The idea of wirelessly connected, fully autonomous, electric vehicles is certainly not new. It has existed for decades. But what was once a collection of unrelated research projects has gained a level of focus and intensity that was unheard of less than 10 years ago. The speed at which these technologies are advancing, combined with new mobility-service business models, is creating a revolution in the automotive industry that is both incredibly exciting and very challenging.
While there are numerous challenges in the race to bring shared, autonomous, connected, electric vehicles to market, many of these challenges are, strictly speaking, non-technical. For example, how do we create a regulatory environment that fosters the development and usage of autonomous vehicles? How do we increase consumer acceptance and demand for electric vehicles? How can automakers cover the costs of adding vehicle-to-vehicle communications devices to their vehicles when the benefits of this technology will be negligible until many years after the initial deployments?
While these types of questions are indeed challenging and important to answer, the focus of this article will be on some of the engineering challenges related to connected and autonomous vehicles, namely the need for speed.
Computing speed—Moore is not enough
Over the last 40 years, computing power has increased by a factor of approximately one million. This phenomenon was foreseen by Gordon Moore in 1965, when he predicted that the number of transistors in an integrated circuit would double every two years.
While the industrial-computer and consumer-electronics industries have taken full advantage of these technological advances, the automotive industry has been much more conservative in its approach to computing power. In the consumer-electronics industry, computing speed is a key differentiator, and older products are quickly overshadowed by newer products with the next generation of computing power. In contrast, the automotive industry is much more focused on safety, reliability, and repeatability, particularly when it comes to the electronic control units (ECUs) that oversee driving functionality (e.g., for braking, engine, and steering).
The trend towards autonomous driving, however, has revolutionized the automotive industry. Safety, reliability, and repeatability are still of the highest priority, but computing power is now just as important. Automakers around the world are adding the newest and most powerful microprocessors to the ECUs that provide the intelligence behind their autonomous-driving functionality.
The AutoPilot computer in the Tesla Model 3, for example, includes an Intel Atom microprocessor and three Nvidia high-performance graphical processing units, in addition to an automotive-grade Infineon Tri-Core 32-bit microcontroller. The processing power of this AutoPilot computer alone exceeds the combined processing power of all the other vehicle control ECUs on the Model 3, to the extent that it requires liquid cooling to keep it from overheating.
And this is only the beginning. Vehicle computers such as this merely provide driving-assistance functionality, not a complete autonomous-driving capability. Consequently, the automotive demand for computing speed will only continue to grow as vehicle driving tasks become increasingly automated.
Networking speed—lessons from the IT industry
Computing speed is only one of the factors in the autonomous-driving equation. To make fast decisions, autonomous-driving computers must have rapid and reliable streams of data to process. These data streams can come from many different sources, including cameras, radar sensors, LiDAR sensors, vehicle-to-everything (V2X) communications radios, the cloud, and multi-dimensional maps. Regardless of their source, these data streams have one thing in common—the resolution of their data is increasing, which necessarily implies that their data rates must increase.
Until recently, data rates for all vehicle-control systems and most automotive sensors were less than one megabit per second (Mbps). These data were transmitted via networks that were specifically developed for automotive applications such as the Controller Area Network (CAN) bus system. Data that was faster than 1 Mbps were either streamed directly from a camera to a display or it wirelessly streamed from the cloud to a smart phone (independent of the vehicle network and electronics).
However, within the last few years, camera and 4G cloud data have been increasingly used as inputs to autonomous-driving computers. In addition, LiDAR sensors, V2X communications radios, and 5G cloud data are expected to be used in the very near future. As shown in Figure 2, cameras are already operating at up to one gigabit per second (Gbps) and are expected to exceed 10 Gbps within the next five years. Furthermore, the total number of autonomous-driving sensors on each vehicle is increasing as automakers seek to provide 360 degrees of surround-sensing capability with overlapping sensor fields of view.
To quickly transfer these data from the sensors to the autonomous-driving computer, the automotive industry has taken a lesson from the IT (information technology) industry, namely the use of point-to-point communications and switched networks. These two IT networking constructs, which form the physical basis for an automotive Ethernet network, can be used to provide the necessary speed, redundancy, signal integrity, prioritization, scalability, and security for autonomous-driving functionality.
Development speed—redefining the automotive industry
Perhaps the greatest challenge related to connected autonomous vehicles is the need for increased development speed. This challenge manifests itself in two ways.
First, the timeline for automotive development has drastically contracted. Ten years ago, a typical vehicle development required about four years. The focus was on flawless execution, while speed and agility were given a bit of a short shrift. Today, the development process can take as few as two to three years. Part of this reduction is attributable to the industry’s need to reduce development costs and the transition to the development of fewer and more global vehicle platforms. In addition, the influence of the consumer electronics industry’s 18-month development period cannot be discounted.
Secondly, vehicles have become incredibly complex. Long gone are the days when a team of mechanical engineers and styling artists could design a vehicle. Today’s vehicles are a complicated collage of sensors, microelectronics, software, electronically actuated mechanical components, high-voltage electrical components, and cloud-connectivity services that attempt to create a user experience akin to that of a smart phone. Very few automotive companies have the technical breadth to develop everything that is needed. And the few companies that do have this broad technical expertise have realized that it’s more cost-effective to focus their efforts on the technical capabilities that will provide differentiation from their competitors.
These two points, taken together, are dramatically changing the automotive industry, moving it from its former siloed product-development process to a new process in which partnerships and collaborations are key. While the automotive industry has been rife with high-profile acquisitions of late, these are merely occurring in areas where automotive companies have seen opportunities to differentiate themselves from their competitors. The vast majority of automotive technological innovation has come about through partnerships and collaborations.
This is the foundation for innovation in today’s automotive industry. Those who adapt to this new collaborative approach will thrive. Those who don’t will quickly become irrelevant.