This is part of IEEE Spectrum's SPECIAL REPORT: WINNERS & LOSERS 2009, The Year's Best and Worst of Technology.
Handle with CARE
This robot uses a vacuum arm to pick a ripe strawberry and gently nestle it into a padded container.
In California it’s hard to drive anywhere without spotting people bending down to pick strawberries—some 10 million pint baskets a day during peak season. Because such farming isn’t practical in labor-starved Japan, the government there is backing a more fundamental solution: robots.
Research began in 2003, with funding from the Institute of Agricultural Machinery Bio-oriented Technology Research Advancement Institution, or IAM-BRAIN. The first demonstration of a prototype came in 2005. Yet although the system may well qualify as a technical tour de force, it’s hard to see how it makes economic sense, particularly outside of Japan.
To understand why it is such a fascinating technical challenge, you need to know a bit about the berry business. Strawberries ripen only on the vine, and they do so in dribs and drabs. A robot therefore must patrol rows of plants day after day. When the robot comes across a candidate plant, its machine vision has to do more than just tell red from green, because a strawberry goes through many shades of red before it is fully ripe. Finally, the robot has to pluck and place the berry very carefully; a little too much pressure will bruise the strawberry, causing it to rot. In short, strawberries are not the low-hanging fruit for robotic harvesting technology.
”It’s not what I would have started with; you have to get so many things right,” says Joseph L. Jones, chief technology officer of Harvest Automation, a company in Groton, Mass., that develops robots for the agricultural market. ”It seems you could start with something simpler—a winter squash, perhaps.
”There are no large-scale uses of agricultural robots doing anything, because it’s really, really hard,” Jones continues. ”People working in agriculture use all our considerable talents to identify an object, our considerable manipulation skills to manipulate it, and we work under all conditions. Robots aren’t up to that, particularly in an unstructured environment outdoors.”
Sure, dumb mechanical harvesters have long ripped hard, unripened tomatoes off the vine to ripen and be sorted later on. And machine vision has gotten good enough to sort fruit passing on a conveyor belt. Robots being developed now will soon move potted plants around greenhouses. But such jobs are easier to do in these contexts than in a berry-growing greenhouse, let alone a planted field.
The original prototype looked like a version of the multimedia cart used in high schools: it had wheels on the bottom and several layers of open shelves, with a laptop computer on the top shelf and a tall arm towering above it, much like the arm of an overhead projector that rises above that media cart. The robot cruised planted rows, illuminating berries with four polarized lights and examining them with three color video-graphics-array cameras. While it conceivably could have gone out in a field, it was designed for the greenhouse environment; outdoors, dust could obscure the machine-vision lenses.
The current version is optimized even more for the indoor environment. It runs on rails built into a specially designed greenhouse; a single arm reaches down to suck the fruit against a tube capped by a soft sponge while a clipper snips the stem. The arm, still holding the fruit by suction, then reaches down and gently places the berry into its own hole on a spongy tray.
When cruising the aisles of the greenhouse, the robot picks, on average, one berry every 10 seconds. That’s nowhere near as fast as a human worker. On the other hand, the robot doesn’t take breaks.
The target cost of the hardware components for one robot—estimated at 7 million yen (about US $72 000 at press time)—equals the seasonal wages of three to four workers. That’s not bad—although that’s Japanese workers, of course; the money would pay for far more labor in California, where the price of an individual strawberry runs as low as a dime, one-fifth the average price in Japan.
But the real expense comes in deploying the robot. Farmers would have to rebuild every greenhouse—raising the beds high enough to let the strawberries hang down for easy picking and installing tracks for the robot. It won’t happen without government subsidies, says Kyoto University professor Naoshi Kondo, a machine-vision specialist who is working on the project under contract to SI Seiko. That’s too bad, because robots would offer advantages beyond a mere savings of labor. Above all, robots collect a lot of data.
”We can code which bed a fruit is picked from, in which greenhouse. If a strawberry later has problems, we can trace it back to one spot and can know when it was harvested,” Kondo says. Such a capability would have come in handy during the recent spinach, lettuce, and chili pepper health scares, which led to vast product recalls.
With the expensive requirement of greenhouse redesign as a major hurdle, does the project make any sense at all? Kondo himself says that at the moment, humans are better at picking strawberries because they are faster and more flexible.
”It’s nifty the way it works,” says Tony Grift, associate professor of agricultural and biological engineering at the University of Illinois at Urbana-Champaign, who recently visited Kondo. ”And I think you could market it in Japan. In America, I doubt it. We aren’t going to grow strawberries in greenhouses anytime soon.”
”If you are interested in developing robots, it is a fascinating task,” says John Billingsley, professor of mechatronic engineering at the University of Southern Queensland, in Toowoomba, Australia. ”If you are interested in harvesting strawberries, it is not the way to do it.” A better solution, he says, would be to just raise a lot of strawberries, harvest them all at once with a dumb mechanical device, and then sort out the good ones. ”It would be cheaper to raise more strawberries than it would be to build expensive robots,” he adds.
Billingsley suggests that the choice is similar to one that an engineer might make when designing a dishwasher. You could make a robotic dishwasher, he says, by taking a standard industrial robotic arm, putting it in front of a sink with a dishcloth in its grip, and programming it to do the task. However, that would make little sense because ”it’s so much easier just to bundle the crockery into a square box and squirt water at it,” he says.
Labor shortages do mean that automation is coming to agriculture, says Billingsley, ”but it has got to be the right kind of automation.”
For more articles, go to Winners & Losers 2009 Special Report.
Snapshot: Robotic Reaper
Goal: To develop a robot that can select ripe strawberries and pick them without damaging them.
Why it's a loser: It works only in greenhouses, which would have to be redesigned to accommodate it. Besides, humans can do the job much more easily.
Who: SI Seiko Co., Maekawa Manufacturing Co., AAI Japan, funded by the Institute of Agricultural Machinery’s Bio-oriented Technology Research Advancement Institution (IAM-BRAIN)
Where: Matsuyama City, Japan
Staff: About a dozen
Budget: Not public, estimated at over US $1.5 million to date