Commercial farms of the future may be staffed by robots that will identify, spray and pick individual pieces of produce from plants, even when their targets are grapes, peppers and apples that are as green as the leaves that surround them.
As scientists in Israel and Europe get closer to this goal, experts say the work has a number of potential benefits. Autonomous agricultural robots could protect human workers from the harmful effects of handling chemicals by hand. And through a system of highly selective spraying, robots could reduce a farm's use of pesticides by up to 80 percent.
Robots could also offer a timely supply of labor in many places, where there simply aren't enough itinerant workers available at the right times in the harvesting cycle. Meanwhile, attempts to create robots that can see, grasp and learn could end up having widespread applications in medicine, video games and more.
And while scientists have been working to develop robots for agricultural labor for more than 20 years, a new project is taking a more cerebral approach. The goal is to teach computers to see like humans do and to get better at their jobs as they work and learn.
"The technology is ready, and now we can start seeing this penetrating into the market," said Yael Edan, an engineer and robotics researcher at Ben-Gurion University of the Negev in Israel. "I would say there will definitely be robots out there in five years -- maybe not be on every farm, and maybe not for every farmer. I think now the time is there."
Modern commercial farms are already full of tractors with automated steering and machines that can milk cows and till soil. But zeroing in on individual fruits or vegetables is a much more challenging task. That's because the outdoor environment is unpredictable and ever-changing.
Each piece of produce, for example, has a unique shape, size, color and orientation, which means that a computer can't be programmed to simply search for a specific image. Shadows and light conditions change throughout the day and night, as well, making an individual object look different under various conditions. And green fruits and vegetables can look much like the leafy bushes or vines they grow on.
To boost a computer's ability to find order within the relative chaos of an agricultural environment, Edan's team, along with a European consortium of colleagues, is working on intelligent sensing systems. One strategy involves multi-spectral cameras that analyze wavelengths of light bouncing off of objects. The idea is to find a consistent pattern that would tell the robot when it is seeing, say, a pepper, no matter whether that pepper was right-side-up or upside-down.
Along with other sensors and programs, the researchers aim to create a robotic "brain" that could then learn from its mistakes and improve as it works.
"We will have an algorithm that will see simple shapes. And when food is partially covered by leaves, it will say: 'OK, let's not use the full-shape algorithm. But since we only see part of the food, let's try to complete the contour,'" Edan said. What separates her team's work from previous projects, she said, is that it incorporates both features of human vision and computer learning.
So far, computers can easily find between 80 and 85 percent of fruits on a plant, the group has found. But their benchmark is 90 percent, and many farmers say they wouldn't use a robot unless it hit an accuracy rate of 99 percent.
Once a robot identifies its targets, the engineers are also trying to design a grasping tool that will grab produce in the right place and pick it with the right amount of firmness. To that end, they are studying human movements and using another set of algorithms to try to imitate what comes so naturally to human hands.
As the project, which began last October, ramps up and begins to produce results, agricultural robots could eventually help farmers around the world, including in the United States, said Bernie Engel, an agricultural engineer at Purdue University in West Lafayette, Ind.
"In many cases, there are challenges finding labor to do some of the harvesting of strawberries and other fruits and vegetables," Engel said. "It's hard work. There's a timeliness factor, where you can't wait a week. You need lots of labor for fairly short periods of time, which creates real challenges for keeping people employed in a sustainable manner."
"If you think about the global population at this point and the need to feed a growing population," he added, "we have to get more efficient at the harvesting and production of these crops."