Scientists in Denmark are developing an agricultural robot for identifying and eliminating weeds. While this might seem like a relatively easy task, it actually requires a lot of machine intelligence to pick out the weeds among the crops. The robot is still in the early stages of development, but the researchers hope that it will ultimately lead to a reduction in the amount of herbicides used by farmers and therefore cut costs.
Called Hortibot, the semi-autonomous robot is a navigational platform designed to have different agricultural tools fitted to it to either mechanically remove weeds or precision-spray them with herbicide. “The original purpose was to build a robot that was simple to use and could be operated by an unskilled worker,” says Rasmus Jørgensen, an agricultural scientist at the Institute of Agricultural Engineering at Aarhus University, in Horsens, Denmark.
Currently, farmers tend to deal with weeds by either spraying entire fields or by using manual laborers to physically remove weeds by hand. But these approaches have their problems, says Jørgensen.
Although labor can be cheap, the cost of training workers can mount up. The high turnover of low-skilled workers in agriculture means that farmers often have to pay to train as much as one-third of their work force each year.
Watch the Hortibot in action.
Indiscriminate spraying of herbicides has an environmental impact, is wasteful, and adds to the cost of farming. And herbicides are normally sprayed using heavy vehicles such as tractors, which in turn cause compaction of the soil. “If the soil is too compact, the roots can’t penetrate it, and water can’t get through,” says Jørgensen. A single tractor and its load can weigh enough to compact soil by as much as half a meter, he says.
The aim with Hortibot is to address these issues by enabling a lightweight robot to carry out the same tasks under the supervision of a single worker with little training. Weighing just 245 kilograms–roughly one-fortieth as much as a tractor–the robot is based on a modified framework of a commercially available remote-controlled slope mower called a Spider.
Standing a meter tall and roughly a meter and a half wide and long, the four-wheeled Hortibot comes equipped with a downward-looking camera. This enables the robot to navigate autonomously in between several rows of crops without damaging them and without the use of any global-positioning technology. Some agricultural machines now use GPS, but it can be unreliable since it depends on the resolution and accuracy of maps, says Jørgensen.
Using a vision-based approach ensures that the robot covers the field more accurately, turning when it reaches the edge of a field to continue winding its way across the entire plot. The human operator is there to guide it to the field and stop it if obstacles emerge. With less than an hour’s training and using a simple control stick, anyone can use it, Jørgensen says.
At a recent Field Robot Event, held in Wageningen, in the Netherlands, Hortibot was able to follow furrows and autonomously turn in the appropriate direction when it reached the edge of the crop rows. While some of the other robots were able to follow crop lines, they were unable to turn. “We have shown that [Hortibot] is easy to work and can make turns without a lot of planning,” says Jørgensen.
The Hortibot team is now planning to equip the robot with modular tools for precision spraying and mechanical weed removal. The aim here is to first identify the weeds, and then either spray them individually without hitting the crops or remove them without damaging nearby crops.
“In recent years, there has been an increasing interest in the use of mechanical intra-row weeders because of concern over environmental degradation and a growing demand for organically produced food,” says Athanasios Dedousis, an agricultural engineer who has been developing such a system at Cranfield University, in the United Kingdom.
But getting it all working is likely to be a huge challenge, says Chris Melhuish, director of the Bristol Robotics Laboratory, in the United Kingdom. “To get a machine to do that is extremely difficult,” he says.
Jørgensen agrees. There has been a lot of research using hyperspectral analysis to try to identify unique signatures of different plants based on the frequencies of light they absorb, he says, but it turns out that their absorption depends very much on the nutritional health of the plant.
So Jørgensen’s lab is taking a different approach. “We use shape analysis,” he says. By analyzing the proximity, orientation, and position of leaves relative to each other, it’s possible to distinguish weeds from crop plants. Difficulties arise when the leaves from different plants overlap or obscure each other, but Jørgensen believes it’s an approach that will eventually prove more reliable than spectral analysis.
Similarly, the mechanical removal of plants is far from straightforward, says Dedousis. Crop rows are not always straight, and plants are not equally spaced. And because some areas can have high levels of weed infestation, the mechanisms for removing the weeds need to be accurate down to about 50 millimeters, Dedousis says.
But the significant challenge is likely to come down to economics, says Melhuish. “At the moment, there is no way [the robot] would be cost-effective,” he says. The estimated cost of Hortibot is about $55,000. With migrant workers in both Europe and the United States willing to work for low wages, it’s difficult to see how the robot could compete, he says.
Hortibot’s cost could be offset by its productivity, says Dedousis. Weeding is repetitive work. Robots can do it round-the-clock and in all weather conditions. For organic farmers, this could be a real boon. “Hortibot is the way towards the future,” Dedousis says. But he predicts that it will take some years before it’s ready for commercial farming.
Indeed, Jørgensen says that before the robot can be commercialized, some fundamental redesigns are necessary. These include increasing the wheel gauge to enable Hortibot to cover more rows of crops with each sweep and increasing the clearance to make the robot suitable for a wide range of crops.
DeepMind’s cofounder: Generative AI is just a phase. What’s next is interactive AI.
“This is a profound moment in the history of technology,” says Mustafa Suleyman.
What to know about this autumn’s covid vaccines
New variants will pose a challenge, but early signs suggest the shots will still boost antibody responses.
Human-plus-AI solutions mitigate security threats
With the right human oversight, emerging technologies like artificial intelligence can help keep business and customer data secure
Next slide, please: A brief history of the corporate presentation
From million-dollar slide shows to Steve Jobs’s introduction of the iPhone, a bit of show business never hurt plain old business.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.