Dave Mercer has been driving trucks across America since 1986. He has hauled ice cream to Nevada, burgers to Oregon, and trailers to Baltimore.
“It can be a rough life,” he says. “You’re away from your family a lot, and raising your kids from the road is a hard life. I’ve talked to a lot of principals on the phone. But it also provided me a good lifestyle. I drove through all the recessions and depressions and don’t recall ever slowing down.”
Three and a half million miles — and one divorce — down the road, Mercer was ready for a change of pace. In 2015, he joined Peloton Technology, a Californian startup developing self-driving trucks using robotics and artificial intelligence. Mercer helps test that technology on tracks and public roads, preparing for the day when truckers might leave their cabs for good.
“We can’t play stupid to the fact that down the road there’s probably going to be a fully automated truck running up and down the highway,” Mercer says.
Head to the California Department of Motor Vehicles (DMV) website and you can read reports of every collision involving the cars enrolled in its autonomous vehicle test program. They are a litany of mundanity. On March 29, a scooter bumped into one of GM Cruise’s self-driving Bolt cars and damaged its left-side wing mirror. Two days earlier, a BMW SUV clipped another Bolt while changing lanes. The big excitement on March 24 was that two of GM’s cars were rear-ended on the same day, resulting in some minor scratches and dents.
What you probably won’t have heard about is the first-ever crash involving a self-driving truck, which occurred last November on California’s Junipero Serra Freeway in Silicon Valley. At 9:45 in the morning of November 9, a semitruck owned by startup Starsky Robotics was involved in a collision with a passenger car. The car hit the truck’s tires and broke a mud flap and had to be towed from the scene. There were no injuries.
This collision only came to light because the California Highway Patrol reported the crash to the U.S. Department of Transportation (DOT), as it must for any incident involving a commercial motor vehicle. Because California does not currently have a testing program for self-driving trucks, developers do not have to file reports on failures and crashes. (Nor are they permitted to test advanced driverless technologies. Starsky did not confirm whether any of its self-driving systems were engaged at the time of the collision.)
Other states are even less restrictive. Arizona and Florida welcome self-driving trucks without any special permits or reporting; in February, one of Starsky’s big rigs drove a seven-mile test run near the Everglades without a human in the vehicle at all. One of its rivals, Embark, completed a coast-to-coast run from Jacksonville, Florida, to Los Angeles in an automated truck the same month.
Unlike the high-profile urban and suburban tests of autonomous passenger cars conducted by Waymo, Uber, and GM Cruise, driverless truck developers have been quietly carrying out their experiments on highways across the Sunbelt. According to figures submitted to the DOT, automated trucks racked up well over 200,000 miles on public roads last year.
A truly driverless truck would change the face of trucking. In a cutthroat industry operating on razor-thin margins, drivers’ wages and benefits are the largest cost, accounting for nearly half of a carrier’s cost per mile. The first company that can eliminate drivers could undercut its rivals and dominate the market.
Automated trucks might also reduce the nearly 4,000 fatalities from large truck crashes that occur each year in the United States, some of which are known to result from driver fatigue. There is even a potential environmental benefit, as smart driving systems reduce fuel use and smooth the transition to alternative fuels and electric vehicles.
But none of this is guaranteed. Cheaper freight transport could stoke demand, wiping out any carbon savings and increasing congestion. Millions of well-paying blue-collar trucking jobs could disappear, and the safety benefits have yet to be proven. Driverless trucks may simply replace today’s human errors with higher-tech mistakes — and an out-of-control 18-wheeler could cause many more fatalities than the failure of Uber’s self-driving car in Tempe, Arizona, in March that resulted in the death of pedestrian Elaine Herzberg.
Automated trucks have only just started to roll, and their final destination is anything but clear.
Trucking is the original internet of things, a $700 billion juggernaut that moves 70 percent of all goods in America and employs millions of drivers, mechanics, loaders, and managers. But it is also a deeply conservative industry, dominated by a few slow-moving incumbents and a long tail of tiny mom-and-pop businesses, using technologies and business models that have remained largely unchanged for decades.
Engineer Josh Switkes was one of the first to realize that trucking was ripe for Silicon Valley’s brand of high-tech disruption. “I got excited about automated trucking 10 years ago for the same reasons people are getting excited about it today,” he says. “It’s a huge industry with very concentrated expenses in fuel, in crashes, and in labor. You have big enterprise customers where you can deploy broadly quite quickly and make huge economic benefits.”
In 2011, Switkes launched Peloton, the startup Dave Mercer now drives for. “At that time, Google had just announced their self-driving car project, but most people saw vehicle automation as something pretty far off in the future,” Switkes says.
Instead of trying to build a complete self-driving truck, Switkes would focus on a technology called platooning. Tour de France cyclists harness the aerodynamic efficiencies of riding in tight packs to travel much faster than they could individually. In the same way, Peloton’s trucks would tailgate each other at a distance of just 50 feet, using a combination of radar, video cameras, and short-range radio messages to automatically synchronize braking and acceleration.
“A midsized truck fleet will spend hundreds of millions of dollars a year on fuel,” Switkes says. “If we can save even a few percentage points of that, it will make a big difference to their bottom line.”
Peloton’s system is likely to go into service this year with several large haulage companies, but it won’t change much about the way truckers or companies operate. Both drivers in a platoon still need to steer and pay constant attention to the road, and the system only operates on highways that have been mapped by Peloton and that aren’t experiencing severe weather or traffic jams.
“It’s not going to be a different job with platooning,” Mercer says. “When I used to drive cross-country, we were running pretty close, but we didn’t have the technology to really keep it safe. This technology adds that safety factor.”
Mercer is excited about the new technology, but he’s skeptical that it will replace human drivers anytime soon. “We’re fooling ourselves if we think we can throw a fully automated truck out there on the highway at this point in time and have the public accept it,” he says. “I wouldn’t want to see it. I wouldn’t want to pull up beside a truck myself and see no driver in there.”
But that’s exactly what another startup, Otto, did early in 2016. Otto was founded by Anthony Levandowski, a brilliant young engineer who built the world’s first self-driving motorbike for a Pentagon competition in 2004 and Google’s first self-driving car in 2008. But Levandowski did not fare well at Google, where he was passed over for a leadership role and disgruntled by what he perceived to be slow progress. He decided to jump ship — taking as many Google engineers as he could with him — to develop much bigger vehicles.
In May 2016, Otto released a video showing a full-size semitruck barreling down a Nevada highway with no one behind the wheel. Otto said it was developing a truck that could be trusted to drive itself on highways without human supervision. That would be a big deal, because federal regulations restrict human truck drivers to 11 hours of driving in any 24-hour period. If drivers could cede control to a robot while they rested, they could travel more than twice as far in a day — slashing that critical cost per mile.
“When we feel safe enough, we’re going to virtually tap on the truck driver’s shoulder and say for the next 100 miles, ‘don’t worry, we got it,’” Lio Ron, Otto’s co-founder, told me at the time. Within months, Uber had acquired Otto for a deal reportedly worth $680 million.
Otto moved fast but it also broke things to get there — namely Nevada DMV’s regulations for self-driving vehicles. Nevada rules required Otto to get an autonomous vehicle testing license before using its self-driving systems on the state roads and to post a multimillion-dollar bond to cover any accidents. In any case, operating an autonomous car — let alone an 80,000-pound semi — without two human safety drivers was strictly forbidden.
“Otto is driving these vehicles illegally and without the required $5 million bond that is needed to protect our citizens,” wrote Jude Hurin, the administrator in charge of autonomous vehicle regulation at Nevada’s DMV at the time. Levandowski would soon have more to worry about, as he was accused of stealing Google’s self-driving technology in an epic trade secrets and patents lawsuit that was settled earlier this year.
Lior Ron eventually admitted in court that he received only $20,000 from the sale of Otto to Uber, as the acquisition failed to meet targets set by the ride-sharing giant. But the tale of a startup just a few months old being bought for more than half a billion dollars had already worked its magic, and, in 2016, automated trucking became the latest Silicon Valley gold rush.
Within months, a small startup called Varden Labs pivoted from making low-speed self-driving shuttles for university campuses and business parks to the newly shiny world of trucking and changed its name to Embark. In a virtual shot-for-shot remake of Otto’s video, Embark revealed its own self-driving truck early last year. Like Otto, Embark shot its video in Nevada, and, like Otto, there was no one behind the wheel.
Embark even went one step further, showing its driverless truck blasting past an RV on a narrow desert highway on the wrong side of the road. “Overtaking with a semitruck on a two-lane highway is a very risky maneuver,” said Paul Ashbourne, an autonomous truck engineer at Embark, at a conference in March 2017. “It’s not something that we would actually do in the real world.” Ashbourne is partly correct. In the United States, about 75 percent of traffic fatalities occur on two-lane roads, many as a result of head-on collisions, and trucks kill about 20 percent more people than passenger cars per million miles driven.
However, the video very much took place in the real world. Documents obtained under public records legislation show that it was shot by filmmakers from the Matrix and Bond movies on Nevada’s Highway 267, near the California border. Although the video was shot on an open highway, Embark says that there were no unaffiliated private vehicles in the area at the time of filming.
“Our view was, coming a bit later to the game, we needed to be able to show off a little bit higher level of technology,” Ashbourne said. A year after Otto, however, Nevada’s DMV did not issue a peep at either Embark’s driverless test or its risky overtaking maneuver. In fact, says Jude Hurin, “Safety was always kept as the highest priority and was not compromised.”
So, what changed? For a start, Embark did not go behind the regulators’ backs. It applied for a testing license, filed a bond, and received special autonomous vehicle license plates for its truck. But a bigger part of the story is what had happened to Nevada’s regulations since Otto’s illegal run.
Shortly after Otto launched, Nevada quietly changed its regulations for testing automated vehicles. No longer would manufacturers need to put two human safety drivers in each vehicle—only one. And that operator would no longer have to sit behind the wheel with immediate access to traditional driving controls. Embark claims that it had a safety driver positioned, unseen, within the cab and ready to take control of the truck in under a second if its systems failed.
When Starsky Robotics arrived on the scene, it upped the stakes once more. While Starsky is working hard on autonomous technology for highway driving, it is also developing teleoperation systems that let drivers take remote control of the truck from an office hundreds of miles away. Self-driving systems aboard the truck handle the easy highway miles, then hand over to a licensed human driver for tricky situations like navigating exit ramps, urban streets, and delivery depots.
“Because of teleoperation, we can already do fully unmanned journeys,” says Stefan Seltz-Axmacher, founder and CEO of Starsky. “ And when our autonomy gets a little better, when we don’t need the person for big stretches of time, we can probably have one driver overseeing 10 to 30 vehicles during an eight-hour shift.”
Seltz-Axmacher says Starsky will follow up its seven-mile unmanned voyage on a closed road with a 100-mile trip on a public highway in Florida later this year. “It won’t be a closed road, but we will have one of the lanes restricted,” he says. “The first time there’s no person in the vehicle, you don’t want weird things to happen.”
It will likely be several years before Starsky’s fully roboticized and remote-controlled trucks are hauling large quantities of goods commercially. While some individual states welcome the technology, the moment a truck crosses a state line, it must comply with a broad range of federal regulations. A recent DOT report concluded that automated trucks “that either require an on-board (non-driving) human technician, or do not require an on-board human at all, are likely to face significant compliance challenges.”
Beyond the red tape, says Seltz-Axmacher, “So far I’m not convinced that the major [manufacturers] will be able to flexibly build or acquire autonomy companies.” The one exception could be Tesla, which is already selling a semi-autonomous system called Autopilot for its Model S, X, and 3 passenger cars. Autopilot can handle lane- and distance-keeping on some roads, but it requires drivers to keep their hands on the wheel and potentially resume driving at very short notice.
There have been many scientific studies showing that human drivers quickly lose their concentration when required to oversee automated driving, and even research on Tesla’s system specifically. Autopilot has been implicated in two fatal collisions in the United States and possibly one in China, where drivers were unable to retake control of their cars in time to avoid a crash. Tesla insists it has taken many steps to prevent drivers misusing their Autopilot systems.
Tesla is now incorporating the same Autopilot system in its new all-electric semi-truck, which is already hauling loads for Tesla’s factories in California and beyond. This is not an approach that Peloton favors. “We put a lot of effort into the making sure the driver isn’t distracted or drowsy,” Josh Switkes says. “It’s important to keep them engaged — having them steering is perfect.”
All the companies I spoke with believed their technologies should ultimately reduce road deaths by giving drivers the time and tools to drive more safely and reliably. But the challenge for regulators is ensuring that future safety improvements do not come at the expense of increased danger today.
“A wide truck has a much smaller margin of error available than a small car, and basic physics suggests that an errant truck could do more damage than an errant car,” says Bryant Walker Smith, a law professor at the University of South Carolina and an expert in automated vehicle regulation. “If these tests are on roads open to the public, then there’s certainly the potential for harm.”
The death of Elaine Herzberg makes that potential seem all the more likely. After the collision, Uber immediately suspended all testing of its fleet of around 200 self-driving cars in the United States — as well as its 13 self-driving trucks, the legacy of its acquisition of Otto. Uber is now focusing on Uber Freight, a smartphone app that quickly connects drivers with loads, similar to its passenger taxi service.
Some rivals, including Waymo, which operates a handful of self-driving trucks, were quick to state that their vehicles would have detected and avoided Herzberg. But while something clearly went wrong in Tempe, mileage data suggests that Uber is far from the least experienced operator of automated trucks today. In fact, data supplied by manufacturers to the DOT shows that Uber’s self-driving trucks covered 105,000 miles last year. That is twice the distance of its nearest competitor, Embark (50,000 miles) and nearly four times as far as Waymo’s trucks (26,738 miles).
If the narrative that self-driving trucks will reduce road deaths is unproven, the equally common perception that they will also decimate jobs seems on similarly shaky ground. Last year, the International Transport Workers’ Federation predicted that automation could eliminate up to 4.4 million trucking jobs in the United States and Europe by 2030. In fact, one of the key drivers for automation is not to fire existing workers, but to help fill job vacancies that carriers are struggling to fill — even with salaries topping $75,000.
“Ever since I’ve been driving trucks, for 32 years, there’s been a driver shortage,” agrees Mercer. “The trucking industry has needed a face-lift for a long time, and Peloton can provide that face-lift without drivers having to be fearful of losing their jobs.”
Steve Viscelli is a sociologist and author of The Big Rig, a book about the decline of American trucking. He points out that truck drivers do a lot more than just driving, including loading and unloading, maintaining the vehicle, fueling, negotiating, and completing paperwork. “If we’re just going to automate the one task of driving, which companies are going to be able to adopt it without having to make major changes to the overall process?” Viscelli asks.
His latest analyses suggest that a much smaller number of jobs are vulnerable to being completely replaced by automation, perhaps 250,000 in the United States. “And about half of those are pretty crappy entry-level jobs that people don’t stay in for long anyway,” Viscelli adds. “I think it could be over a decade before we see any kind of job impact at all.”
Even if the coming wave of automation does not eliminate many human roles, it could make it trickier to attract new truckers, says Karen Levy, an assistant professor at Cornell University and author of an upcoming book about high-tech trucking. “A lot of people get into trucking precisely because they want more autonomy and don’t want someone looking over their shoulder the whole time,” she says. “There’s a romance about trucking, even though it’s objectively not a great job.”
Levy points out that many drivers get paid by the mile rather than by the day. If robots are driving the easy highway miles, people could be left with the harder and less well-paid suburban routes. And while the Uber Freight app gives truckers more of the flexibility they crave, it might squeeze out more jobs among brokers and carriers.
Self-driving trucks are coming. Even if Silicon Valley’s current fleet of startups are acquired or fall by the roadside, the big truck manufacturers — Daimler, Volvo, and Paccar — are revving up robotic programs close behind. Autonomous trucking does not lack for technology, capital, or enthusiasm, and its promise for reducing road deaths and pollution is plausible.
But what could be a challenge is public and political acceptance. Experimental self-driving trucks are driving beside us on highways today with little transparency, regulation, or oversight. Uber’s fatal collision in March shows how quickly attitudes can change and business plans flounder in the face of perceived hubris or incompetence. If the next accident one of these trucks is involved in is more than just a fender-bender, robotic trucking could stall before ever really gets moving.