By the end of this decade, trucks powered by supercomputers will be hauling freight across the country on some of the safest roads in a century. Driver assistance technologies on commercial and passenger vehicles are more prevalent each year, creating the potential to reduce road crashes caused by human error. As artificial intelligence (AI) grows more powerful, there won’t be drivers operating some commercial vehicles running on major freight corridors.
The efficiency that could come with Class 8 autonomous vehicles (AVs) is nearly endless. But how safe can what equates to an 80,000-lb. robot be motoring among human-controlled vehicles on the interstate? That is a question that Ariel Wolf, general counsel for the Self-Driving Coalition, welcomes.
“The thing I try to emphasize is the baseline,” Wolf told FleetOwner. “What’s the baseline today? We’ve accepted — because we have no alternative — the level of death and destruction that occurs on the roads today. It’s just something we’ve grown accustomed to. It’s not just the 36,000, 37,000 deaths; it’s the 4 million injuries. People walk away with lifetime permanent injuries.”
At least 31,000 people have been killed each year in U.S. motor vehicle crashes since the end of World War II, according to the National Safety Council. “The conversation should be built on that baseline,” Wolf said. “People just end up jumping to the question: Is the technology safe or not? I think we know the technology will make the roads safer.
“While we address the important questions that come with new technology, particularly in the transportation sector,” he continued, “we need to look at what it is we are living with today. And then ask ourselves — given all these questions and all these important issues that we’re raising — will the deployment of this technology make us safer, more mobile, more efficient going forward? I think the answer is a clear ‘yes’ when it’s framed that way.”
Photo: Kodiak Robotics
‘No longer conceptual’
Right now, AVs are in the “crawl, walk, run approach,” said Richard Beyer, vice president of engineering and R&D at Bendix Commercial Vehicle Systems. “You’re starting with the easier types of traffic like divided highways in areas where you don’t have too much crazy weather.”
This is why the Southwestern U.S. was fertile ground for nascent commercial AV testing last decade. But the testing of self-driving commercial and passenger vehicles has expanded into the more temperate climates of the Midwest and East Coast in recent years. On public roads, divided highways offer the most potential for commercial vehicles. And it’s not just testing.
“It’s certainly no longer conceptual,” said Daniel Goff, head of policy for Kodiak Robotics, an autonomous technology company focused on long-haul trucking. “We deliver freight every day for customers — real paying customers. That’s really exciting. People don’t necessarily realize that their latest iPhone might have been delivered by an autonomous truck.”
Those self-driving Kodiak trucks still have a human behind the wheel to monitor the vehicle, just in case. But they are setting the foundation for America’s driverless future.
Unlike a self-driving passenger car, autonomous trucks don’t need to be capable of handling highways and surface streets. “Our model is a highway-focus model,” Goff told FleetOwner. “Our view of the technology is that you need to optimize for either highway driving or surface street driving. They are just very different applications.”
One of the most significant risks on surface streets for AI-driven vehicles is people doing unexpected things, such as a child chasing a ball into the street. It’s one of the few areas where humans still have an advantage over those AI-powered supercomputers. “People are pretty good at the unexpected,” Goff noted. “Computers and robots have a really hard time with the unexpected.”
Divided highways are a different challenge. Self-driving long-haul trucks are more focused on seeing far enough down the road to be prepared “to stop an 80,000-lb. vehicle if something happens,” Goff said. “Long-range perception is really important.”
That long-range view isn’t just about the lidar, radar, video, and other technologies that power the AI controlling autonomous trucks. Developers and OEMs are counting on AI to help keep supply chains secure as the trucking industry faces a growing driver shortage.
“What’s easy about the commercial vehicle piece of the equation when we talk about automated driving is the benefit is huge for transportation,” Bendix’s Beyer told FleetOwner. “There is already a shortage of drivers today. When you don’t have enough drivers, that means there are loads not being delivered because you can’t deliver.”
Taking away those human burdens — such as hours of service limitations — could be one of the significant benefits of automation, Beyer noted. Even with drivers still in the cabs of AVs, they could rest or do other work as the trucks roll autonomously across the country. “That efficiency gain would be a big benefit to the fleet moving goods,” he added.
“The total cost of ownership and the payback for autonomy is ripe,” Beyer said. “There is a big payback for commercial vehicles. Finding that sweet spot is what’s driving a lot of development. For the startups that were originally looking at passenger cars and robo-taxis — those are coming too — the payback is longer out.
“With commercial vehicles, if you look at the miles moving goods on divided highways in North America, there are a lot of miles that artificial intelligence could really bolster and help move goods through those corridors. This would have a significant payback even though you can’t cover every single street in the U.S.,” Beyer added.
A dashcam video on Kodiak’s website shows entirely human-disengaged runs back and forth between Dallas and Houston on Interstate 45. The video includes all 829 miles of driving over 13 hours and 11 minutes at 10-times speed and shows the Kodiak truck hanging out in the right lane of I-45 unless it has to move over for traffic entering the highway.
Kodiak and other self-driving trucking companies see a depot-to-depot model emerging first.
“We actually envision partnering with various types of infrastructure providers for what we call ‘transfer hubs’ or ‘truck ports,’” Goff said, adding that is a fancy way to describe near-highway parking lots where a human driver would pick up a load from a distribution center, drive it to a truck port and switch the trailer out with an autonomous truck, and send the cargo on its way. The truck would drive autonomously along a divided highway to the truck port closest to its eventual destination, where another human driver would meet it and finish the delivery on surface roads.
This model for trucking is “something we are all going to see pretty soon,” Wolf, of the Self-Driving Coalition, believes. “It’s one of those straightforward use cases that has a lot going for it.”
From testing to reality
But how does that divided highway, truck port model scale up from its current testing to truly driverless trucks?
“To start with, we want to see it grow quickly, but it’s going to take some time,” Wolf of the Self-Driving Coalition said. “Secondly, there are going to be for some period of time some level of monitors and technicians in the cabs.”
Bendix’s Beyer believes the industry is on pace for truly driverless trucks on U.S. highways by 2030. “It’s a larger subject because the vehicle itself needs to change pretty significantly from an architecture perspective,” he said. “You have the AI players out there preparing to do the driving function for human drivers through the computer, or artificial intelligence. But for the rest of the vehicle itself, the architecture has to be able to handle not only the inputs — doing that is fairly easy — but have the redundancy that’s required to keep the vehicle moving, even when you have some single-point failures.”
He explained that a truck with no human inside to control it can’t suffer a failure of any kind in the middle of a highway. “You have to be able to finish the job to get the vehicle to a safe location, typically, to an exit, to a rest area, to the next depot,” Beyer said. “Depending on where you are on the highway, that could be minutes. That could be a half-hour. So, what you need to have is an architecture for the vehicle, especially for the safety systems like steering and braking. You have to have those fully operational, even in backup mode, to get the vehicle into a safe position where you’re not affecting highway safety because you just parked in the middle of the road. That work is ongoing.”
Wolf said it’s not just about redundancies but data based on the current AV testing going on across the country. “It’s going to come down to the data generated by the testing and limited deployment that is underway right now,” he explained. “Does that data warrant regulators to step in and say redundancies are needed? It really has to be data-driven.”
Wolf said there are countless hypothetical issues to consider, but what’s most important is safety records. “Safety is the key message in the AV truck space,” he said. “The public shouldn’t take a company’s word for it. Let the data show how safe it is.”
Wolf added that comes with “a regulatory environment that embraces the technology and prioritizes safety. Those two things are not opposite each other; those two things go hand-in-hand.”
“At least for the near term to mid-term, you’re still going to have somebody in the driver’s seat for most of these vehicles,” Beyer said, “whether they’re called a safety driver or something else. Then, at some point, you’re going to see those vehicles without people inside them — for at least the corridor driving when the technology is ripe and has proven itself.”
This would also be coupled with more public understanding and “societal acceptance of the technology,” Beyer said.
That acceptance, Wolf anticipates, will come more quickly than some might expect as the U.S. public gets more used to AI in its everyday life over the rest of this decade.
“But remember,” Beyer stressed, “aircraft have been autonomous for over 30 years. They can basically land and take off on their own. But every plane I’ve ever got on — even in the last few years — has a pilot and a co-pilot. And the reason they still have a pilot and a co-pilot is that there are still conditions that occur during a flight that an automated system isn’t designed to handle — those real weird anomalies. Then, you’re really happy you have a pilot and a co-pilot in there. For 95% or 97% of the flight, you probably don’t really need a pilot or co-pilot. But it’s because of regulation, and it’s because of society acceptance and things like that this still makes sense.”
Kodiak’s Goff said he doesn’t want to make a firm prediction of when AI-powered tractors like those his company are developing will be running without safety drivers on roads such as I-45 in Texas, but he said it would happen before 2030.
“This is a technology that is coming — at least in certain lanes,” he said. “The first driverless trucks are going to be on highways sooner than a lot of people think.”
How does a fleet prepare?
Wolf suggested fleets interested in getting into or just preparing for the AV future to follow the regulations. He said the Federal Motor Carrier Safety Administration is “really going to be front and center in a lot of this” in the coming months.
The Self-Driving Coalition leader said interested fleets could see this technology advance quickly this decade. And it’s not too early to start talking with AI trucking developers and traditional OEMs about the self-driving future.
“The technology is definitely coming,” Bendix’s Beyer said. “Just like the battery-electric and fuel-cell type vehicles are coming because of environmental impacts, automation is coming. “We’re in an interesting time with commercial vehicles. The technology changes that are happening in the next five to 10 years is probably more change than we’ve seen in the last 30 or 40 years — in a much-shrunken timeframe.”
“I keep telling engineers around here that I’d love to be 18 again and starting a career because it’s a very exciting time for the industry. But it is very, very complex, and that’s why you have some of these AI players in here,” Beyer said.
“When you look at these driver replacements, these are not simple computers. These are very high-end computers that have their own cooling systems because they’re running so hot and have so much data processing — they’re doing video processing, lidar processing — to do these corridor evaluations to understand how to maneuver the vehicle. It’s not simple to replace a driver.”
While it’s not a simple road ahead for truly driverless trucks, there’s a lot of traffic on the path to autonomous Level 5 driving that could transform freight movement on U.S. roads this decade.