The Future of Autonomous Transportation Requires AI Supercomputing in the Car and in the Cloud

Serkan Arslan, Director of Automotive, NVIDIA EU

Serkan Arslan, Director of Automotive, NVIDIA EU

Developing an AI-powered Vehicle
Autonomy will fundamentally improve the way the world moves — from shared and personal vehicles, to long- and short-distance travel, to delivery and logistics. The most significant benefit of this transportation evolution is safety. Today, 94 percent of crashes are caused by human error. By removing the human driver, autonomous vehicles can prevent or mitigate the majority of traffic crashes. For self-driving cars to deliver this crucial benefit, they require a massive amount of compute performance. The more compute, the more sophisticated the algorithm, the more layers in a deep neural network can be run to enable the car to see, think and act. The computational requirements of fully autonomous driving are enormous—easily up to 100 times higher than the most advanced vehicles in production today. But the next generation of autonomous cars will not only rely on energy-efficient, high-performance hardware, it will also need sophisticated software which needs to be developed, validated and certified within the car company. Autonomous driving is more than just replacing the human driver, it’s about creating an AI driver that is much safer than a human. According to RAND Corporation, to drive even 20 percent better than a human requires 11 billion miles of real-world experience. That translates to more than 500 years of nonstop driving with a fleet of 100 cars — an impossible task. That is why a new IT infrastructure is needed to seamlessly support the development of the car, the development of the software stack for autonomous driving and the simulation environment to virtually test and validate the vehicle.

The Next Generation of Vehicles will be Software Defined

Today’s car companies are experiencing a digital transformation. Cloud-to-car computing and car development by creating a digital representation of the vehicle require a completely new look at today’s IT infrastructure. Software is becoming an integral part of the entire development process, from the virtual design of a car to the industrial design through the mechanical and electrical design, to autonomous driving development. The latter relies heavily on using artificial intelligence to train deep neural networks, and then exhaustive simulation to test and validate the entire hardware and software stack before putting vehicles on the road.

Cloud-to-car computing and car development by creating a digital representation of the vehicle require a completely new look at today’s IT infrastructure

Processes once defined by CDOs now need to be integrated into the IT department and run by the CIO. Creating a stable platform for developers amid an extremely dynamic ecosystem is crucial. Developing an end-to-end production pipeline can help ensure innovation, while not losing sight of the actual vehicle production. As the automotive industry is transitioning from hardware- to software-defined vehicles, the average software and electronics content per vehicle is rapidly increasing. Currently, the vehicle’s software functions are powered by dozens of electronic control units that are distributed throughout the car. Each is specialized — one unit controls windows and one the door locks, for example, and others control power steering and braking. By combining this into one high-performance, energy-efficient compute platform, next-gen vehicle architecture will be more capable and cost effective. Centralizing and unifying compute in the car will make it easier to integrate and update advanced software features as they become available, whether for self- driving or AI-powered user experiences. In addition to in-vehicle supercomputing hardware, massively parallel processing solutions can power the data centers used to solve critical challenges faced in the development of safe AVs. A single test vehicle can generate petabytes of data each year. Capturing, managing, and processing this massive amount of data for not just one car, but a fleet, requires this type of high-performance computing architecture and infrastructure.

Using the Virtual World to Train Cars for the Real World

Simulation presents a powerful solution to what has been an insurmountable obstacle: driving millions of miles and reacting to difficult scenarios. By tapping into the virtual world, car makers can safely and accurately test and validate autonomous driving hardware and software. A simulated test environment is more than just a virtual car on a virtual road. It takes model building as intensive as those for movies, and as detailed and accurate as the blueprints for the city roads and highways the car will eventually drive on. And not only does this world need to look realistic, it must also obey the laws of physics. It needs to be designed to virtually test any potential environment and driving situation, with the ability to ingest world, vehicle and traffic scenario models. A robust simulation solution includes “hardware-in-the-loop” testing with a digital feedback loop, allowing developers to test both autonomous driving software and the hardware platform it runs on. In addition to testing, simulation is also a vital tool for vehicle validation. To be truly safe, vehicles must be able to react correctly, reliably, and safely in every possible driving situation. Experts from TÜV SÜD, a German technical inspection association, estimate that each fully automated driving function comprises 100 million such situations. Given this, scalable verification and test methods are needed to cope with this enormous number of scenarios. Simulation is a method that makes these goals attainable. TÜV SÜD, NVIDIA, and Austrian drive system developer AVL are working together to validate and establish simulation for such a process.

The path forward for both the technology and the business model is far from fixed. But certainly a few things are needed to succeed in the development of the future car: A highly scalable platform that can enable all levels of autonomous driving and a software platform that combines deep learning, sensor fusion, and surround vision to enable a safe driving experience. This unified architecture needs to extend from the data center to the vehicle and provide an end-to-end solution that will conform to national and international safety standards. It can be trained on a GPU-based server in the data center, then fully tested and validated in simulation before seamlessly deployed to run on the AI computer in the vehicle. And finally, safety must be the focus at every step— from designing to testing and, ultimately, deploying a self-driving vehicle on the road.