Autonomous vehicles are born in the data center, which is why NVIDIA and Deloitte provide a strong foundation for developers to implement robust self-driving technology.
At CES this week, the companies detailed their collaboration, which aims to alleviate the biggest pain points in AV development. Deloitte, a leading global consulting firm, is partnering with NVIDIA to offer a range of data generation, collection, intake, curation, labeling and deep neural network (DNN) training services with the NVIDIA DGX SuperPOD.
Building AVs requires huge amounts of data. A fleet of 50 test vehicles running six hours a day generates 1.6 petabytes daily – if all this data were stored on standard 1 GB flash drives, they would cover more than 100 football pitches. Still, it is not enough.
On top of this data collected, AV training and validation requires data from rare and dangerous scenarios that the vehicle may encounter but which may be difficult to encounter in standard data collection. This is where simulated data comes in.
NVIDIA DGX systems and advanced training tools enable streamlined, large-scale DNN training and optimization. Using the power of GPUs and AI, developers can seamlessly collect and curate data to extensively train DNNs for autonomous vehicle perception, planning, driving, and more.
Developers can also train and test these DNNs in simulation with NVIDIA DRIVE Sim, a physically accurate, cloud-based simulation platform. It leverages NVIDIA’s core technologies – including NVIDIA RTX, Omniverse and AI – to deliver a wide range of real-time AV development and validation scenarios.
DRIVE Sim can generate high-fidelity synthetic data with truth using the NVIDIA Omniverse Replicator to train vehicle perception systems or test decision-making processes.
It can also be connected to the AV stack in software-in-the-loop or hardware-in-the-loop configurations to validate the complete integrated system.
“The robust AI infrastructure provided by NVIDIA DGX SuperPOD paves the way for our customers to develop transformative autonomous driving solutions for safer and more efficient transportation,” said Ashok Divakaran, Connected and Autonomous Vehicle Lead at Deloitte.
A growing partnership
Deloitte is at the forefront of AI innovation, services and research, including AV development.
In March, it announced the launch of the Deloitte Center for AI Computing, a first of its kind center designed to accelerate the development of innovative AI solutions for its customers.
The center is built on NVIDIA DGX A100 systems to bring together the supercomputing architecture and expertise that Deloitte customers require when becoming AI-powered organizations.
This collaboration now extends to AV development by using robust AI infrastructure to develop solutions for truly intelligent transport.
NVIDIA DGX POD is the foundation that delivers an AI computing infrastructure based on a scalable, tested reference architecture with up to eight DGX A100 systems, NVIDIA networks and high-performance storage.
To further scale AV development and speed up time for results, customers can choose the NVIDIA DGX SuperPOD, which includes 20 or more DGX systems plus networking and storage.
With Deloitte’s many years of work in the automotive industry and investment in AI innovation, combined with the unprecedented calculation of NVIDIA DGX systems, developers will have access to the best AV training solutions for truly revolutionary products.
Deloitte’s leadership in artificial intelligence is paired with a broad and deep set of technical capabilities and services. Among its ranks are more than 5,500 system integration developers, 2,000 data researchers and 4,500 cybersecurity practitioners. In 2020, Deloitte was named the Global Leader of Global System Integration Services by International Data Corporation.
Deloitte also has extensive experience in the automotive industry, serving three-quarters of Fortune 1000 car companies.
With combined experience and cutting-edge technology, NVIDIA and Deloitte offer robust data center solutions to AV developers that include infrastructure, data management, machine learning operations and synthetic data generation.
These services begin with Infrastructure-as-a-service, which provides control of the DGX SuperPOD infrastructure in an on-prem or co-location environment. Experts design and set up this AI infrastructure as well as deliver ongoing infrastructure operations for streamlined and efficient AV development.
With Data-Management-as-a-Service, developers can use tools for data capture and curing, enabling scaling and automation for DNN training.
NVIDIA and Deloitte can also improve data researcher productivity by up to 30 percent MLOps-as-a-Service. This turnkey solution implements and supports enterprise-quality MLOps software to train DNNs and accelerate accuracy.
Finally, NVIDIA and Deloitte make it possible to cure specific scenarios for comprehensive DNN training Synthetic data generation-as-a-service. Developers can take advantage of simulation expertise to generate high-fidelity training data to cover the rare and dangerous situations that AVs need to be able to handle safely.
Equipped with these invaluable tools, AV developers now have the ability to ease some of the biggest bottlenecks in DNN training to deliver safer and more efficient transportation.