CES 2025: Deere Advances Off-Highway Autonomy

John Deere revealed in Las Vegas its second-generation tech stack and four new autonomous machines.

Open-field autonomous demonstration at Gilroy, California, test site employing a 9RX 640 tractor. (SAE/Ryan Gehm)

With the pungent smell of garlic in the air and the threat of tremors from three major fault lines, a select group of journalists invited to John Deere’s Gilroy, California, test center in mid-November got to experience the company’s new autonomous technology and machines in advance of their public reveal at CES 2025 in Las Vegas.

The “cross-eyed” orientation of the four cameras in this pod enables triple overlap. (SAE/Ryan Gehm)

Deere experts are adamant that autonomy can help solve challenges in four distinct areas: in the field, in the orchard, in the quarry and on the grounds. An increase in global food demand and the need for more infrastructure – coupled with significant worker shortages in agriculture, construction and professional landscaping – are driving autonomy’s growth in these off-highway sectors.

The latest autonomous machines – a tractor for large-scale ag operations, a tractor for high-value crops, an articulated dump truck (ADT) and a battery-electric mower – leverage much of the technology revealed three years ago when Deere introduced its first autonomous tractor and started to iterate on it since then, Jahmy Hindman, senior vice president and chief technology officer for John Deere, said at the CES pre-briefing in Gilroy.

Deere’s autonomous 5ML low-profile utility tractor adds lidar sensors to counter unique challenges found in orchards. (SAE/Ryan Gehm)

“This is a full autonomous solution – it’s not augmented, if you want to think about it that way, with a teleoperator,” Hindman said. “There is human engagement with the machine, there has to be. It happens in the form of an operation center through a mobile device today. That’s important for customers to be able to see what the machine’s doing, understand how much productivity they’re getting from it, modify the settings on it if they want, and understand from a tendering perspective if that machine needs something like fuel, for example. So there is a digital umbilical cord between a customer and their machine, but the machine is doing the job without human intervention.”

Deere presented the machines in order of maturity. The most mature technology is incorporated on the large tractor tillage solution, Hindman said, and the least mature – “the one we’ve been working on the shortest period of time” – is the machine for commercial mowing.

New tech stack for open-field ag

The articulated dump truck is Deere’s first autonomous entry in the construction sector. (SAE/Ryan Gehm)

Deere’s autonomous tractor for open-field ag operations uses a second-generation autonomy kit. Why did Deere and its Santa Clara, California-based subsidiary Blue River Technology decide to engineer a new tech stack? Because the initial solution was designed simply to prove that removing an operator from the cab was technically feasible.

“We made tradeoffs to just prove that this worked, to run a machine safely and efficiently in a customer’s field,” said Willy Pell, CEO of Blue River Technology. “There were limitations that would prevent us from scaling to more operations.”

Stereo cameras in the first tech stack operated “in a rigid way” to maintain orientation across vibration, shock and temperature changes, Pell explained. “What happens is as you try to get accurate depth at longer ranges what you want to do is make those cameras go further apart, [but] eventually you run into mechanical limitations.”

The solution was to employ camera arrays, or pods, placed atop the cab that offer triple overlap, and they are not rigidly mounted relative to one other, Pell said. “You can solve for the positions and orientations in real time. Every single time we take a picture, we’re solving for those tiny angular errors,” he said. “What we’ve done is we’ve taken mechanical intelligence and turned it into software intelligence.”

Customer testing for Deere’s battery-electric autonomous mower begins in early 2025. (SAE/Ryan Gehm)

Ease of repair is another benefit of the new pod arrangement. In the first generation, stereo cameras were placed out front on the counterbalance weight bracket, as well as at the rear of the tractor. “One of the problems [with that front location] is that you have to run this wire all the way through the hood. It takes 16 hours to go and plumb that line through,” Pell said. “If your wire wore out or broke, that would be a long repair time whereas [the new kit] has a very short repair time.”

Asked if lidar was a consideration for this large-ag application, Pell said, “One of the things we want to do is have this technology be adopted quickly. If you can pull off a problem with just cameras, it’s much easier to have those cameras roll out of the factory. It’s cost-effective.”

Sixteen individual cameras (compared to 12 cameras in Gen1) arranged in four pods provide a 360-degree view around the tractor. Two Nvidia Orin purpose-built processors power Deere’s VPUs (vision processing units), the core of what will “power all autonomy for John Deere,” Pell said. Two VPUs that are passively cooled are used across the 16 cameras in the large-ag use case. Leveraged for tillage first, Deere’s goal is to have a fully autonomous farming system for corn and soybeans in the U.S. by 2030.

“We needed to be able to see further so we can go faster. There are a lot of operations you do in agriculture that go a lot faster than fall tillage, so we have a 40-percent speed increase,” Pell said, noting a 12-mph (19-km/h) top speed. “We had to create something that would generalize across all the different implements that farmers pull. We don’t want to just automate a single job; we want to automate all the major jobs that tractors do. This is why we invested in this.”

The unit can see out to 25 m (82 ft) and beyond to cover both wide implements and sufficient stopping distance for obstacles. Telematics modems are factory-fitted to the tractors, which connect via 4G to Deere’s cloud service. For rural areas that do not have sufficient terrestrial connectivity, Deere offers Starlink as an add-on that mounts to the top of the cab. The system automatically selects whichever signal is stronger at the time.

The Gen2 kit is retrofittable to tractors back to model year 2020, taking about eight hours to do a full install. For a model year 2025 tractor that already includes all the wiring harnesses, it’s an easier and less time-consuming process to just bolt on and plug in the four camera arrays on the roof.

“We’ve been moving the chess pieces for some years now where all the mechanical connections, all the electrical connections and all the communication connections have been built into the tractor cab awaiting this change,” Pell said. “Tractors coming off the line right now are autonomy ready.”

The Gen2 kits have already accumulated more than 50,000 autonomous acres with paid pilot customers, said Aaron Wells, director of engineering at Blue River Technology. He said that 35 Gen2-equipped large tractors were operated by about 15 customers for the first season. A “limited launch” for production customers starts in 2025, he said.

Unique needs in the orchard

Tractors that help to protect high-value crops in orchards can perform repetitive spraying up to 10 hours a day at 2.5 mph (4.0 km/h), six to eight times during a growing season. Because of the canopy density in orchards, traditional GPS systems can struggle with accuracy. The trees also impact the orchard autonomous kit’s technology configuration.

Deere’s autonomous 5ML low-profile utility tractor for orchards leverages both cameras and lidar sensors. The three pods are placed lower to avoid contact with branches. The front pod incorporates a lidar sensor and three cameras; the side pods each feature lidar and two cameras. A rate controller is also key to autonomy in the orchard, reading the tractor speed and controlling the rate of spray.

“There’s a lot of similarities between the autonomous tillage solution and the autonomous orchard solution,” said Igino Cafiero, director of high-value crop autonomy at Deere. “They’re both built on the same fundamental John Deere second-generation kit, so it’s the same compute, those VPUs, same cameras, same user interfaces that you’ll see mostly between them.”

The addition of lidar is the main difference. “The lidars allow us to see depth…and then define our path in real time. It isn’t a preprogrammed path – it can make micro-adjustments as we go,” Cafiero said. “What this allows us to do is drive down that orchard row, know where the trees are, and we can see what is or isn’t an obstacle, which is a lot trickier in this dense, sort of chaotic environment.”

Ouster provides the lidar sensors on the tractor. “The value the machine is providing to the growers is materially more than the cost of the sensors in this application,” Cafiero said.

Ag data feeds construction, landscaping apps

The articulated dump truck is Deere’s first autonomous entry in the construction sector. Targeting quarry applications, the ADT also shares technology with the ag autonomy system. The autonomous stack mounted to the roof includes six stereo cameras (12 total) and two VPUs, along with two StarFire receivers. The unit provides 360-degree coverage.

Autonomous ADTs are currently operating with customers but are still in the development stage. “Our intention is to have this autonomous ready, so it will come off the factory line [in Davenport, Iowa] ready to be outfitted with the autonomy stack,” said Kasey Kelly, product manager for articulated dump trucks and scraper systems at Deere. “We continue to bring out new models of trucks on a regular development cycle, so we’re matching that up with the development of the autonomous solution.”

The zero-turn-radius mower for commercial landscaping is the other new application employing elements of Deere’s Gen2 autonomy stack. The battery-electric mower, which has 21.4 kWh of energy on board, uses eight cameras, one VPU and two GPS receivers. It can operate up to 10 hours per cycle and recharge in six to eight hours on a 240V outlet. It’s capable of operating in a stand-on manual mode in addition to the fully autonomous mode. Customer testing begins in 2025.

Safety is always the utmost priority for Deere, but that’s especially so for autonomy in commercial landscaping where interactions with people will be more prevalent, Hindman acknowledged. “We’re still working through safety requirements for this. Yes, it will be higher [safety standard] than the other ones we’ve worked on for that very reason,” he said.

How will that be achieved? “More data,” said Brad Powers, director of engineering at Blue River Technology. “This is an AI system or machine-learning-based system, so we train it more to be able to handle the higher incidence rate of human [interactions].” And Deere has no shortage of data.

“We’ve made a tech stack that can meet the needs [across industries] from the autonomy side, the machine learning models, even data,” Pell said. “Data from a tractor are being used to train models on a dump truck and on a mower. This is a big advantage. Collecting this data, labeling it, finding all these edge cases, this is hard work. It wouldn’t be that we would ship an autonomous mower using just agriculture data – you fine-tune it at the end. But we have this base of data that allows us to get into these new markets safely and effectively, really quickly.”