Opinion: Despite Advances in Self-Driving Technology, Full Automation Remains Elusive

In the visions of techno-utopians and techno-pessimists, the coming decades are filled with taxis, 18-wheelers, and family automobiles all driven by computers. This, despite the fact that today’s self-driving vehicles still struggle with mundane tasks like left turns and highway merges.

VIEWPOINTS
Partner content, op-eds, and Undark editorials.

If the past is any guide, however, the cars and trucks of the future may be more dependent on human help than we imagine. More than 50 years ago, automation efforts began on a different type of vehicle — one that suffers far fewer technical challenges, because it moves on rails. I speak, of course, of trains. Decades later, trains of varying levels of automation are now widespread, but humans continue to dominate train cabs. The reasons have less to do with technological challenges than with social and economic pressures. Those same pressures could prevent cars and trucks from ever being fully autonomous.

Experiments with automated subway lines and freight trains date back as far as 1962. Some systems from that era were quite robust; Dallas-Fort Worth’s driverless “Airtrans” people-mover, inaugurated in 1974, served until 2005. Today, driverless-capable track mileage in urban metros is doubling every decade. More than 70 metro lines in 40 cities worldwide are now fully automated, and many more lines support “semi-automatic train operation,” where human drivers open and close doors and press “go.”

Intercity trains have been slower to buy in — big swaths of open track are less predictable than underground or elevated guideways — but Germany and France plan automated passenger lines by 2021 and 2023, respectively, and a driverless freight line in Australia is on track to fully launch by the year’s end. In the U.S., even manual trains now must use “positive train control,” which automatically keeps trains clear of one another and within speed limits.

But even in nominally driverless train systems, humans often pull strings from behind the scenes. Much like Airtrans, France’s “drone trains” will be remotely driven by human “téléconducteurs.” Even the most autonomous systems are supervised by remote operators who can take control of the train if needed. In fact, says Ashley Nunes, a research scientist at MIT’s Center for Transportation and Logistics, “we have yet to find a single example of a safety-critical system or setting where there is no [real-time] human oversight. We can’t find one.”

Why has it proven so difficult to take the last few humans out of the loop? Part of the reason is that humans have superior intuition for detecting anomalies and responding to exceptional circumstances. Trains are rarely expected to halt for obstacles — the obstacles are expected to keep out of the way — but drivers who are familiar with a route can often catch small problems before they turn into big ones. In New York City, for instance, locomotive engineers and train operators often call in about an unusual sparkle ahead or strange motion over some stretch.

A similar dynamic is emerging in self-driving cars. Although the companies that make autonomous vehicles (AVs) aim for maximum vehicle autonomy, they widely accept that cars will occasionally need to call in human assistance for finer judgment. Silicon Valley startup Phantom Auto offers “tele-operation” for nominally autonomous cars, and manufacturers including Nissan, Waymo, and General Motors are configuring their vehicles to consult humans in difficult scenarios.

But humans’ enduring role in rail stems from more than just their driving skills. Even a fully self-driving train would need people to install and service equipment, respond to failures, and physically couple and decouple cars and cabs. Some trains even need on-board staff at all times: if a passenger train breaks down, for instance, “it’s far better having [a] person on the train and available than leaving 1,300 or 1,700 people by themselves…awaiting someone to arrive,” says Gerard McFadden, engineering director of England’s Govia Thameslink Railway.

Again, self-driving road vehicles will likely face a similar situation. Automated freight trucks, for instance, will still need people to make repairs, manage inventory, and load and unload deliveries — tasks currently handled largely by drivers. Self-driving taxi fleets, meanwhile, will need substantial support teams to address rider concerns or interact with law enforcement.

All that is to say, “driving” comprises a broad suite of activities, some of which are more easily handled by computers than others. And as often as not, the factors that ultimately dictate which activities get automated when have less to do with technology than with economics and politics.

On the economics side, automating rail reduces costs by making it possible for a handful of human workers to oversee a large fleet of trains taking trips at higher frequencies than could be achieved under manual control. For rapid transit and short-range people-movers, where convenience and capacity are king, the potential payoffs of those efficiency gains are huge. That’s less true of longer-range passenger trains, which run less often and require on-train conductors to interact with customers. Freight is in a similar boat: As Dale Lewis, former director of strategic analysis for CSX, puts it, a freight train can already move up to 300 truckloads across the U.S. with just seven two-person crews, “so what’s the payoff for getting the last two people off the train? It’s not a very big number.” A rare exception is the soon-to-be-automated train in Australia, run by mining giant Rio Tinto, which is worth it only because shift changes in the middle of the endless outback are so costly.

Likewise, the areas where autonomous vehicles are making inroads are those where potential efficiency gains are greatest: long-distance trucking, taxi services, and local shuttles. The high upfront investment in automation pays off quickly for large fleets that can be operated continuously with minimal human oversight. Automation is less economical in the private car market, which is why most driverless AV purveyors are not eyeing such sales.

But perhaps the biggest impediments to rail autonomy have been political. For one, there’s the liability issue: Eliminating the driver could leave rail operating companies to shoulder the blame for any incidents. But regulation also plays a role. In the U.S., the regulatory environment does not favor rail automation; in fact, a proposed Federal Railroad Administration rule would mandate a two-person minimum crew in most locomotives.

That proposal belies the influence of another political heavyweight in the automation fight: the labor unions and the jobs they defend. “Even if you had [legal] permission” to cut engineers or conductors, says Lewis, “you’d have to negotiate it with each of the respective unions.” The situation recalls that of “firemen”— the crewmen who fed locomotive steam boilers — after diesel replaced steam: Backed by unions, firemen held on as mandatory crew for over two decades. The jobs issue is especially salient in public transit, Nunes says: “When you’re talking about the public purse, the prospect of jobs being lost is politically very difficult to swallow.”

Autonomous cars and trucks face similar political obstacles. Current rules, holdovers from the pre-automation era, mandate that AVs be equipped with human-friendly features such as pedals. Though that may change soon, legislative bills to encourage autonomous technology in the U.S. have stalled in the face of public skepticism and opposition from safety advocates and unions.

The push to automate trains, then, reveals two truths about autonomous vehicles. First, transportation of any sort is a beast with many arms, not all equally subject to automation. And second, the extent to which the vehicles are automated will largely be a function of economic tradeoffs and political support. With Google sibling Waymo already running a small self-driving taxi service, it remains to be seen — and determined by us — which driving-related roles will be automated and how soon.