Eric Volkman investigates whether we’re treading a dangerous path to Vision Zero.

It seems counter-intuitive, perhaps even scary. A number of people and organisations involved in the push towards autonomous driving have stated that the best way forward is to put self-driving cars on the road before they’ve been perfected.

This point of view got a significant boost last November. That’s when influential US researcher RAND Corporation published a study that suggested some form of early adoption would reduce fatalities in the long term, while improving autonomous systems through real-world use. “There are two benefits” said Nidhi Kalra, the study’s co-author, director of RAND’s San Francisco Bay Area Office, and senior information scientist. “first, we start saving lives sooner and second, we improve the technology faster… It's a bit like saving for retirement – the earlier the better.”

A big selling point in this argument is safety. RAND’s model, which can be adjusted for a set of variables on the organisation’s page devoted to the study, indicates significant savings in human lives even with very early-stage rollouts. If a user inputs variables that are fairly aggressive among the ones offered, such as self-driving cars to be put on the road in 2020 when they’re only half as safe as those operated by humans, the model still predicts a considerable saving of life. The 2020 estimate for annual traffic fatalities in the US is roughly 37,000 (the same figure for 2017). This tally, the model suggests, would increase to just over 40,000 before seeing a dramatic drop in the early 2030s to around the mid-20,000s. Selecting less aggressive options produces more profound savings.

That initial rise is an acknowledgement that a sudden, system-wide introduction of self-driving cars would make the roads less safe at first, which could very much hurt the cause of autonomy. Neil Boudette, an automobile reporter for The New York Times, points out that: “There's a huge difference between programming a machine to make a perfect latte and programming a car to navigate through an intersection or teaching it to read stop signs, recognise pedestrians and detect oncoming traffic.” An early rollout could, thus, fall anywhere on the scale from unreliable to catastrophic.

Adds Boudette: “I think the auto industry has to put AVs on the road in restricted, limited tests to learn to understand how they operate and how other drivers react to them.” A key advantage to this incremental approach is that the scope for fatalities is much narrower. Also, it gives human drivers, and autonomous passengers, more time to get accustomed to sharing the road with self-driving cars.

Yet, the early-bird approach recommended by RAND has its proponents. Among these is Caleb Watney, an analyst at think tank R Street Institute, who has written a study on autonomous rollout that drew broadly similar conclusions. According to him, at least some of the resistance to putting such vehicles on the road sooner rather than later is owing to the trust issue. We are still spooked about machines driving us around but this probably won’t last.

“I think people are inherently sceptical when you take control away from humans,” Watney said. “A really good example can be found with the switch from manual to automated elevators, people freaked outwhen human elevator operators were replaced! It took some time but, eventually, people realised that automated elevators were safer and cheaper than the human-driven alternatives.”

Perhaps part of the current fear has to do with perception. At times, the only news regarding autonomous solutions that seems to reach the public involves accidents and fatalities. “The spate of issues we're seeing with Tesla's Autopilot, which is not even an autonomous system, where people are misusing the systems or arguing that the Autopilot failed suggest that the public could view the technologies as unsafe and not have much appetite for them,” said Kalra.  
There’s an expectation-versus-reality factor here. “People think machines should be perfect. Like a calculator or computer,” said Boudette. “They are used to having machines handle tasks and performing them flawlessly and doing them over and over exactly the same way but driving involves a very chaotic and unpredictable environment.”

However, it’s still early days and the numerous pilot and limited-scope projects – from Waymo’s trials to Navya’s self-driving shuttle loops at the University of Michigan and Las Vegas – potentially could help put people at ease with autonomy. The success of these efforts indicates that auto makers could potentially have good, if not perfect, autonomous systems ready for rollout in the very near future. Additionally, the US federal government has started the ball rolling on the legislation front, writing and introducing laws on autonomy.

So, many of the pieces are in place. Early-rollout believers assert that we’ll make much quicker progress and reduce the casualty count when we get self-driving cars on the road, rather than keeping them at limited use. “If we force driverless cars to do all their testing on private safety courses, without any access to real-world situations that happen on public roads, it could be decades before we reach similar levels of performance,” said Watney.


Connected & Autonomous Vehicles

14 May 2018 - 17 May 2018, Santa Clara, USA

From vehicle electrification and infrastructure to the evolution of ADAS and vehicle automation to enhanced connectivity and new mobility models, no rock will be left unturned.

Related Reads