To build a “hand” for a robot apple picker, Joe Davidson first had to learn how to pick apples.

After all, he was an engineer, not a farmer.

“Picking apples seems like a very simple thing to do,” Davidson said. “Replicating the patterns and sense of touch that professional pickers have is very difficult in a machine.”

Davidson, then a Washington State University graduate student, was taught to pick by growers. He spent days in the field using sensors to record himself squeezing, rotating and pulling apples off branches.

Now Davidson is an Oregon State University assistant professor. He and Professor Cindy Grimm are running a project dedicated to robot hands.

They were awarded $440,000 from the USDA in 2020 for a four-year study.

Previous coverage

Apple exports still double that of last season

Last year they received an additional $250,000 from the Washington Tree Fruit Research Commission to continue developing their harvesting gripper in collaboration with Wageningen University and Research in the Netherlands.

Artificial trees and fruit inside their OSU lab mean studies take place year-round, regardless of growing season.

“If you want to use AI and machine learning, you need lots and lots of data,” Davidson said.

He predicted that in the next five to 10 years, robots will pick apples commercially. It may take much longer before machines harvest the majority of acreage for the nation’s top fruit commodity, with $3.2 billion in farm-gate revenue.

“We’re playing the long game,” Davidson said. “Automation has to happen.”

Rise of the machines

Robot apple harvesters are being developed and field tested by universities and companies around the world as growers face a worker shortage and rising labor costs.

But the rise of machines remains years away from orchards.

For now, robots can’t compete with human labor for cost and picking efficiency.

“There is progress being made but picking by human hand continues to be the most effective way to harvest our crops,” said Brianna Shales, spokesperson for Stemilt Growers, a large Washington-based producer.

Engineers will continue to make advancements, said Manoj Karkee, director of WSU’s Center for Precision & Automated Agricultural Systems.

In the past decade, his lab has received roughly $1.5 million in USDA funding for robotic apple harvesters.

“We are closer than ever,” said Karkee, who is working with Israeli firm FFRobotics on a robot that moves between rows, six arms on each side.

“It will take a few years until robotic solutions are widespread in the industry,” said Avi Kahani, co-founder and CEO of FFRobotics.

Leading candidates

FFRobotics and two other companies are leading the race to create robotic apple pickers: Tevel, an Israeli firm that uses flying drones, and Advanced Farm Technologies, a Davis, Calif.-based company that pioneered a robotic strawberry harvester before adding apples to its portfolio in 2022.

This season, Advanced Farm’s robots picked 300,000 apples during tests in Washington, said Peter Ferguson, director of business development.

Its robots navigate rows using Light Detection and Ranging sensors called LIDAR. Both sides of the machines have three arms — each with a camera for finding fruit and a suction cup to pick apples.

Ferguson said his robots have occasionally picked 2,000 apples per hour.

“Making that more consistent, where it’s the norm, where we’re always picking apples at that rate, will be really important,” he added.

Apple, pear production

Global apple, pear production edges higher

Speed and cost

Karina Gallardo, a WSU economics professor and extension specialist, said picking speed was the most critical improvement that could jumpstart robots, followed by rate of damaged apples, and missed fruit.

Timothy Delbridge, OSU assistant professor of applied economics, said speed has been tough to increase, because agricultural settings lack uniformity.

With an uneven fruit load, “There’s going to be some time when one arm sits idle,” Delbridge said.

Robotics expert Professor Stavros Vougioukas of the University of California-Davis said engineers must figure out load management and orchestrate the arms so they’re always working.

For an analysis, Gallardo and a team that included Vougioukas assumed robots could pick roughly one apple every three seconds.

At that rate, a robot could harvest in about 14 hours what would take 120 man-hours, she said.

Using a machine would save $461 per month compared to using manual labor per acre, she estimated. The analysis didn’t include upfront costs of buying or contracting robots.

Delbridge said many robotics firms are considering a fee-for-service model.

Vougioukas said robot pickers still need to get much closer to the cost of manual harvest.

Picking efficiency

For machine harvesting apples, variety isn’t as critical as orchard architecture, experts said.

Some Washington orchards have small-canopied trees trellised vertically in an almost two-dimensional fashion, creating green walls dotted with fruit.

Davidson said orchards evolved that direction to make fruit easier to reach for humans. But that style also works best for robots.

“The more farmers plant that way, the more prepared they will be for the future,” Ferguson said.

Extra pruning and thinning paid off in tests, he added.

Even with the best maintained orchards, top machines can only pick 60% to 70% of apples, Karkee said.

Cameras can’t detect apples obstructed by leaves or hidden behind other fruit. Robots also have difficulty grabbing clusters or apples tucked behind main limbs or trellis wires.

Some also snap branches, damaging trees, Karkee said.

Apples that dropped on the ground and are bruised is a concern, too.

“Another challenge is, once you get it off the tree with the robot, how do you put it in the bin without damaging it?” Davidson said.

Advanced Farm addressed this problem before its second season with a conveyor system inside the picker, replacing a slide on which apples rolled into bins.

Different approaches

Engineers have been developing robot fruit pickers since the 1980s, but Karkee said widespread research started about 2010 thanks to affordable technology, more powerful sensors and machine learning advancements.

Despite progress, there’s no agreement on the best design — even for robot grippers.

FFRobotics’ machine has hands with two wide fingers that gently clamp over an apple almost like a cup holder.

Both Advanced Farm and Tevel use vacuum suction cups instead.

Davidson is working on a design that combines both ideas, with three vacuum suction cups to make the initial attachment, plus three fingers.

“We tested this up in the field in Washington in the fall,” he said.

Davidson added that concept variety is a strength for robotic harvest. “It needs lots of different ideas and we’ll see what works long term,” he said.

Adapting to other uses

Efforts are underway to automate harvest for other produce worldwide because of the same labor issues facing the United States.

If a robot apple picker proves successful, Karkee believes it could be adapted for different crops. The picking mechanism would need modifications for factors such as how delicate the fruit is.

“The rest of the machine would be fairly applicable,” Karkee said.

The most valuable robots could be machines that can not only harvest, but also assist growers year-round with tasks such as pruning, thinning and pollinating crops.

WSU and OSU robotics labs are tackling some of those issues.

“It’s not just modifying the same machines to pick other fruit. Long-term, we need multipurpose robots,” Davidson said.

Development not a straight line

Advanced Farm was formed in 2018 and first focused on robots to harvest strawberries. Over the years, their robots have picked 5 million berries, and UC-Davis even came out with a breed of strawberry, UC Monarch, designed for robot pickers, with long stems and smaller canopies.

Delbridge said harvest efficiency went down for Advanced Farm’s strawberry picking robots later in the season, when plants were bigger and fruit was harder to find.

The company also realized it could create better efficiencies for farmers with an automated packing line. Workers no longer had to pick and pack, allowing them to cover 30% more ground in the same amount of time, Ferguson said.

Such a pivot is normal. “That happens a lot in research,” Davidson added.

For example, an OSU study on robotic pruning used automated estimations of branch girth. Researchers transferred that tech to measure trunk thickness — an indicator of nutrition — for a precision fertilization project.

Exploration of ideas is part of the development process, Vougioukas said. Much of his work has been devoted to harvest aids such as transport robots that deliver produce from fields to trucks.

“We don’t really know where we’re going to, but all these technologies are great, and they’re all meant to help growers grow more food,” Vougioukas said.

Things take time

The impacts of automation will take time and play out differently for agriculture than other economic sectors, Delbridge, the OSU economist, said.

That’s in part because many growers rely on foreign workers with H-2A visas. Delbridge said there will be less of an impact on the domestic workforce, and many employees will be retrained to oversee robots.

“I don’t think we’ll be in a situation where we’re going to have farm workers sitting around trying to find work,” Delbridge said.

Even if a robotic apple harvester were ready immediately, it would only impact a few orchards initially, experts said.

“It would take years, even for a mature technology to gain adoption,” Vougioukas said.

Delbridge said robotic harvest and automated agriculture isn’t some switch that will magically flip “on.” There will be intermediate steps.

“We have been seeing productivity gains for the last 80 years, and I think they’ll continue to come, but I don’t think we’ll see a silver bullet,” he said.