CMU robotics projects sorts plants for less


Share with others:


Print Email Read Later

The co-owner of a large California nursery hires 1,000 workers each season to trim and sort strawberry plants for shipment to berry producers.

But her problem is getting proper documentation for the immigrant work force, then assuring that those 1,000 "trimmers" sort the plants uniformly. Another concern for Liz Elwood Ponce of the Lassen Canyon Nursery in Redding, Calif., is the high cost of the labor force.

The situation made her stop and think.

There must be a company or university somewhere that could produce a strawberry-plant sorting machine to make the process faster and cheaper with better results.

And her Internet search spat out the name of one research university that seemed eminently qualified to create a robotic sorting machine to usher her business into the modern era.

Ms. Ponce's call to the Carnegie Mellon University's National Robotics Engineering Center in Lawrenceville drew immediate interest, especially when NREC officials learned she represented a consortium of five nurseries willing to sponsor the project.

Christopher Fromme, the NREC project's manager and lead engineer, said vision technology already exists to grade fruits, vegetables and berries. But no such technology existed to sort strawberry plants, or any other plants, for that matter. That's because of variations in each plant and changes that occur with age.

"That's why we're doing it," Mr. Fromme said. "We don't get the easy problems."

And this project wasn't easy.

But NREC already has completed the first and most difficult step in a five-phase project. The prototype classifies and sorts harvest plants more consistently and faster than workers can, with a comparable error rate, according to an NREC news release.

While the technology being used is proprietary, Mr. Fromme said it accomplishes the task with machine learning. That means a human sorter tutors it, then the equipment takes over with robotic technology and computer formulas known as algorithms to replicate what the human does.

"The machine learning with vision-system technology is the unique aspect in that the system is taught what to do by expert human sorters," Mr. Fromme said. "We are attempting to transfer expert knowledge from person to machine."

In short, the equipment evaluates root structures, crown size and number of petioles among other features to gauge whether the plant meets quality standards.

The equipment is necessary because strawberry producers replace all plants each season to maximize yields and quality.

That also means the nurseries can sell only those plants that meet producers' standards. Plants, which are shipped with their roots bare, must be harvested, trimmed, sorted, packaged and placed in boxes.

Mr. Fromme said each strawberry plant, much like a snowflake, is unique. A conveyor transports the plants through the machine, which classifies the plants according to size, variety and stage of growth. It also decides which ones qualify for shipment and which ones must be discarded.

The NREC team, Mr. Fromme said, put the robotic equipment through a field test for 10 days last October under realistic conditions. A news release notes that it sorted more than 75,000 plants at a rate of 5,000 plants an hour -- a speed several times faster than humans can achieve.

But NREC officials aren't yet satisfied. They hope to boost sorting speeds to 20,000 to 30,000 plants per hour, with an error rate equal to or better than what humans can achieve.

While the technology would reduce the workforce at each nursery, Mr. Fromme said higher-paying jobs would be created to build, operate and maintain the equipment.

With the prototype completed, the NREC team will focus on adding other robotic equipment that will require changes in the harvesting process. For instance, strawberry plants have entwined roots and runners that connect the plants underground. "New equipment is necessary to break up the plants to be separated from other plants without damaging them," Mr. Fromme said.

It could take five years to get the entire system in operation.

"They will get a very quick return on their investment," Mr. Fromme predicted for the five sponsoring nurseries that raise about 85 percent of the strawberry plants used in California. "That's the beauty of the system. We spent a lot of time making this as simple as we could. We made a lot of trade decisions that will make this system agriculturally robust."

Meanwhile, Mr. Fromme said the machine-learning approach that NREC developed could apply to other plants that are shipped with bare roots (or without soil). Those could include roses, flower bulbs, peonies and asparagus among other flowers, fruits and vegetables.

And, in retrospect, Ms. Ponce said, she picked the right university to solve plant production problems for her Lassen Canyon Nursery -- the largest such nursery in California with 1,000 acres in production

"This is a great project, and I'm really looking forward to what they will come up with in the end," Ms. Ponce said of NREC. "I hope it meets expectations."

In the meantime, she's thankful Mr. Fromme responded to her Internet search with genuine interest.

"Lucky me," she said.




David Templeton can be reached at dtempleton@post-gazette.com or 412-263-1578.


Advertisement
Advertisement
Advertisement

You have 2 remaining free articles this month

Try unlimited digital access

If you are an existing subscriber,
link your account for free access. Start here

You’ve reached the limit of free articles this month.

To continue unlimited reading

If you are an existing subscriber,
link your account for free access. Start here