Robotic harvester would be costly
An automated harvester would need to be affordable and robust, as well as technically feasible.
Robotics expert Dr. Sanjiv Singh says fully automated robotic harvesting systems are looking more feasible technologically than a few years ago, but whether they would be robust enough for orchard work and affordable to the growers is still in doubt for the time being.
Singh, a researcher at Carnegie Mellon University’s Robotics Institute, is leading a multistate project on Comprehensive Automation for Specialty Crops, which aims to develop methods to improve production efficiency, identify threats from pests and diseases, and respond to food safety hazards.
What it doesn’t aim to do—at least, not yet—is develop a robotic apple harvester.
Singh said when the CASC project began in 2008, the scientists involved decided not to work on automatic harvesting. It would be such a huge endeavor, they agreed, that it would overwhelm the rest of the project. Not only that, but what would be the use of developing a machine that cost $2 million dollars? Or could only pick one apple every ten seconds (10 times slower than a human)?
“The gee-whiz factor runs cold pretty quickly,” he said. “We could have done it, but we needed it to be all three of these things for it to be a success: affordable, robust, and technically feasible. Those are the reasons I personally think it’s not a short-term or even medium-term solution. It would require some significant funding, and I don’t know if there are deep enough pockets to go after that.”
CASC faculty decided instead to work with DBR Conveyor Concepts in Michigan who are developing a harvest-assist machine that improves harvest efficiency by eliminating the need for workers to climb ladders.
That’s the type of technology that will reach growers quicker, Singh points out. Tests with the DBR last season in Washington State showed that workers were 20 percent more efficient working with the machine than when picking fruit conventionally. Singh said the developers and scientists think it reasonable to expect that the machine could increase worker efficiency by 100 percent within three years, which would mean the grower could get by with half the number of workers.
“That’s something that people can go to the bank with,” he said. “There’s a lot more that can happen to the machine.”
Meanwhile, scientists at Carnegie Mellon are wondering, four years after the CASC project began, whether the time is ripe to begin looking at fully automated harvesting. Tree architecture is changing, with more growers adopting formal training systems where there is less obstruction of fruit by the foliage, and the cost of technology continues to decrease.
“We may start to consider this,” Singh said. “We’ve been doing some experiments in our lab, trying to see how hard it is.”
For example, they’ve done trials with industrial robotic arms—the kind that would survive repeated motions—to find out how precisely they can locate something and reach out and grab it.
“When you watch them, they’re cartoonish,” Singh said. “Right now, they have slow, deliberate motions, and the kinds of grabbers we have are pneumatic and don’t look quite right.”
Whereas a person who’s good at picking might pick an apple per second, these robotic arms might take ten seconds per apple.
But the good news is that the robotics are far cheaper than they used to be. Five years ago, one robotic arm might cost $150,000, which didn’t make economic sense. Now, they are ten times cheaper.
Singh said these are informal experiments to gather preliminary data that could be used in writing a proposal.
“We want to make sure we don’t oversell the technology,” he said. “We want to make sure when we come out there that people think it’s credible. We’ve taken a pretty conservative approach to this.”
Once a project is launched, he thinks it could take three to five years to develop a system with the right accuracy, speed, and price.
One of the important aspects of a harvesting system is the ability to locate the apples on the trees.
Six years ago, Vision Robotics of San Diego, California, began a project to develop a robotic apple harvester. Other manufacturers who had attempted robotic harvesting had run into problems with end effectors that tried to locate fruit as they picked it. Vision Robotics took a two-step approach. One robot would act as a scout, develop a three-dimensional map of the fruit and interfering branches, and work out a picking strategy for a separate machine that would come along later and pick the fruit.
The Washington Tree Fruit Research Commission provided $283,000 initially to Vision Robotics to develop the orchard scout, while the California Citrus Research Board paid a similar amount to the company to develop a citrus scout. Vision Robotics was never able to secure funding to develop the harvesting part of the system, but saw potential for the scout as a crop estimating and yield mapping tool.
When the CASC program began, Vision Robotics became a collaborator. Carnegie Mellon’s Robotics Institute has since taken over the scout project. Singh said they are working on making it more compact and lighter so it would be cheaper and easier to pull with a tractor. They have also integrated the scout with an autonomous orchard vehicle that they’ve developed. The university is looking for a partner to commercialize it.
Tony Koselka, founder and chief operating officer of Vision Robotics, said his company is still interested in developing a robotic harvester, but it would be an expensive project that would be best tackled by a consortium because various scientists around the country are making progress on different aspects of it. It would be a multimillion-dollar project.
He believes it is technologically feasible, and perhaps even cost effective, but he thinks it would be difficult. Manipulating the apples without damaging them is the hardest part. Apples can’t just be plucked from the tree. It’s challenging to have a machine grab apples from a variety of different angles and replicate the way a human lifts and turns them to detach them from the tree.
The project would be easier if all the orchards had formal training systems where all the fruit was visible and hanging vertically, he said. The scout would be 100 percent accurate, and the harvester could have shorter arms and could pick all the fruit from the same orientation.
Having a scout to map the fruit before harvesting might not be necessary in orchards where all the fruit is immediately visible, but a scout would make it more efficient, Koselka said. “The biggest factor is how fast you can get the fruit off the tree. If the scout gets you 25 percent more efficiency, that’s enough to pay for the cost of it.”
Initially, a robotic harvesting system might not save growers money, but it would reduce their labor needs. Koselka said the economics are less favorable in apples than in citrus, however, because citrus is harvested for ten months of the year versus a few weeks for apples. Machines for apple harvesting might need to be multipurpose so they can be used for other jobs such as pruning, he said. Vision Robotics has developed a robotic pruner for grapevines, which should be commercialized within the next two years.