Self-Driving Cars?

Autonomous vehicles face enormous hurdles, but there are many shades of automation before that.


The dream of hopping in your car and lounging on a bench seat that faces sideways or backward while it delivers you to your destination is a really nice idea. It’s akin to upgrading from economy class seat on an airline to your own private jet. But this idea also faces some huge hurdles.

First and foremost is liability. If an autonomous vehicle gets into an accident, who’s responsible? If it’s the automaker—after all, there are no drivers involved—this will be a difficult concept to sell to stockholders. And if it’s the vehicle owner, it may be equally hard to sell to the companies that will insure them. In fact, it likely will require a revamping of state insurance laws just to allow this to happen in the first place.

This leads to hurdle No. 2. While autonomous vehicle technology has been proven to work, it has not been proven to work flawlessly. Reaching that milestone can happen in two ways. One is to basically turn the vehicle into a supercomputer, with enough memory, storage and raw compute power to analyze all possible interactions and interpret unpredictable driving habits. This basically turns the vehicle into a giant artificially intelligent machine. And while it’s true that an autonomous vehicle can react more quickly than a person—much more quickly, in fact—it reacts in very predictable ways to what it has been programmed to do. It’s the old line, ‘To a hammer, everything is a nail.’

The second approach is to sharply limit the number of possible reactions—basically to divide and conquer the various pieces and process them centrally—something that works much better when all the cars on the road are autonomous and communicating with each other. This is the dream of all automotive technology companies, of course, but it’s offset by the reality of people hanging onto cars for a long time. The average age of cars on the road in the United States is 11.4 years, according to IHS. And that’s the average. Some are much older. Even assuming that all cars are autonomous, laws would still have to be re-written to provide special permits for self-driving.

(There is a good discussion of some of these developments in a just-released report by the Insurance Information Institute on Self-Driving Cars and Insurance, which can be found here.)

One thing that becomes clear in this discussion is that automated responses and autonomous are very different. Sensors that set in motion automated responses if the driver nods off or that take control of the car in a skid, or even which allow a car to self-park at the press of a button, are all major advances in automotive safety and automation. Cars will continue to make progress in this area, even using other vehicles to communicate well beyond the line of sight to avoid multi-car pileups or even single-vehicle accidents.

Whether this ultimately removes all drivers from the road, potentially even turning cars into driverless taxis rather than personal vehicles, is something that will take years to play out. But given the number of cars on the road—60 million new cars per year, at least count, times 11.4 years per car—coupled with the amount of advanced electronics going into new cars, and you can see why this discussion generates so much interest. It will continue to be a huge market opportunity for a vast number of chips (multiply the total cars by 8 to 10 SoCs per vehicle and many more sensors). Even the steps leading up to fully autonomous cars are significant, and the opportunity will continue to grow with each new feature, regardless of whether we ultimately get to self-driving vehicles.