—by Matt Milkovich
Five years ago, when Renfu Lu’s research team tested its in-field sorting machine in Michigan apple orchards, they received feedback from growers, and some suggested that instead of humans feeding apples into the sorter, why not add a robotic harvester and create a fully automated picking and sorting system?
So, Lu, a research leader with the U.S. Department of Agriculture’s Agricultural Research Service, and his Michigan State University collaborators went back to the drawing board to design a more ambitious machine. This fall, they were back at Schwallier’s Country Basket in Sparta to demonstrate their robotic harvester prototype. Within the next five years, they plan to have a commercially viable, fully automated mobile platform that can harvest apples and sort them in the field.
“There are still major technical issues to overcome, but I’m confident we can meet that goal,” Lu said. “Every year will see significant improvements in performance.”
Lu discussed the project’s next steps during the September demonstration. He plans to power the platform by electric battery. He wants each two-armed robot module to reach a picking speed of 2 to 3 seconds per fruit, roughly the equivalent of human pickers.
“Our single-arm robot module has reached a picking speed of 3.5 seconds per fruit, so we feel confident that our two-arm robot module can reach the target picking speed after the system is further optimized,” he said.
The demo harvester was pulled by a tractor. A graduate student used a laptop to adjust the platform’s height and displacement. The team ran the manipulation arms at 60 percent of their maximum speed. MSU mechanical engineering professor Zhaojian Li expects the harvester to be fully automated by next season.
The project received a four-year, $3.5 million grant from the USDA’s Specialty Crop Research Initiative in 2023. Participants include researchers from MSU, Penn State University, Washington State University and Montana State University, as well as Michigan companies Precise Manufacturing and Proto-Tec Inc.
The project’s combination of robotic harvester and in-field sorter is novel and could help solve the growing problem of labor shortages, said ARS National Program Leader Jonn Foulk, who spoke at the demonstration.
Grower Phil Schwallier liked that the harvester caused very little fruit bruising.
“The catching frame and bin-filling apparatus are very gentle,” Schwallier said. “The apples look like they can be put right on your kitchen table.”
The demonstration did not include the sorter, but when the machine is fully integrated, a conveyor will move the harvested apples to the in-field sorting module on the same platform, where it will sort the fruit into fresh and processing bins. MSU researchers are analyzing apples picked by the harvester, to check for bruising, but Lu said he is confident the system handles fruit as gently as human pickers.
Lu’s harvester uses a vacuum system. Guided by a computer system, each robotic arm uses suction to attach the apple to the soft rubber end effector. The vacuum system runs constantly but ramps up to full capacity only when the end effector nears the target fruit. Once the apple is attached, the vacuum automatically ramps down again, for greater energy efficiency, he said.
Unlike gripper-style end effectors, which require the perfect approach to an apple, the vacuum system can pick even if the approach isn’t ideal, said Li of MSU.
There are similar harvester concepts being developed, but as far as the Michigan collaborators know, none include an automated in-field sorting component.
Lu recently saw a vacuum-based robotic harvester developed by California company advanced.farm in action in a Washington state orchard, and he made some comparisons for Good Fruit Grower.
The advanced.farm harvester has six arms, compared to Lu’s eight, but each arm is more dexterous than those of the USDA model. When the advanced.farm arms pull apples off a branch, they sometimes pull off spurs, too. Lu’s vacuum arms suction the apples and twist them slightly for easier detachment, he said.
There also are differences between the perception systems that guide the robotic arms. Advanced.farm uses a stereovision system: two cameras close to the end effector. This gives more accurate information on fruit position, Lu said. The Michigan prototype’s perception system is mounted on the mainframe, not the arms. It identifies all fruit within a larger working space and then uses that information to efficiently guide the arms. The camera sits farther away from the canopy, however, so information about the location of specific fruits isn’t as accurate, particularly if they’re covered by leaves or branches, he said.
To deal with that, the Michigan research team is developing an active laser scanning algorithm to improve the harvester’s ability to localize target fruits, Li said. •
Leave A Comment