NVIDIA DRIVE Platform Enters Commercial Operation

Subscribe To Download This Insight

3Q 2019 | IN-5541

In June 2019, NVIDIA announced a partnership with Volvo to develop fully autonomous industrial and commercial trucks. The development will leverage on the NVIDIA DRIVE platform, which is capable of Simultaneous Localization and Mapping (SLAM) and can plan its route on a High Definition (HD) map using Artificial Intelligence (AI). Early last year, Cohda Wireless ported its Vehicle-to-Everything (V2X) software to the NVIDIA DRIVE platform, enabling wireless 802.11p and 5G connectivity to expand the detection range beyond the car’s on-board sensors. As such, the platform can process sensor data from other cars and infrastructure. The NVIDIA DRIVE platform is a modular and scalable solution that offers the NVIDIA DRIVE software stack; AGX hardware; AutoPilot, which is a Society of Automotive Engineers (SAE) Level 2+ self-driving solution; and Constellation, a data center solution that simulates cameras, radar and Light Detection and Ranging (LiDAR) inputs to test Autonomous Vehicles (AV). The partnership between Volvo and NVIDIA is the first to utilize the whole of NVIDIA DRIVE platform’s End-to-End Solution (E2ES).

Registered users can unlock up to five pieces of premium content each month.

Log in or register to unlock this Insight.

 

NVIDIA Takes the Wheel for Volvo

NEWS


In June 2019, NVIDIA announced a partnership with Volvo to develop fully autonomous industrial and commercial trucks. The development will leverage on the NVIDIA DRIVE platform, which is capable of Simultaneous Localization and Mapping (SLAM) and can plan its route on a High Definition (HD) map using Artificial Intelligence (AI). Early last year, Cohda Wireless ported its Vehicle-to-Everything (V2X) software to the NVIDIA DRIVE platform, enabling wireless 802.11p and 5G connectivity to expand the detection range beyond the car’s on-board sensors. As such, the platform can process sensor data from other cars and infrastructure. The NVIDIA DRIVE platform is a modular and scalable solution that offers the NVIDIA DRIVE software stack; AGX hardware; AutoPilot, which is a Society of Automotive Engineers (SAE) Level 2+ self-driving solution; and Constellation, a data center solution that simulates cameras, radar and Light Detection and Ranging (LiDAR) inputs to test Autonomous Vehicles (AV). The partnership between Volvo and NVIDIA is the first to utilize the whole of NVIDIA DRIVE platform’s End-to-End Solution (E2ES).

This is not the first time Volvo has worked with NVIDIA. In the past, Volvo utilized NVIDIA DRIVE hardware but developed its own autonomous driving software through a joint venture with a Tier 1 supplier, Autoliv, for its passenger vehicles. As for commercial vehicles, Volvo has decided to adopt NVIDIA’s full stack solution and is now moving away from testing and toward operations utilizing all of NVIDIA DRIVE platform’s assets for its commercial fleet.

Volvo will deploy autonomous operations on two fronts: its conventional and electric-powered trucks. Its rollout of autonomous conventional trucks will see seven Volvo FH16 trucks transporting limestone from a Brønnøy Kalk AS mine to a port. Its electric-powered autonomous trucks, called Vera, will move containers from a DFDS logistics center to APM Terminals’ containers in Gothenburg, Sweden.

5G Might Disrupt Edge Model for Autonomous Vehicles

IMPACT


ABI Research’s Commercial Telematics Market Data (MD-COMT-120) predicts there will be more than one million global SAE Level 2 commercial vehicle shipments by 2030, with SAE Level 3 and 4 shipments reaching over 200,000 within the same period. To achieve mission-critical success for these higher SAE levels, processing power needs to migrate to the edge. This would allow for on-the-fly AI inference of sensor data, localization of the vehicle on an HD map, and route planning to occur via multiple trained Deep Neural Networks (DNNs) or Proof-of-Concept (PoC) Recurrent Neural Networks (RNNs). It would also mitigate safety issues that may arise from loss of connectivity or low latency. The successful deployment of the NVIDIA DRIVE platform, whose latest hardware is powered by two Xavier System-on-Chips (SoCs), represents the current need for edge computing for successful autonomous commercial deployment and the growing importance of SoCs.

SoCs are needed to power AI functions for AVs and integrate individual Electronic Control Units (ECUs) of legacy vehicles into a singular platform consisting of Graphics Processing Units (GPUs), Central Processing Units (CPUs), and, in some cases, Digital Signal Processors (DSPs) and Image Signal Processors (ISPs). This would reduce power consumption, which is important for Electric Vehicles (EVs) such as Vera, and simplify Over-the-Air (OTA) software processes, as explained in ABI Research’s Next-Generation Automotive Infotainment (AN-5151) Technology Analysis Report. As Original Equipment Manufacturers (OEMs) and Tier 1 suppliers currently lack the expertise to develop their own hardware, partnering with third-party vendors such as NVIDIA is sound. However, ABI Research finds that OEMs such as Volvo do not have much time to develop their own full stacks, as market leaders such as Waymo, GM Cruise AV, and Uber ATG are already ahead of the competition with their own autonomous driving stacks. As such, according to ABI Research’s Autonomous Vehicle Ecosystem Analysis (AN-4996), new Intellectual Property (IP)-based business models in which OEMs co-develop autonomous driving solutions, like NVIDIA and Volvo have, will become the new norm.

However, this shift toward the edge might be stymied by the advent of 5G and future technology that further increases bandwidth and connectivity reliability and decreases latency for AI inference and training. ABI Research believes processing in the cloud could lead to significant reductions in the amount and cost of hardware required at the edge and increase processing power because more SoCs would be able to be run on a central server. We would see Mobile Network Operators (MNOs) that provide connectivity at the edge start to play a bigger role in AV and thus become critical members of the ecosystem. Acting as the bridge between MNOs and the automotive industry, the momentum behind the 5G Automotive Association (5GAA) is a clear indication of the push toward the integration of cellular connectivity in autonomous vehicles.

Still, New Opportunities Abound

RECOMMENDATIONS


Moving forward, ABI Research believes that hardware providers should provide open platforms that allows OEMs and Tier 1 suppliers to completely customize the type of software they must commit to. OTA software processes should be enabled by default so that autonomous vehicles can get the most updated software, protocols, and HD map reference, thus offering scalability right at the point of sale. SoC makers should also be aware of the energy consumption their chips require to meet the demands of a growing EV market. For example, NVIDIA DRIVE AG Xavier can deliver 30 Tera Operations Per Second (TOPS) of performance with only 30 watts of power.

However, SAE Levels 3 and 4 of autonomous commercial vehicles are forecast to have a penetration rate of less than 5% by 2030 (MD-COMT-120), and there are still many challenges for stakeholders in the AV ecosystem to face before reaching these levels. As such, most operational iterations, such as NVIDIA DRIVE AutoPilot, are at an SAE Level of 2 or the so-called “2+,” and it will take some time before autonomous driving can break out of operating within the constraints of a repeated route, as exemplified by NVIDIA and Volvo’s partnership for logistic delivery trucks.

Beyond this, new business opportunities are arising as hardware providers become more entwined with OEMs in the autonomous vehicle ecosystem, such as the ability for map vendors to provide HD maps as a service. A mission-critical autonomous vehicle would require geo-coded meta data that includes all the road geometry as well as specific information and marking on lanes so that the vehicle would know where to stop even if the roads are covered in snow. This is an important aspect of fully autonomous driving, as car sensors may not be able to pick up faded lanes or inconspicuous markings due to snow or ice. Car sensors can also feed real-time data to map vendors, ensuring that an updated HD map is readily available for autonomous cars to respond to changes in infrastructure and roads. As such, partnership relationships between mapping companies and autonomous system vendors are important. Currently, NVIDIA has partnered with Baidu, HERE, TomTom, and Zenrin, which are creating HD maps for roads. HD map companies would also need to develop maps compatible with the software that autonomous system vendors roll out, ensuring that their mapping solutions support the strategies of autonomous system vendors for the long term. One such map strategy would be to leverage OTA services, which ABI Research believes should be a default offering in autonomous cars. An example of this would be HERE’s cloud-enabled Open Location Platform (OLP), which allows developers to purchase and sell Application Programming Interfaces (APIs) on a marketplace integrated into the platform. This would allow for future business models, such as EV owners purchasing a map layer that shows charging station locations from right within their vehicles.