What are the development challenges in autonomous vehicle technology?
This road was part of my drive from my Christmas vacation back to Stockholm. A beautiful winter road, with lots of snow on both the trees and the road. I even had to wear sunglasses to avoid snow blindness. But just as beautiful as it is in this picture, I am certain that all of the existing autonomous platforms – no matter if we are talking about Tesla, Waymo, Uber ATM, or GM Cruise – would fail to take me safely through this trip. As good as all these systems work on a warm Californian road, it would be difficult for them to navigate a snowy Scandinavian one.
Cost verses price, and the pace of development
To understand the current challenges in autonomous vehicle technology development, we can also look at Volvo. In 2016, they announced that 100 families would have their autonomous cars on the streets of Gothenburg by end 2017. Now in December 2017, Volvo talks about much fewer families and the cars will have a very limited autonomous function. Not only is the cost of the hardware high, but the development of the hardware is happening at rapid speed. Another example is Velodyne, who recently announced that their lidar sensor dropped in price from $8000 to $4000.
The improvements in calculating power and energy consumption are happening so fast that car vendors, who are used to building cars that would last a decade or more, are facing new challenges. When you buy a car most of the functions are already installed with no or limited possibilities to do software and hardware upgrades, and to take autonomous vehicle technology to the next level, on the fly (or drive – if you will) software update functionality is needed.
Big data and need for speed
But even if the promise of commercial availability of autonomous cars has slowed down a little, development is still ongoing. This development builds on big data and machine learning. We know how many miles the current leaders of autonomous development has driven to date thanks to public announcements, but it is still only a fraction of all miles cars drive every day around the world, in all kinds of road conditions. It’s valuable data, nonetheless.
First today, then tomorrow
What we probably need to do now is to slow down the visionary thoughts of the future, and handle today’s challenges. If we are to use machine learning to improve autonomous vehicle technology, we need to collate more data from the vehicles than we do today. In turn, this would require faster, more resilient networks to support that growth. Many cars today connect using 2G, 3G and 4G, but if the data we want to transmit would include video and images, 5G will probably be needed to support that. Don’t get me wrong: 4G is very powerful and can handle most of the expected growth for the next coming years. It’s only when we add functions that require what 4G was not built for that we need 5G.
More and better hardware, more software, more data, more machine learning, more miles driven, more knowledge in driving autonomously on a snowy winter road in Sweden. It’s not an easy task we have before us. And since we did press the brake pedal late 2017, let’s make the best use of it to further investigate what is needed, so I can rest my eyes on the surroundings, rather than wearing sunglasses to see the road side snow.
Check out more of the latest automotive industry trends and insights from Ericsson.