Their pivot from self-driving cars to self-driving trucks is the latest sign of how difficult it is to fully take our hands off the wheel, industry experts and researchers said.
"It is becoming more and more evident that automating trucks and transportation on highways is a more realistic goal for AVs than solving the general urban traffic automation problem," said Dimi Astropolous, senior scientist at Carnegie Mellon's Robotics Institute and the National Robotics Engineering Center.
Nearly two decades after another CMU professor, Red Whittaker, built an autonomous Hummer that drove itself 132 miles across the Mojave Desert, mass implementation of self-driving cars remains a distant goal. Experts blame safety, business economics and technological complexity for the delay.
In the meantime, they say it makes sense for pioneers to pursue the most promising business ventures as a means of continuing development.
On Sept. 7, Mr. Salesky and his team announced the launch of driverless truck company Stack AV, which has already secured an investment from Japanese tech conglomerate SoftBank, said to be at least $1 billion, according to Matt Smith, chief growth officer at the Allegheny Conference on Community Development. The startup follows in the footsteps of two other Pittsburgh-based driverless truck startups: Aurora, worth $2 billion last year, and Locomation, which shuttered earlier this year with only a couple of trucks tested in the region.
Softbank's backing gives Stack AV more of a fighting chance than Locomation, said Philip Koopman, associate professor of electrical and computer engineering at CMU. But it's unclear whether the advancements in trucking will bleed over into car technology.
"It depends who you ask," Mr. Koopman said. "For a while, the claim was that software could be shared between the two."
The business case
Experts describe the trucking industry as ripe for autonomy: An existing shortage of about around 78,000 drivers is set to double by 2031, according to the American Trucking Associations, while continuing e-commerce growth is increasing the demand for truck-based shipping.
Softbank may have a closer view of the situation than most: Japan is confronting its own logistical delays as a series of pro-worker regulations known as the "2024 problem" compound an existing truck driver shortage.
"The business case for making money with trucking is more straightforward," Mr. Koopman said. "You can book in advance; there's more dollars per ride involved."
Trucking has a higher cost per driver compared to robotaxis and an existing supply chain of potential customers. The routes are consistent and replicable. But the model is still relatively untested from a commercial standpoint.
"I don't know how they're going to make money," Mr. Koopman said.
Waymo, Google's AV spinoff that Mr. Salesky helped develop, shut down its trucking program in July to focus on robotaxis, which it described as a more viable route to commercial success.
Image DescriptionAurora, one of Pittsburgh's two driverless truck companies, is set for a commercial launch next year after already demonstrating that it can remove the driver from trucks driving between Dallas and Houston, Texas. (Courtesy Aurora)
Cars take a back seat
Aurora, Pittsburgh's other driverless truck company, is set for a commercial launch next year after already demonstrating that it can remove the driver from trucks driving between Dallas and Houston, Texas. The company has said its software could eventually carry over to cars.
Through a partnership with Toyota, Aurora is testing a small fleet of driverless Sienna minivans in Pittsburgh and Texas that employ a similar computer "driver" which includes real-time perception, mapping and control.
Stack AV has not formally announced plans to apply its technology to cars, though Mr. Salesky did describe trucking as "a good entry point" for autonomous vehicles in an interview with TechCrunch.
His team has deep experience in driverless cars, including from thousands of street tests carried out on Pittsburgh streets.
While some of that intellectual property was likely lost to the automakers that disbanded Argo, a trove of Argoverse data is still available online at CMU's Center for Autonomous Vehicle Research. The center was previously named after Argo and created with a $15 million commitment from the startup in 2019.
Losing out on the visual indicator of futuristic cars navigating Pittsburgh's streets doesn't mean that the region is losing its entrepreneurial prowess, said Audrey Russo, CEO of the Pittsburgh Technology Council.
"Not everyone here gets to work on products that are seen right in front of our faces. I actually appreciate the fact that their products will be in other places because I think that helps Pittsburgh," she said. "That helps tell a richer story."
And experts say Pittsburgh — with its dense and complicated infrastructure, featuring brick, cobblestone and even wooden roads — was always an ambitious place to test driverless technology.
"People think San Francisco is bad — Pittsburgh says 'hold my beer,'" Mr. Koopman said.
It would be unlikely to see another taxi fleet pop up here anytime soon, said Cetin Mericli, the former CEO of Locomation, given the costs associated with building and maintaining the vehicles.
"Particularly for Pittsburgh, it's as much of a business decision as a technical decision," he said.
Argo was launched during a period of "optimism" and "peer pressure" that pushed startups to spend more and move faster than they might have otherwise, Mr. Mericli said.
"This is one of the beauties of being in high tech," he said. "Sometimes things work, other times they turn out to be much more challenging than you can anticipate. We now know infinitely more than we did about the possibilities and challenges."
Image DescriptionExperts said that former Strip District-based driverless car startup Argo AI was launched during a period of "optimism" and "peer pressure" that pushed startups to spend more and move faster than they might have otherwise.(Pittsburgh Post-Gazette)
Safety keeps the spotlight
Driverless trucking technology has "lifesaving potential," Mr. Salesky told Bloomberg the day of Stack AV's announcement, echoing claims he and other AV manufacturers have made for years.
Computer drivers don't get tired or distracted. They can be trained on government-created crash models. And for trucks, the routes are even more replicable.
But the consequences of a single failure are also much greater.
"People will say it's simpler and easier," Mr. Koopman said. "I don't think that's strictly true. It's not as chaotic but the stakes are way higher."
The researcher said he's been impressed by Mr. Salesky's commitment to safety to date.
"Their track record is: This is very complicated, very hard. We want to take the time to do it right."
Experts say a commitment to safety will be a key part of future regulatory approvals and positive public perception.
Under Mr. Rander's leadership, Uber's robo-taxi unit struggled to regain momentum after a high-profile fatal crash in Arizona in 2018.
More recently, Cruise has had to answer for testing problems in San Francisco, which prompted outrage on social media and at municipal meetings. That backlash is "sensationalized" and overblown, Cruise CEO Kyle Vogt told The Washington Post the same day as Stack AV's announcement.
Last year, Aurora and Argo's teams helped to convince Pennsylvania lawmakers to approve the driverless bill, which allows testing on public roadways.
That doesn't mean the companies have proven their tools are safe.
"The safety problem is receiving a lot of attention in autonomous driving, but so far a definitive solution has not emerged," said John Dolan, a principal scientist at the CMU Center for Autonomous Vehicle Research. The problem is especially acute for trucks given their size and momentum and the damage they can cause in accidents, he said.
Insight from the field
Mr. Mericli of Locomation, said it's "very prudent" for Stack AV to start with trucks.
"When we were starting we were pretty skeptical and unconvinced about the unit economics of robotaxis. Even if you could solve the technology, how long would it take to bring the cost down?" he said.
And solving the technology was never a given. A fully autonomous car is the "holy grail," of AV engineering, Mr. Mericli said.
That's part of why he pivoted to his own development to focus on supply chains.
Mr. Mericli is now leading Atlas Robotics, an autonomous forklift startup headquartered in Pittsburgh with offices in Turkey. The lifts will operate more safely and efficiently than human operators and will help alleviate a workforce shortage in the palletized storage industry, Mr. Mericli said.
"Unlike trucks, they are confined within four walls, but they are just as pivotal a transformation," he said.
A former post-doctoral fellow at CMU, Mr. Mericli described Stack AV's founders as brilliant innovators and close friends.
"They're wonderful people and one of the best suited teams to tackle this problem. I'm clapping as hard as I can for their success." Mr. Mericli said. "Hopefully we will fix the issues of the supply chain from cover to cover."
© 2023 the Pittsburgh Post-Gazette. Distributed by Tribune Content Agency, LLC.
