Autonomous vehicles: driving into a very big cloud

Today, you cannot pick up a newspaper—much less an automotive industry journal—without seeing an article related to autonomous or connected vehicles. What was relegated to the distant future less than a decade ago has exploded in the past five years or so.

Autonomous driving holds many promises for a better mobility future. In 2015, more than 35,000 people died on U.S. roads in 6.2 million auto accidents. That number could be significantly reduced. Those too young or too old to drive could benefit. The cost of on-demand mobility could be reduced when the driver (the costlier element of a taxi, for example) is eliminated. Autonomous fleets could be even more fuel-efficient.

The engineers of Google are largely responsible for introducing autonomous driving into the popular culture. Spun off as Waymo in early December 2016, Google’s team working on self-driving technologies has accumulated about 2 million self-driven miles to date—largely on city streets, and equivalent to 300 years of human driving experience, they report.

Virtually all of the Google engineers trace their roots to Carnegie Mellon University (CMU). Last year, CMU’s engineering center celebrated 30 years of research in self-driving vehicles and put post-steel-era Pittsburgh, the city where SAE International is headquartered, back on the map of the automotive industry. Uber chose Pittsburgh for its autonomous driving operations, and in September launched its own self-driving fleet of Fords and Volvos there, capturing not only the city’s attention but the world’s.

As with any new technology, often it is hard to comprehend all that is “behind the curtain.” When you fly, the air traffic control system that allows the airplane to automatically take off or land in inclement weather, or the avionic systems that are behind the autopilot function, are invisible to you.

Here are some statistics to start putting the autonomous, connected car’s less visible aspects into perspective:

  • A recent Delphi experimental car traveled from San Francisco to New York over the course of 9 days with automated driving in operation for 99% of the journey. It generated about 3 terabytes of data.
  • The Google car creates about 1 gigabyte of data per second as the dozens of sensors take measurements, receiving and transmitting data. For example, the car’s Velodyne 64-beam laser is accurate up to 100 feet, can rotate 360 degrees, and is capable of taking as many as 1.3 million readings per second. Using the average amount of time a driver spends in his car per year (about 600 hours), the Google car would generate about 2+ petabytes of data per year.

Petabyte (equivalent to 1000 terabytes) is only the start of a whole new set of terms we will soon have to comprehend. If the world’s fleet of about 1.2 billion vehicles were all to be autonomous someday, the term to describe the 1000^7+ bytes of data generated gets us into zettabytes, an amount of data that tests our limits of comprehension. For comparison, global internet traffic will pass one zettabyte for the first time this year, after increasing fivefold over the past five years.

Understanding the amount of data involved is just the beginning of the challenge. At issue is how this “Big Data” will be used—for example, in helping secure what is the most promising benefit of autonomous driving: improved safety. The data must be effectively processed and analyzed for the benefits to be realized.

The Commission on Autonomous Vehicle Safety and Testing, of which I was a member, recently took on this challenge. On Jan. 5 it issued a report that recommends best practices for a responsible approach to testing and validating autonomous vehicles for the road. One of the key recommendations is that the auto industry should form a consortium to jointly compile and analyze the data generated by autonomous vehicles. Today, each auto company or systems supplier is conducting its own testing per its own standards, and not sharing the “confidential and proprietary” data being collected.

It is going to be vital that this data is not only assembled and shared for joint learning—in an anonymous manner—but that it is also continuously monitored for predictive analytics. On the one hand, the artificial intelligence (AI) software that powers the autonomous vehicles continues to “learn” and evolve. On the other hand, with no one commanding the vehicle, a failure of a component like an integrated circuit could create safety problems for many vehicles before it is noticed.

In fact, AI-based diagnostic software will be required to monitor the AI-based autonomous vehicle “system.” Safety regulators like the National Highway Traffic Safety Administration will have to develop the skills to monitor and manage an entirely new approach to ensuring the safety of Americans traveling in autonomous vehicles.

And perhaps the most daunting part of this data will be ensuring its security. Already, there have been reports of hacking of cars through everything from tire-pressure sensors to the multimedia networks. SAE’s recently released cybersecurity standard for vehicles (J3016) was designed to address this need, but a whole new set of capabilities will be required for this new age.

SAE will have to embrace this new world and support the development of engineers who can not only develop the autonomous technologies that will drive our vehicles in the future, but also support “Big Data” expertise and the cybersecurity competence.

As I keep saying, what an amazing time to be an SAE engineer.