CV Autonomy: Picking the 'Low-Hanging Fruit'

Increasingly automated commercial vehicles offer challenges and opportunities in simulation, data management and standardization.

John Deere recently acquired Silicon Valley-based startup Bear Flag Robotics to accelerate automation on farms. (Bear Flag Robotics)

Functional safety is a major reason why autonomous-vehicle (AV) development timelines have been pushed back, and why SAE “Level 2-plus” driving automation has become more of a focus for many companies in the near term. “There still is some focus on Level 4 for the future, but it’s basically on the backburner,” said Suga Ragunathan, director at KPIT, a software development and integration company that assists OEMs and suppliers with advanced driver-assistance system (ADAS) and AV validation.

TuSimple has received 6,775 reservations for SAE Level-4 International LT Series autonomous trucks in partnership with Navistar. Production is expected to begin in 2024. (TuSimple)

Do industry experts realistically envision the deployment of fully-automated Class 8 trucks? This question was posed at the SAE COMVEC 2021 conference. The consensus: “Yes, but…” Anupam Gangopadhyay, engineering director for controls and software at Navistar, summed it up: “It’s all in the ODD [operational design domain].”

Navistar does not yet offer full SAE L2 capability, Gangopadhyay said. Lane-centering, which proactively keeps the vehicle centered within its lane, is not yet available on its trucks but “that’s obviously coming. It’s in the works,” he said. “As we go to L4, for us it’s going to be a truck driving on a freeway that is manageable without touching areas where our ODD cannot support it.”

Stephan Olsen, general manager at PACCAR Innovation Center, agreed. “It’s not going to be a light-switch event,” he said. “It’s going to start in the well-understood ODD [scenarios] – I-10 [which runs from Santa Monica, Calif. to Jacksonville, Florida], Texas, New Mexico and Arizona where there’s good weather, defined A-to-B points. As the technology matures, it’ll grow from there.”

Oshkosh will manufacture the U.S. Postal Service’s next generation of postal delivery vehicles, which eventually could benefit from increased levels of autonomy. (USPS)
FedEx has teamed with Aurora and PACCAR to actively test autonomous operations on parcel routes in Texas between Dallas and Houston. (Aurora)

The “middle mile” ODD is not necessarily “low-hanging fruit,” but it is “the next fruit to be picked” in L4 on-highway autonomy, according to Olsen. “Today, there’s a driver shortage – that’s one of the top three issues that any fleet operator will tell you is a challenge,” he said. “If autonomy can solve that middle mile, that leaves the local routes for the human drivers where they can get home at night and see their families.”

Automation in off-highway applications, particularly in the military, can pose some unique challenges not encountered on roadways. “When you’re in an off-highway environment, you may not know what your boundary conditions are, you may not know what obstacle detection you’re looking at, depending upon the decisions that have to be made,” said Michael Puroll, chief engineer – system architect at AM General. “With military, it gets even more difficult. Sometimes decisions have to be made about, do you go through an obstacle? Is it a friend-or-foe situation? You don’t necessarily want to stop an automated vehicle in all conditions. So, the what-if scenarios become much greater in off-road situations.”

Parsing terabytes of data, millions of scenarios

Commercial trucks can collect more than 10 TB of data per day, some experts claim. “That’s a lot of data to store,” KPIT’s Ragunathan said. “We cannot use all of the data, so it is really important to understand what we want to collect, what conditions that we want to focus on.”

Bear Flag Robotics, an agriculture technology startup based in Silicon Valley, stores data on-board and pulls hard drives every week. “By acre, 2 terabytes in a week are a lot of data and we cover a lot of ground,” said Daniel Carmichael, farming operations manager for the company, which was recently acquired by John Deere to accelerate automation on farms. The companies have outfitted a Deere 8RT tractor with Bear Flag’s perception kit, which includes an inertial measurement unit (IMU), a three-computer compute system and a mobile control interface.

A 5G cellular infrastructure could help with data transfer, but likely not for some time in agriculture. “I think it’s going to be awhile before we get that kind of connectivity on the farms in some of these rural areas,” Carmichael said. “On the farming side, it’s not quite as critical to get that data real-time as it would be in a combat situation,” said Clayton Janasek, senior staff engineer at John Deere. “With that in mind, things like WiFi – if you can get a mobile hotspot, you can get the data off that way. There’s always a USB option. Looking forward, satellite connectivity for the data is an intriguing idea as well.”

Ben Duprey, senior modeling and simulation engineer at Mechanical Simulation Corp., said the goal should be to learn from other players in the environment and then make decisions based on that. “Long-term, I shouldn’t have to worry about constantly acquiring data just in case,” he said. “At some point, if that data – and more specifically, if that information – becomes outdated in some way, I no longer have to maintain it. I’m more interested in making sure that my decision-making is based on the most current information and then maybe I supplement it with some historical cases.”

Standardizing use cases to help speed up AV development is a possibility, but many experts believe that at this stage, standardization would stifle important discoveries. “Standardization will certainly help create a level playing field, it’ll put us all on a common understanding of what’s required, but it’s also important to recognize that we’re still in the early days of developing these systems,” PACCAR’s Olsen said. “I’m looking at this through the lens of a developer and the work we do with our developers – there’s a lot of exploration yet to be unearthed that could create opportunities in how we set regulations. Locking down at this time is probably premature and might prevent some opportunities to be discovered.”

“We don’t want the regulation to be driving the technology. We want technology to have some room to grow,” said Brendan Chen, chief engineer of unmanned systems at Oshkosh. “The role of government is to make sure that there is a minimum level of expected functionality; regulations are meant for providing that safeguard.”

Partnering to problem solve

Collaboration and partnerships are another strategy increasingly being employed to accelerate AV development. “Just the sheer complexity of the problem leads to the fact that none of the OEMs can solve it alone,” said Peter Waeltermann, president of dSpace and moderator of a COMVEC session on validating ADAS and AVs.

PACCAR’s “go-to-market plan” is to collaborate with autonomy developer Aurora. “What we concluded in making that decision is that we want to leverage what we do best – building and validating trucks that are going to perform well,” Olsen said. “Developing autonomous driving stacks is not what we do best.”

Truck OEMs do not want to invest millions of dollars re-creating what many self-driving technology companies have spent the past 10 or 15 years working on. “For quite some time, perception will be the playground for other players who have been working on it for [many years] and have developed their skills and knowledge,” Navistar’s Gangopadhyay said. “Everything else in the vehicle, even the motion control, and of course, body control, brake systems, those are traditional vehicle systems. Once you draw the line at motion control, everything beyond that – perception and maybe planning – is where partners will play a big role.”

Clear and constant communication between all the different parties is extremely important, Chen said. “The perception system doesn’t fail on its own; the truck doesn’t fail on its own,” he said. “Failures occur at the interfaces typically, where if you don’t have a clearly defined interface control document, or expectations are not set correctly, that’s when the ball is dropped.”