The Elusive Hunt for a Robot That Can Pick a Ripe Strawberry

Traptic’s creators say it can pick 100,000 strawberries a day. It will now work exclusively in Bowery indoor farms, marking the first use of robotic arms at the company, which relies heavily on computer vision, sensors, and technology to grow lettuce for customers like Safeway and Walmart. Bowery intends to move robotic arms between indoor strawberry rows—as it does in fields—using automated vehicles. In addition to harvesting, Bowery will explore use of robotic arms to pollinate strawberry flowers and do maintenance work like thinning or pruning leaves.

Bowery was founded in 2015 and has operations near cities including New York and Philadelphia. Backers include GV, formerly Google Ventures, and individuals like Uber CEO Dara Khosrowshahi and former Amazon consumer CEO Jeff Wilke. In May 2021, Bowery raised $300 million at a $2.3 billion valuation and announced plans to expand operations to the Atlanta and Dallas-Fort Worth areas early next year.

 

Last year, Bowery opened Farm X, a New Jersey research facility where the company explored vertical farming of cucumbers, strawberries, and tomatoes. Farm X opened shortly after Bowery hired Injong Rhee, who previously worked on machine learning projects at Google, as chief technology officer. Rhee said Bowery will use Traptic’s technology to begin selling strawberries this spring, adding that the tech is “mature enough to get us there.” Bowery envisions making Traptic a core part of Bowery’s fruit and vine operations.

 

Rhee said Traptic stood out from other companies in the space because its robot doesn’t touch the strawberry. Instead it grabs strawberries by the stem, pulling strongly enough to detach the fruit but lightly enough to avoid damaging the rest of the plant.

Such tasks are an ongoing challenge. A recent study reviewed nearly 50 harvesting robot projects from 2000 to 2020 and found that harvesting robots have yet to see widespread use, in large part because most robots still can’t do a better job than a human.

 

Hugh Zhou is lead author of an analysis of AI progress in fruit-harvesting robots and their commercial viability. The study was carried out by researchers developing an apple-harvesting robot at Monash University in Australia. Based on what’s possible today, Zhou said he can envision a scenario in which robots pick 70 percent of easy-to-classify strawberries and humans pick the remaining crop. It’s just in recent years that HarvestCROO Robotics and a handful of other companies have advanced their systems to the point that they’re picking strawberries at rates competitive with people.

 

Zhou says the makers of fruit-harvesting robots are heavy on demo videos and light on data. The majority don’t share damage rates publicly, or how well their vision systems perform when picking strawberries from clusters or strawberries partly hidden behind leaves, two primary reasons computer vision systems fail to pick fruit. Crushing or bruising berries is still a common problem, and accidentally harming one strawberry while attempting to pick another is a problem that is often overlooked and rarely mentioned.

 

Companies working with robots to harvest fruit say there are reasons many haven’t moved out of the prototype stage and only a handful of machines are in operation today.

 

In April 2021, AppHarvest, operator of the largest greenhouse in the US, acquired Root.ai, a Somerville, Massachusetts-based company whose Virgo machine picks grape tomatoes with a small, three-pronged robotic gripper. That technology is now used to pick strawberries and cucumbers as well, using four- and eight-fingered grippers. As part of the deal, Root.ai CEO Josh Lessing became CTO at AppHarvest.

Picking rates have doubled since the acquisition, Lessing says. He says the company now wants to reduce the cost of the robots as it moves toward finalizing Virgo hardware in 2023.

 

Today Virgo picks healthy fruit better than most people, but it needs to improve its ability to detect ripeness and damage less fruit before being deployed in widespread use. Reducing the damage rate is tied to more use of soft grippers and passive forms of robotic control, says Lessing, a former research director at Soft Robotics.

 

Proponents of agricultural computer vision systems argue that being able to predict when fruit is ripe will lead to improved sales, reduced waste, and yield gains as the global population expands to 10 billion.

 

Robots can also help grow and market expensive specialty fruits. Robots working in tandem with automated vision systems can monitor crops 24 hours a day to predict the ideal time to pick a ripe, red strawberry. The startup Oishii raised $50 million last fall for its vertical farming operation as it works to replicate the process of growing sweet Omakase strawberries typically found in the Japanese Alps. A box of 11 berries goes for $50.

 

Advanced Farm operates 10 robotic picking machines at Blazer Wilkinson strawberry farms in Central California. Each machine can pick roughly 100 pounds of strawberries an hour. Tarps drape along the top and the side of a machine that sits atop two rows of strawberries at once. The tarps keep out light and help cameras and computer vision systems classify fruit and control robotic arms—light can affect the computer vision system, so the machines mostly operate at night.

 

Each time a ripe strawberry is identified, a silicone robotic hand with a suction cup in the middle moves in, grabs the strawberry, and then uses three fingers to twist it away from the stem and place it in a bin. Advanced Farm designed nearly 50 versions before deciding on the current design of its picking system.

 

Advanced Farm cofounder Kyle Cobb says the company’s robots are “in the ballpark” but are still more expensive than human strawberry pickers. “It’s one of those problems that we’ve mostly gotten through, but like with all problems with robotic harvesting, the last few percent are going to be hard to build up,” he says.

John Wilkinson of Blazer Wilkinson says he became interested in robotics a few years ago in response to labor shortages. He says the technology is still in the research and development phase, but he thinks it will ultimately become essential.

Hillary Thomas is research and technical director at farm conglomerate Naturipe. She says companies working on strawberry-picking robots have made big strides in reducing rates of damage since Naturipe started testing robotic harvesting in operations in 2016. Naturipe works with robotics companies, including Traptic and Harvest CROO Robotics, and she says each company can now reliably pick marketable fruit that meets company quality specifications.

 

Factors such as the cost of these machines and their output will determine whether the robots replace farm workers, but Thomas said there’s no longer any question that robots can overcome challenges to successfully harvest strawberries. Whether robots are adopted by growers in the fields of California will come down to the cost per pound of delivering berries to consumers.

 

Thomas says she can envision human-machine scenarios in which robots pick at night and people pack during the day. Grape harvesters in California fields already work alongside Burro robots. Robots are also being introduced to carry out other dedicated tasks on farms, like weeding, pruning, pollinating flowers, and painting fruit in UV light to protect it from mold or mildew.

 

Zhou, the Australian researcher, says that despite today’s shortcomings, the machines are improving. Recent advances signal that “soft robotics combined with deep learning algorithms might be the solution to the last mile of this fruit-harvesting challenge,” he says.