With the recent launch of Tucson Future, the development of self-driving trucks has once again attracted attention.
1. Some recent progress in autonomous trucks
Volvo Trucks Self-Driving Vehicle Products
Sweden’s Volvo Group has been working on self-driving trucks for many years, with results including platooning trucks, self-driving garbage trucks, a commercial demonstration operation in 2018, and a self-driving dump truck for short-dump transport of lime mines, and cabless self-driving in 2019. The concept card Vera has been put into trial operation at the port.
Self-driving Peterbilt 579 test truck with Aurora Driver
The American company Paccar has strategically cooperated with the American autonomous driving start-up Aurora since January 2021 to establish a partnership for the development and marketization of autonomous trucks.
The Swedish company Einride has been developing the Einride pod, an electric truck that can be controlled remotely, including the AET1, which operates in enclosed areas, and the AET2, which can be driven on public roads at a closer distance, with deliveries scheduled to begin in mid-2021; AET3 for autonomous driving, AET4 that can operate on major roads and highways, etc. are scheduled to be delivered in 2022-2023. Among them, the AET4 is an L4-class model that can be driven on a wide range of roads.
TuSimple is advancing the development of self-driving Class 8 heavy-duty trucks, which have recently come to market. Tucson has previously announced that it will partner with Navistar in North America to launch the industry’s first Level 8 autonomous truck in the North American market in 2024.
Zhijia Technology has been advancing the development of its automatic driving system PlusDrive for rear-mounted trucks, and plans to cooperate with FAW Jiefang to develop and mass-produce self-driving trucks in 2021.
The American company Locomation mainly develops the Autonomous Relay Convoys, which is a platooning technology of two trucks. The rear car automatically follows, and the driver can rest. American transport company Wilson Logistics has announced that it will introduce Autonomous Relay Convoys to 1,000 trucks in 2022.
2. NVIDIA’s self-driving truck circle of friends
The above-mentioned self-driving truck-related companies all have a hidden attribute, that is, the self-driving development of each company adopts the solutions of the American semiconductor company NVIDIA.
NVIDIA Self-Driving Truck Partners
Volvo Group established a partnership with NVIDIA in 2019 to jointly develop self-driving commercial vehicles and decision-making systems. The two parties will carry out end-to-end AI development, simulation and in-vehicle computing projects, which is NVIDIA’s deepest and most extensive development cooperation project. Initially, the development team will rely on the Drive AGX Pegasus platform, which will handle the computing power inside the vehicle, while the Drive AV software stack will handle sensor data and route planning. The team will also use Nvidia’s Drive Constellation simulation platform for testing.
Paccar announced in 2017 that it is partnering with NVIDIA to develop solutions for autonomous vehicles.
Isuzu’s technology development also uses the NVIDIA DRIVE AGX platform. In 2018, Isuzu Motors announced that it would work with Nvidia to develop autonomous driving technology, especially to help autonomous driving systems learn how to apply data collected from passenger cars to trucks and other large commercial vehicles.
Einride announced in December 2020 that it will be equipped with the latest NVIDIA DRIVE AGX Orin on the next-generation POD, through the most advanced processor to achieve the redundancy of the autonomous driving system.
Einride’s next-gen POD
TuSimple’s financing includes a contribution from NVIDIA, and its fleet of self-driving trucks also uses NVIDIA DRIVE.
Zhijia Technology announced in March 2021 that its next-generation autonomous driving system for large trucks will use NVIDIA DRIVE Orin.
Locomation also uses NVIDIA DRIVE Orin.
3. NVIDIA autonomous driving related products
1. NVIDIA’s self-driving car chips include:
2015 – Drive PX, computing power (single-precision floating-point computing power) 2Tops, supporting L2-level autonomous driving
2016 – Drive PX2, computing power 8Tops, supporting L3 autonomous driving
2017 – Drive Xavier, computing power 30Tops, supporting L4 autonomous driving
2019 – DRIVE AGX Orin, computing power 254Tops, supports L5 autonomous driving and is compatible with L2~L4 autonomous driving,
2021 – DRIVE Atlan, 1000 TOPS, for Level 4 and 5 autonomous vehicles – and the next generation of highly assisted vehicles.
At the recent GTC 2021 conference, Nvidia unveiled its latest smart car and autonomous vehicle chipset, the DRIVE Atlan, which promises to be an “AI data center on wheels.” DRIVE Atlan aims to supply Level 4 and 5 Self-driving cars – and the next generation of highly assisted cars, are expected to begin sampling to automakers and AV developers as early as 2023.
DRIVE Atlan is the industry’s first “1000 TOPS SoC”. In contrast, NVIDIA DRIVE Orin – a chip solution announced at the end of 2019 but not yet shipped, has a computing power of only 254 TOPS; the Xavier solution in 2020 is only 30 TOPS.
Atlan SoC has the next generation GPU architecture after Ampere, new Arm CPU core, new generation BlueFieldDPU (Data Processing Unit), ASIL-D safety island, etc. At the same time, it also supports 400Gbs (400,000 Gigabit) network and security gateway, which can meet the needs of high-speed communication. The computing power of Atlan’s single chip is 4 times that of DRIVE ORIN (254TOPS, which was announced in 2019 and has not been officially installed in mass-produced vehicles), which is stronger than that of most L4-level autonomous vehicles.
The NVIDIA DRIVE Orin, which will be put into production next year, is not limited to computing autonomous driving, and will become the central computer of the car in the future. The Orin SoC can undertake the computing of the four domains of in-vehicle instrumentation, infotainment, driver monitoring, and autonomous driving. The four domains are virtualized and isolated from each other, and are designed with an architecture that supports functional safety and information security. Atlan is compatible with software stacks written on previous generation chipsets such as Orin or Xavier.
DRIVE Atlan will begin sampling in 2023, and the SoC is expected to appear in 2025 models.
2. NVIDIA DRIVE software:
The NVIDIA DRIVE™ platform includes an on-board computer (DRIVE AGX) and a complete reference architecture (DRIVE Hyperion™), as well as a data center hosted simulation platform (DRIVE Constellation™) and a deep neural network (DNN) training platform (DGX™). The NVIDIA DRIVE™ platform also includes a rich software developer kit (SDK) designed to accelerate autonomous vehicle (AV) development.
The NVIDIA DRIVE™ software stack is open source and helps developers efficiently build and deploy a variety of advanced AV applications, including perception, localization and mapping, planning and control, driver monitoring, and natural language processing.
DRIVE OS is the foundation of the DRIVE software stack, the first secure operating system for accelerated computing. It includes NvMedia for sensor input processing, the NVIDIA CUDA™ library for efficient parallel computing, NVIDIA TensorRT™ for real-time AI inference, and other developer tools and modules that provide access to the hardware engine.
The NVIDIA DriveWorks™ SDK provides middleware functionality critical to autonomous vehicle development on top of DRIVE OS. These capabilities include a Sensor Abstraction Layer (SAL) with sensor plugins, data loggers, vehicle I/O support, and a deep neural network (DNN) framework. Modular and open, the tool is designed to comply with automotive industry software standards.
The DRIVE AV software stack includes perception, mapping, and planning layers, as well as a variety of deep neural networks (DNNs) trained on high-quality real-world driving data. These rich perceptual outputs can be used for autonomous driving and mapping functions. Within the planning and control layer, the NVIDIA Safety Force Field™ computing Module checks the operation of the main planning system to protect the vehicle from collisions.
DRIVE IX is an open software platform that provides in-cabin perception for innovative solutions for AI cockpits. DRIVE IX provides perception applications to access functions, as well as DNNs for advanced driver and occupant monitoring functions, AR/VR visualization, and natural language interaction between vehicle and passengers.
Eighth generation Hyperion platform
The DRIVE Hyperion 8 platform automotive reference architecture contains the sensors, high-performance computing and software needed to develop autonomous vehicles. All of these tools are validated, calibrated and synchronized and work out of the box. Hyperion is a fully operational and production-ready open source autonomous vehicle platform. It can reduce a lot of the time and cost of equipping cars with AI capabilities and the technology needed for autonomous driving.
NVIDIA DRIVE Hyperion Developer Kit
Hyperion is equipped with all the hardware needed to validate autonomous driving systems at peak performance. Includes 2 NVIDIA DRIVE Orin system-on-chips (SoCs) that process data from 12 exterior cameras, 3 interior cameras, 9 radars, and 2 lidar sensors in real-time to enable safe autonomous operation of the car.
The entire suite of tools is precisely synchronized and calibrated for 3D data acquisition, saving developers the time it takes to configure and run a test drive of an autonomous vehicle.
Hyperion provides everything needed for on-road validation of smart car hardware by adding a complete sensor configuration on top of centralized computing. And because Hyperion is compatible with the NVIDIA DRIVE AV and DRIVE IX software stacks, this makes it a critical platform for evaluating and validating autonomous driving software.
Developers can start using this latest open source platform in the near future. The eighth-generation Hyperion will be available to the NVIDIA DRIVE ecosystem later in 2021.
4. NVIDIA’s Advantages and Influences
In the process of NVIDIA’s contact with major car companies, there are five key indicators that affect the choice of autonomous driving chips.
1. The computing power and efficiency of deep learning.
2. Support multiple different types of sensor input.
3. The convenience of software development. Nvidia offers a variety of development tools, such as software development using CUDA, and even GeForce and Tesla GPU hardware. When the product is officially mass-produced, it can be easily ported to autonomous driving hardware.
4. Obtain functional safety certification. At present, NVIDIA’s mass-produced Xavier autonomous driving chip has obtained the ASIL-C functional safety certification of Germany’s T?V.
5. Provide a complete solution. Some car companies develop their own autonomous driving systems, hoping to get more open software and hardware solutions; some car companies hope to buy a complete set of technologies to integrate into the car, both solutions NVIDIA can provide.
More than 370 companies or researchers participated in NVIDIA’s open platform for autonomous driving development, and many development companies integrated their own technologies with the DRIVEAGX system. Extensive collaboration with other companies’ products is easily accessible through NVIDIA.
In addition to high-performance hardware and highly scalable software, such an open platform also increases the value of NVIDIA. NVIDIA, which is behind the car company, has made great contributions to promoting practical development.
The Links: LQ121S1LG42 SKKT323/16E