A bunch of companies in Silicon Valley, Detroit, and around the world are racing to build cars that drive themselves. And they have a big, important disagreement about what these vehicles will look like.
The disagreement centers around the steering wheel — and specifically whether cars will have one or not. On one side of the debate are conventional car companies who largely view self-driving capabilities as a new feature to add to their existing fleet of self-driving cars. On the other side of the debate is Google, which has already built a self-driving car prototype that's designed to operate in a fully autonomous mode.
This is partly a debate about safety; each side claims its own approach will save more lives. But it's also a clash of business models. If human beings can take over in situations the computer can't handle, it will be easier for car companies to gradually introduce self-driving capabilities to their existing vehicle fleets. On the other hand, companies like Google and Uber are starting from scratch anyway, and they can more easily adopt on-demand business models that are more compatible with fully autonomous operation.
The debate matters because regulators are currently drafting safety standards for autonomous vehicles. Reuters reports that California's standards may require cars to have a steering wheel, while federal regulations may allow fully autonomous operation.
Why fully autonomous cars might actually be safer
If you're worried about the safety of self-driving vehicles, you might expect it to be safer to allow human drivers to take over in an emergency. It's easy to imagine nightmare scenarios where a self-driving vehicle suddenly malfunctions — or is hacked — and a helpless passenger is hurtled to his death.
Including an option for human-driven operation also allows self-driving technology to be conservative about when it takes control. In situations where the software isn't sure about the right thing to do — say, a construction site or a blizzard — it can hand over control to a human driver.
But some experts say that cars capable of switching between automatic and self-driving can actually be more dangerous than purely automated driving. That's because the most dangerous moment is the instant a vehicle switches between an autonomous and human-driven state.
If a driver wasn't paying attention just before the switchover, he's more likely to make a misjudgment about how fast the car is moving, which vehicles are around, and so forth. Human drivers might also get confused and assume the car is driving itself when it isn't. And over time, as cars become more and more automated, human drivers might get out of practice, making them less safe drivers if they're forced to take over in unusual situations.
This isn't just a theoretical problem — it's something that has cropped up with the autopilot feature on airplanes. As a 2014 New Yorker article pointed out, a number of crashes have occurred because pilots simply weren't paying close enough attention as the plan largely flew itself. That caused them to make crucial mistakes when they were forced to take over control.
One study found that with higher levels of automation, "pilots' ability to make complex cognitive decisions suffered a palpable hit. They were less able to visualize their plane's position, to decide what navigational step should come next, and to diagnose abnormal situations."
With the plane doing most of the driving, pilots had more trouble concentrating on the task in front of them. Their minds tended to wander That made them less well-prepared when an emergency required them to exercise good judgment and quick thinking.
Car companies are just starting to introduce partially self-driving cars onto the market, so we don't yet know if the same kind of problem will crop up on our roads. But it provides a powerful argument for advocates of allowing fully autonomous driving.
The steering wheel debate is also about business models
It's not a coincidence that Google has been leading the fight to allow fully self-driving vehicles, while many car companies prefer the gradual route. These technological approaches dovetail with the business strategies each company is likely to pursue.
Car companies are mostly in the business of selling finished automobiles directly to customers. And if you're buying a car, you naturally expect it to work in all circumstances. Yet building a car that can work in all circumstances — in rural as well as urban areas, in rain and snow, on mountain passes and in construction zones — is a difficult technical and logistical challenge that might take many years to fully solve.
There are two big issues here. One is that certain environments are just inherently challenging for an autonomous vehicle to navigate. A big snowfall, for example, will render most of a self-driving car's landmark-detection capabilities useless. Driving through a construction zone is another case where fully automated driving could be challenging, since lanes aren't as clearly marked as on a normal road.
The other issue is that self-driving software may need detailed maps to navigate safely. And mapping the entire country will take a lot more effort than mapping major populated areas.
A mixed-mode vehicle provides a nice solution to this dilemma. The car can drive in situations where it's safe to do so — when it's not snowing, not in a construction zone, and in an area where maps are available. The rest of the time, the human driver can take control.
Fully autonomous vehicles are going to have to take a different approach. If the vehicle isn't confident it can drive safely in a particular situation — in snow, in remote areas, on dirt roads, and so forth — it's going to have to refuse to drive altogether. And that means cars like Google's prototype that have no steering wheel are going to have some major limitations, especially in the early years.
Of course, that's going to be a hard sell if Google is trying to sell its vehicles directly to customers. Who wants to own a car that can only drive to certain parts of the country or certain weather conditions? But this wouldn't be such a big problem for an Uber-style on-demand service, since people who take taxis typically rely on a mix of transportation modes to get around anyway.
Over the long run, this is likely to emerge as the biggest cleavage in the self-driving car market. Uber is working on its own self-driving car technology, and we can assume they'll focus on building an on-demand service like the one they already have. GM and Lyft have a partnership to work on self-driving technology, suggesting that GM may be interested in shifting toward a more on-demand model.
We can expect these on-demand vehicles to come without steering wheels. And without the need for a driver's seat, they may wind up looking a lot different than a conventional car. They might be smaller, have rear-facing seats, and offer less storage space. With their ability to go back and recharge after a few trips, they may have smaller batteries and a shorter range — and as a result, be more affordable.
On the other hand, we can expect some car companies to focus on an incremental strategy, gradually adding self-driving capabilities to conventional vehicles.
My guess is that in the long run, the fully automated, on-demand model will come to dominate, especially in urban areas. But for a while, we're likely to have a mix of partially and fully automated vehicles, with fierce competition between them — both in the consumer market and in the halls of government.