Between 2009 and 2015, Google spent $1.1 billion on autonomous vehicles for its Waymo subsidiary, which have so far roamed more than 10 million miles of city streets — though none of those miles have yet involved a paying customer. Tesla, which builds the data collection it uses to improve its autonomous technology into every vehicle it sells, lost more than $1 billion as a company last year alone. In pursuit of viable self-driving cars, the companies have had to navigate a web of regulation that varies from state to state, apply for permits, and risk getting banned from roads if they fail to follow the rules. Neither of their self-driving technologies is ready to be set loose on public streets without a human safety driver behind the wheel or navigating previously approved routes.
But Zack James faced few barriers when he started his autonomous tractor company in 2017. His roughly 200-pound tractors are closer in size to a riding lawn mower (and about half the weight) than they are to traditional combines and sprayers, and the open metal frame vehicles work together in a swarm. After coming up with the concept while in law school at the University of Michigan, it took James about a month to fabricate the first prototype, which he soon tested on a family field in Crown Point, Indiana, standing by in case the tractor encountered an obstacle it couldn’t yet navigate.
Instead of sleek Teslas or robot Ubers, the first truly driverless vehicles are more likely to look like James’ tractors: rolling placidly over a cornfield at a max speed of 7 mph.
What James did not do before building and testing his prototype in the real world was raise a billion dollars. Or apply for a permit. Or consult ethicists about how to handle split-second decisions about whose life to save in case of an impending accident, otherwise known as the “trolley problem.” And yet — so far, so good. “I crashed one into a tree one time,” James says. “But both were fine: Nothing happened to the vehicle or the tree.”
Silicon Valley entrepreneurs have dazzled stockholders and consumers with visions of fully autonomous vehicles that can wander out to work as robotaxis when not needed by their owners. But this version of a no-humans-required future of transportation is likely still years, perhaps even decades away. Instead of sleek Teslas or robot Ubers, the first truly driverless vehicles are more likely to look like James’ tractors: rolling placidly over a cornfield at a max speed of 7 mph.
The early version of James’ tractor, called the Rabbit, disperses seeds for cover crops, which farmers plant in between their cash crops to improve soil health. Other early versions of small autonomous farm robots focus on dispensing fertilizer, mowing and slashing, and weeding. Often they can be outfitted with different attachments so that the same tractors perform different tasks.
These small autonomous tractors won’t revolutionize transportation, but they do aim to solve an important problem in agriculture. The typical giant farming tractor treats an entire field the same way, spreading pesticide or fertilizer on every inch of soil whether or not that inch benefits from the chemicals or not. The tractors can cost as much as $600,000, making them impractical for small farms and an expensive asset that sits idle for long stretches on larger operations. The hope is that swarms of small, autonomous tractors could alleviate labor and fuel costs for farms and enable sustainable practices that are currently considered impractical, such as “precision agriculture,” in which farmers diagnose and cultivate plants individually, spraying pesticides only on weeds and allocating fertilizer based on how much a plant needs.
Developing autonomous tractors isn’t simple — farm fields are unstructured, uneven environments that are arguably harder to navigate than paved roads — but several factors make it easier than developing autonomous cars. Most of the tractors’ speeds top out at about 10 mph, giving them much more time to process data and react than a car traveling at freeway velocities. A tractor’s tasks are also much narrower than a vehicle ferrying human beings. “‘Harvest the field’ or ‘find which crop is sick,’” says Girish Chowdhary, the director of the Field Robotics Engineering and Sciences Hub at the University of Illinois, Urbana-Champaign. “This is a more well-defined task than ‘take my kids to school.’” Farm fields, unlike roads, also do not typically contain unpredictable pedestrians or human drivers.
Meanwhile, even among those who are developing autonomous tractors, it’s not clear who, if anyone, has created standards or regulations for autonomy on farm fields. Larger autonomous tractors may fall under the same regulations as autonomous cars when they move from field to field using the road, but smaller tractors can be transported in a trailer. The International Organization for Standardization, a nongovernmental body that publishes standards on everything from occupational health and safety to medical devices, has guidelines for tractors with auto-guidance features but not for those with no driver. When I asked a spokesperson at the U.S. Department of Transportation, which publishes a set of voluntary guidelines for autonomous vehicles on roads, if the agency had regulations for autonomous vehicles on farms, they referred me to the Department of Agriculture, which did not return my request for comment. The U.S. Occupational Safety and Health Administration (OSHA) confirmed that it does not have regulations regarding autonomous tractors, though state OSHA agencies may make them. One state, California, has chosen to do so, requiring them to have an operator with access to “vehicular controls.”
Chowdhary, who is also the cofounder of a startup called EarthSense that sells an autonomous agriculture robot he developed to phenotype plants for breeders, says he’s heard some people refer to autonomous tractors as “autonomous cars without the liability.”
In states like Florida and Arizona, autonomous carmakers enjoy nearly as lax regulations, but the risk of putting a tractor in an empty field while it’s still in development pales in comparison to the risk of putting autonomous technology on the road before it’s ready. Last month, the U.S. National Highway Traffic Safety Administration opened an investigation into the 12th accident that may have involved a Tesla using its autopilot feature, which can summon a car from a garage and automatically change lanes on a freeway. At least three of those 12 accidents have been fatal. “If I needed to,” James says of his Rabbit tractor, “I could grab it and flip it over.” (Larger-scale autonomous tractors or those that work alongside humans for, say, fruit-picking, pose more safety questions.)
“If a little baby comes in the corn field is it going to stop? Probably not. But the point is that the chance of that happening is low.”
The freedom to send autonomous tractors over farm fields without much, if any, regulatory interference or risk of injuring humans helps speed the development of better technology. “The challenge with agriculture is that the environment can be very different in different fields,” Chowdhary says. “Hence, the only way to ensure that the robot autonomy is robust is to expose the algorithms to many different situations.” U.K.-based Small Robot Company currently tests its small autonomous agriculture robots at 20 farms throughout the country. “The problems you encounter vary from farm to farm,” says the company’s chief marketing officer, Sarra Mander. “[In] some farms there may be a road running through the middle. Most have a footpath. We needed to get a real-world idea of what the problems are. It’s much easier to encounter a problem early on than try to retrofit it later.”
In 2019, Small Robot Company began offering weed mapping to selected farms for a trial. By collecting data on where weeds were sprouting, the robot — which Small Robot Company named Tom — helps farmers make decisions about where to treat the field in the following year or whether to spray for certain types of pesticides. Later the same year, Small Robot Company introduced a service that will identify weeds and kill them without pesticides by literally zapping them. It’s a start, but the tractors still require someone to be watching them work to handle glitches, and there’s years to go before the company’s tiny fleets of tractors fulfill the vision of a field that basically runs on autopilot, with robots handling the weeding, spot applications of fungicides and fertilizers, and strategic planting.
It’s also not the case that autonomous tractors are completely without risk, which is why firms like Small Robots Company have begun participating in groups creating industry standards for how the robots should interact with people. Standards created by an industry group are a common route to the regulation of emerging technology and one that is perhaps more appropriate for a 200-pound robot rolling at walking pace through an open field than a 3,000- or 4,000-pound car driving fast enough to be a lethal weapon. “If a little baby comes in the cornfield is it going to stop? Probably not,” says Chowdhary. “But the point is that the chance of that happening is low.”
Chowdhary says he’d define his phenotyping tractor’s autonomy at level 1, according to his own roadmap, which he based on the level of human involvement required to get the job done. (“It doesn’t cook breakfast,” he says, “but it can go through the rows of corn by itself.”) Next year, he hopes to achieve level two, when a single human will be able to supervise many robots. What he would consider “fully autonomous,” a level 4 on his scale, is that the robots would maintain a field on autopilot throughout a whole growing season.
“I wouldn’t say there is full autonomy,” he says of swarm farm robots. “but I would say that the path to full autonomy is unhindered in cornfields.” Unlike the supposedly open road.