Self Driving Cars - Whats your take?
The Very Human Problem Blocking the Path to Self-Driving Cars
IT WAS A game of Dots that pushed Erik Coelingh to rethink his entire approach to self-driving cars. Coelingh, Volvo’s head of safety and driver assist technologies, was in a simulator, iPad in hand, swiping this way and that as the “car” drove itself, when he hear an alert telling him to take the wheel. He found the timing less than opportune.
“They gave the message when I was close to getting a high score,” he says. Jolted away from the absorbing task, he had no idea of what was happening on the “road,” or how to handle it. “I just realized,” he says, “it’s not so easy to put the game away.”
The experience helped confirm a thesis Coelingh and Volvo had been testing: A car with any level of autonomy that relies upon a human to save the day in an emergency poses almost insurmountable engineering, design, and safety challenges, simply because humans are for the most part horrible backups. They are inattentive, easily distracted, and slow to respond. “That problem’s just too difficult,” Coelingh says.
And so Volvo, and a growing number of automakers, are taking you out of the equation entirely. Instead of developing autonomous vehicles that do their thing under most circumstances but rely upon you take the wheel in an emergency—something regulators call Level 3 autonomous capability—they’re going straight to full autonomy where you’re simply along the ride.
Google figured this out around 2012, when it decided that full autonomy—no steering wheel, no pedals, no human backup—was the best way forward. Almost everyone else has embraced this way of thinking, abandoning the step-by-step approach and promising to begin launching fully robotic cars within a few years. The shift came as automakers recognized the difficulty of the “handoff”—getting the person behind the wheel to take control at a moment’s notice.
Automakers also saw only incremental improvements in safety, convenience, and value by advancing from Level 2 autonomy—cars that can keep their lane and handle rush-hour gridlock—to more sophisticated systems that still require human intervention. Going straight to levels 4 and 5 and offering a fully autonomous vehicle creates new markets, and new opportunities to challenge the likes of Uber and Google.
The Handoff Conundrum
It should be noted that these designations, defined by SAE International, are are squidgy, and don’t directly correlate to specific vehicles automakers are developing. For the sake of this discussion, Level 3 autonomy defines cars capable of basic decision-making like when to change lanes or pass other vehicles. The human at the wheel can check out entirely to, say, play an iPad game, but must be ready to take control if something goes amiss—a sensor fails, for example, or the car’s map doesn’t quite match the terrain.
Level 3 seems like a natural evolution of the tech you find in Tesla’s Autopilot, which demands vigilance even if not everybody obeys. More work for the robot, less for the human. But it’s a Herculean challenge for engineers and designers. “Having a human there to resume control is very difficult,” says Bryan Reimer, an MIT researcher who studies driving behavior. Once relieved of the burden of constantly paying attention, people are quick to lose focus, and getting them back on task is difficult. Imagine you’re watching the final moments of The Shining when someone suddenly turns on the light and tosses you a Rubik’s cube. How quickly could you register what’s happening, let alone attempt to solve the puzzle? Now you see the challenge of the handoff.
To make Level 3 work, the car must verify its human hasn’t, say, dozed off or strapped a VR headset to his face. This involves installing cameras and sensors to monitor things like head position and gaze direction. It means providing visual, aural, and haptic alerts to get the person’s attention. And it requires making absolutely sure the autonomous technology is robust and sophisticated enough to handle any situation for the 5 to 10 seconds needed to for the human to realize what’s happening and take control.
Simply put, solving this problem is almost as difficult as figuring out how to make cars drive themselves. That’s why Google—whose autonomous effort is now called Waymo—almost immediately abandoned any thought of building anything but a fully autonomous car. It started with a system that could handle highway driving with human oversight. Google’s engineers soon realized those humans were lulled into paying zero attention, and that they were all but useless in such circumstances. So they started pursuing full autonomy.
A tech company like Google or Uber can go for the moonshot, but automakers tend to be more conservative. They prefer small steps, gradually refining and introducing new technology to prepare consumers for the changes ahead. And so most of them planned to progress steadily through the ranks of Level 2, 3, and so forth. Ford was among the first to break ranks, announcing in late 2015 that it would skip Level 3. “We’re really focused on completing the work to fully take the driver out of the loop,” Ken Washington, the automaker’s head of research and advanced engineering, said at the time.
Beyond being difficult to achieve, Level 3 autonomy is difficult to justify. If every car on the road featured Level 2 capabilities, fatal automobile collisions would drop by 80 percent, according to Delphi, one of the world’s largest automotive industry suppliers. Level 3 doesn’t advance the ball much further, so why bother? Full autonomy, on the other hand, brings safety improvements while also bringing mobility to people who cannot drive, automating deliveries, and creating other opportunities.
The Money Question
Still, any automaker willing to throw enough time, money, and engineers at the problem can solve the Level 3 conundrum. Audi has all of those things in great quantity and plans to bring Level 3 capability to its flagship A8 sedan in 2018. The car will handle stop-and-go traffic to start, with highway capabilities to follow. Audi, which offers some of the best user interfaces in the business, has spent many years and many dollars studying the handoff, which it considers a top priority. Its solution, while not yet complete, will include driver monitoring systems and a combination of alerts the human will see, hear, and feel.
Still, even Audi sees the future, and so it is quietly pursuing a parallel track toward full autonomy. Like everyone else in the industry, it sees new markets and new competitors in a game that is rapidly changing. Car-sharing and other alternatives to ownership are growing increasingly popular, and upstarts like Uber see an opportunity to push established players aside. That’s why Ford, which more than any other outfit made automobiles ubiquitous, is rebranding itself as a “mobility” company and General Motors is buying robo-car startups and working with Lyft to develop an autonomous ridesharing network.
It is difficult to overstate the impact Uber and Lyft have had on this. The financial allure of ridesharing services that don’t rely upon human drivers has changed the calculus. “It’s an arms race to provide the mobility on demand,” says Delphi CTO Jeff Owens.
Uber wants to drop the human chauffeurs who gobble up the majority of its customers’ fares. Google could free up time in the car for riders to use its other services. Cities who host these fleets of robo-cars get a new way to tackle costly congestion and free up parking spaces for other uses. And the automakers—Ford, GM, BMW, and others—might slice off their chunk of an industry whose potential value Boston Consulting Group pegs at $42 billion a year by 2025.
But none of that works until they take you entirely out of the picture—and let you finally get that Dots high score.