Why Tech and Car Companies Have Such Different Visions For Self-Driving Vehicles

It's about culture: iteration versus disruption, incrementalism versus speed, now versus tomorrow. JL

Timothy Lee reports in Vox:

Car companies expect to develop self-driving capabilities gradually over time, with software gradually taking over human capabilities. In contrast, Google (like Uber, which started testing self-driving cars last year) hopes to come out of the gate with fully automated vehicles, which would open the door to their use in ride-hailing services.
For much of the 2010s, Google was the only company putting serious resources into the technology underpinning self-driving cars. That started to change in 2016, as almost every major automaker announced technology partnerships to develop their self-driving capabilities.

In 2017, we’re going to start to see the fruits of those announcements on our streets. By the end of the year, there will be a lot more self-driving car prototypes on the streets of cities like Detroit, Pittsburgh, and Austin. And a lot of them are pursuing a vision for self-driving vehicles that’s different from that of Google, long seen as the technology leader.

Most car companies expect to develop self-driving capabilities gradually over time, with software gradually taking over human capabilities. In contrast, Google (like Uber, which started testing self-driving cars last year) hopes to come out of the gate with fully automated vehicles, which would open the door to their use in ride-hailing services.

At this week’s Consumer Electronics Show, almost every major carmaker announced plans to accelerate its self-driving car efforts. Many showed off their latest prototypes, and some announced new partnerships. And almost all are vowing to have their cars on public streets before the end of the year. The big question facing many of them, however, is whether the gradualist approach they’ve chosen is the right one.

Car companies love the gradualist approach.

Gradualists use a five-level system created by the auto industry trade group SAE International to describe the shift to fully self-driving vehicles. Many automakers are already introducing driver-assist features like advanced cruise control — in which cars detect when they are approaching other vehicles, and modify speed — and stay-in-lane capabilities (SAE levels 1 and 2). The next step is partially automated systems (levels 3 and 4), culminating in “level 5” systems that are fully automated.

But others — including Uber and Waymo, the new name for Google’s self-driving car business — think this whole approach is a mistake. They hope to develop vehicles that are fully self-driving from the outset. Indeed, they argue that partial automation is a safety hazard, since the shift from human to software control is a major source of safety problems.

The idea behind level 3 and level 4 automation is that the car would handle routine tasks but would have the ability to hand the vehicle off to human drivers in situations the car’s software isn’t sure how to handle. But there are a couple of potentially big problems. One is that the automation of routine tasks makes it much less likely that the human driver will be paying full attention in the seconds before the car wants to hand over control to the human driver. If the issue requires immediate action, the human driver is much more likely to make a mistake than he would have if he had been driving (and hence paying close attention) all along.

This is a problem that has been well understood in the aviation world for years: The seconds after a plane switches from autopilot to manual control are often the most dangerous because of the heightened risk that the pilot is not fully prepared to take control of the airplane, and pilots’ skills get rusty if they rely on autopilot to guide the plane most of the time. So advocates of full car automation argue that we need to fully take human drivers out of the loop in order to reap the gains from automation.

And if cars are fully autonomous from the outset, that would enable their use with Uber-style ride-hailing apps. This would have some significant advantages. For example, the service could be introduced one city at a time as maps and other infrastructure are created in each locale.

But the gradualist approach is a better fit for car companies’ existing business model of selling cars to consumers through dealerships. People who buy cars expect them to work everywhere, so a car that doesn’t work in certain parts of the country — or certain weather situations — would be a nonstarter. The gradualist approach also allows car companies to build on their existing car designs.

Unsurprisingly, therefore, tech companies tend to be more enthusiastic about jumping straight to full autonomy, while car companies are more inclined toward gradualism. You can see this divide at work in speeches given by company executives at CES this week.

At a Wednesday event, Gill Pratt, the CEO of the Toyota Research Institute, the home for Toyota’s self-driving car research, insisted that full “level 5” autonomy was far in the future. “We’re not even close” to the breakthroughs necessary for full autonomy, he said. “It’ll take many years and many more miles, in simulated and real world testing, to achieve the perfection required for level 5 autonomy.”

Instead, Toyota is working on getting its cars to “level 4” — capable of handling the routine aspects of driving while a human driver keeps a watchful eye for emergencies.

But Lyft executive David Baga was more optimistic. Speaking at a CES panel, he predicted that half of Lyft’s cars would be fully autonomous within five years and Lyft’s fleet would be fully autonomous within a decade.

Hyundai is showing off a prototype self-driving car at CES this week. It has focused on using less expensive sensors in its self-driving cars in a bid for affordability — a clear sign that it expects its cars to be sold, not hailed on demand.

Some car companies are hedging their bets

In an announcement last August, Ford vowed to have cars with level 4 autonomy — where, again, the driver would have to be ready to take over in unusual situations — by 2021. At the same time, Ford has created a subsidiary called Ford Smart Mobility that will start to develop mobility services such as car sharing, which will give Ford some options if full autonomy arrives sooner than expected.

GM signed a deal with Lyft a year ago to develop self-driving cars. Fiat Chrysler is working to incorporate Google’s self-driving technology into some of its vehicles. These partnerships will serve GM and Chrysler well if an immediate jump to full self-driving becomes feasible earlier than expected.

Last summer, BMW announced a partnership with Intel and Mobileye, a leading maker of autonomous vehicle technology. In its press release announcing the deal, Intel endorsed the gradualist approach, predicting that it would introduce level 3 and level 4 vehicles before reaching level 5. At CES, the trio announced that they plan to start putting test vehicles on the road in the second half of 2017.

Over the next year, we’ll get a much clearer sense of how the self-driving car business will actually shake out. With so many companies entering the market, Waymo — the Google company — will feel a lot more pressure to get its technology into the market after more than seven years of development.

Uber will have to address complaints about the way its recently introduced cars drive. For example, when it briefly tested its vehicles in California (before being asked to stop by regulators), it faced charges that it mishandles turns that crossed a bicycle lane, potentially endangering cyclists.

Until now, most car companies have just announced partnerships and showed off concept cars at trade shows. But most have now committed to regularly testing their vehicles on public streets. We’ll start to learn which car companies — if any — have the technical chops to create their own self-driving technology, and which will be forced to license self-driving capabilities from more sophisticated players