Ford Motor Company: Data and the future of autonomous vehicles
Autonomous vehicles have captured our attention and with good reason, as self-driving cars promise to change our relationship to automobiles and the love affair with cars.
The technology needed to bring autonomous vehicles to life is complex. For example, the following market landscape, from Vision Systems Intelligence, includes these components:
Autonomous vehicles technology landscape. Image credit: Vision Systems Intelligence
This data ecosystem is highly complex and involves multiple parties in both the private sector and in government. As private companies develop technology and algorithms, they must partner with federal, state, and local governments that control the roads and make the policy decisions that will allow autonomous vehicles on the road.
It will take years to implement this complex environment completely. The Society of Automotive Engineers created a standard that describes progressive levels [.PDF] for vehicle automation:
SAE levels of autonomous vehicles
Given the complexity and importance of the automotive industry and how it is changing, I invited three of the smartest experts in the world together for a discussion on this topic.
The all-star discussion took place as episode 240 of the CXOTalk series of conversations with top innovators:
Watch the video embedded above to see our conversation and read an edited transcript of critical short sections below.
Tell us about data and the automotive ecosystem?
David Bray: [Autonomous vehicles] produce a lot of data. Do you see that data being stored and processed by the car, the road or infrastructure, or somewhere else? What will the ecosystem looking like in three years?
Paul Ballew: Yes, processed by the car; but also could be within the ecosystem proper. Certainly, will be in a entral environment as well. The technology enables us to do things on the data side that allows you to go down the development of autonomous-type vehicles.
It’s prompting us to push forward with edge data analytics and edge data management. So, I think the answer to your question is going to be all of those factors together.
With these massive data analytic challenges, you’re going to have to have a more diverse ecosystem to enable that.
If you’re really going to go beyond autonomy but have smart vehicles that are interfacing with each other, which could have all sorts of other individual and societal benefits, then the ecosystem must go down that path as well. That includes smart infrastructure and related activities.
We’ve been agnostic, humble, as the technology is changing from a data and analytics standpoint. The ecosystem is not going to miraculously hook up every vehicle to some pipe and collect every data element and somehow build data centers that are the size of the state of Texas.
That’s not the strategy because it’s not feasible. But, the good news is that technology is letting us now do other things that offset those concerns.
Do we need to rethink our understanding of automotive data?
Evangelos Simoudis: There’s a need for new frameworks around which to think about data. There is transportation infrastructure data, data from the vehicle, from passengers in the vehicle, from other vehicles, as well as data providers.
It’s a very complex ecosystem. We tend to think about the data vehicles produce, but this goes far beyond that. I’m glad that Paul mentioned that we need to rethink data management. This is not all about cloud-based storage. You need policies on what data to keep in the car, what to push outside of the car.
You have to understand that we are not talking about a single cloud. This is not about Ford’s cloud versus BMW’s cloud or GM’s cloud. Ford wll have a cloud, but Delphi will also have a cloud, and weather.com will have a cloud. So, even for data that’s going outside the vehicle to this type of infrastructure, the data decisions are very complex.
Over the past twenty years, as cars have become more software-dependent, auto makers have become aware of the importance of data. There are already quite a few sensors in vehicles today. But the type of data, complexity of data, volume of data, the big data that we’re talking about — in an environment where we have autonomous and eventually driverless vehicles and on-demand mobility — is stupendous compared to what we deal with today.
And, that requires new thinking.
Paul Ballew: We agree. We brought together an organization to do this, in part to bring that new thinking forward. Certainly, it applies to autonomous vehicles, but also to other things that we have underway such as IoT.
When we talk about the Internet of Things in an industrial setting and plants, you can’t go down a conventional data management approach. Even if you use cloud-based storage for physical data centers, it’s still impractical and makes absolutely no sense in terms of a sustainable model.
We describe it as “modern master data management.” Data management is essential; data structures and related activities. But, it is a unique way of thinking. How you ingest and curate and leverage those data assets to support business objectives.
We’ve got to go beyond the conventional thought processes, not just automotive, but in the way we think about data management: build a central environment, call it a data lake, put some type of identifiers, and all those things with it. The world is moving well beyond that.
What are the issues around vehicles as the data platform?
Evangelos Simoudis: Over the past twenty or so years, the automotive industry has been thinking of vehicles as platforms. In fact, much of the terminology that they have been using, including Ford, has been around platforms.
We’re talking about different type of platform. It’s no longer a platform for electromechanical devices with some computing, but it’s a platform of sensors and actuators with an immense amount of computing power and quite a bit of storage.
Some people have described it as “robots on wheels.” When you think about on-demand mobility and applications such as ride-hailing, you can think of robot taxis and robots on wheels. Long-haul tracking, another projected application for autonomous vehicles, are very robotic.
We have a very different kind of platform than we’ve used to date in vehicles.
Paul Ballew: Any of us that have grown up in the industry, when we think of platforms, it’s the physical architecture of a vehicle. There was a small utility platform. There’s a mid-car platform. And the evolution of the last few years have been talking about Vehicle-as-a-Platform.
When we describe a platform now, it as an interface point, an insight-generating point, or the ability to leverage and connect vehicles.
That word has evolved in a very short period in our industry. Now, when we describe it as a platform, we have to pause a moment and describe what we mean. Never thought I’d have to put an operational definition around “platform” in the auto industry because it was common terminology.
What about other issues, such as data for training AI systems and interoperability?
Evangelos Simoudis: Having data to train the systems that will go into these autonomous platforms is a much bigger deal than we even thought. In fact, today, one of the investment theses in my firm is identifying start-ups that do simulations, because even the companies that are fielding test vehicles can only collect very small amounts of data. They have very small fleets and the amount of data that can be collected physically tends to be a relatively small sample of what we will need to effectively train the artificial intelligence systems that will provide the autonomy to these vehicles.
Today, Waymo and Tesla probably have the most data, but even that is a very small amount compared to what is needed. So, creating larger collections, whether from other contributors that have actual data or creating simulated data to train systems, will be a very big deal on our journey towards a driverless future.
David Bray: We’re looking at the next decade for autonomy to fully mature. But at the same time, there are already advances here and now and things coming along the way.
One of the nuts that we have to crack is solving interoperability amongst the data. As one who has participated in human standards groups, they usually have three- to four-year time horizon to create standards for that. That doesn’t strike me as reasonable in terms of setting data here.
We’re probably going to need some semi-autonomous mechanisms to make sense of data from different devices, different vehicles, and have some interchange among that. Because, if we rely only on the human condition, we’re going to be slowed down by ourselves.
I do think we’re going to see advances over the next decade and I think full-fledged autonomy – probably seven to ten years, with advances along the way.
