which focuses on the research and development of vehicle-specification-grade autonomous driving system solutions, with its main product being the software-hardware-integrated autonomous driving computing platform (ADCU?-?Autonomous?Driving?Computing?Unit).
This is the only general-purpose platform solution in the Chinese market that supports high-level autonomous driving (L3 and L4), integrating algorithmic software, architectural design, automotive safety, and chips, which can achieve high efficiency and optimization, safety and redundancy, and controllable cost and power consumption, and can meet the needs of large-scale front-end mass production.
During the GTC conference, Macroview Intelligent Driving plans to mass-produce L1\L2 level self-driving passenger cars with OEMs in the fourth quarter of 2021, providing advanced assisted driving functions including auto-parking, AEB (automatic emergency braking), and ACC (adaptive cruise control).
Meanwhile, Macroview Intelligent Driving has adopted NVIDIA?Xavier?GPUs in its L3+ high-level automated assisted driving system, which has landed in high-speed trunk logistics, intelligent public transportation, and travel.
Honking Intelligent Driving is also working with several headline mainstream commercial vehicle manufacturers and tech companies to develop L3-level self-driving trucks to help logistics companies save labor, reduce fuel consumption and improve driving safety. According to the plan, its L3-class trucks can save fuel by 5 percent, cut labor costs by half, and greatly reduce the accident rate.
Greg?Estes, NVIDIA's global vice president of corporate marketing and developer programs, delivered a speech in which he said, "These can be in the field of automated driving driving can be quickly landed, but also thanks to NVIDIA's support in investment, marketing and technology."
First, NVIDIA allows them access to NVIDIA's technical resources, and they can have a personal connection with NVIDIA, which is very important for startups, by communicating with technical people far better than consulting documents.
Second, startups want to be able to scale up, and we know that many startups start with local GPUs and then want to scale to the cloud.
So NVIDIA was able to provide them with programs that greatly increased GPU availability, and NVIDIA offered grant programs and discounted pricing to help them get started and grow to maturity.
Lastly, a very important item is visibility.
NVIDIA helped them with marketing programs, working with the startups to build success stories and relying on NVIDIA's marketing resources to raise awareness of the startups.
Greg?Estes said, "In a lot of cases, part of that is dedicated communication with the VCs, and NVIDIA does the matchmaking between the VCs and the startups, again, to raise the visibility of the startups."
NVIDIA has also made innovations in self-driving technology.
First, in terms of end-to-end data collection for autonomous driving, NVIDIA will collect data from a large number of sensor devices, train the models in a data center on a large DGX?SuperPOD, and generate trained neural network models for deployment into the car.
Before deploying these models into cars, NVIDIA has to simulate and test them through hardware-in-the-loop simulation.
NVIDIA will use the actual AI hardware in the cars running these models to simulate the synthesis of the information seen by these models, including a perception component that can use all the sensors to sense the environment around the car, a planning component, and a prediction component.
These components predict the behavior of other cars, pedestrians, and other traffic participants in the scene in order to plan ahead based on how other traffic participants are likely to behave.
To handle different levels of autonomous driving workloads, NVIDIA will use a variety of products and solutions based on the Ampere architecture that are built specifically for edge applications.
For assisted driving functions, embedded chips based on the Ampere architecture delivering 10 trillion operations per second and consuming as little as 5 watts of power can be used to handle the task.
For L2-level autonomous driving, Orin, which offers 45 watts of power consumption and 200?TOPS per second, is available to handle the workload.
For L5-level driverless cabs, which are fully automated, NVIDIA will make a product with ?100 TOPS?per watt of computing power.
The simple summary is that companies that get the NVIDIA Startup Showcase have the most direct access to the most direct channel to get the capabilities that NVIDIA offers not only limited to AI technology, but is cost-effective.
Greg Estes revealed in an interview, "NVIDIA plans to project has covered nearly 7,000 AI startups around the world, NVIDIA hopes that through the NVIDIA Startup Showcase can be in the product development, prototyping and deployment of the key stages of the development of the startups, each member of the enterprise is able to continue to obtain for its tailored to help equity that provides essential tools for startup development."
Currently, the automatic driving technology is in a long period of rapid development, after the first half of the financing, technology integration, the second half of the automatic driving is bound to enter the technology landing stage, how can startups in the new round of technology outbreak fast breakout, in addition to having a hard own technical strength, more need to be empowered by NVIDIA, such as a mature AI company.
This article comes from the author of the automobile home car family number, does not represent the views of the automobile home position.