Pamplin Media Group – Grapes of math: New technology for vineyards

GFU students are developing an autonomous robot that could radically change the way vineyard owners predict their harvest.

They call it the “Vitibot”. The name isn’t official, but a student designed a sticker with the name on it and it stuck – a clever combination of ‘viticulture’ (the science of grapes) and ‘robot’.

One of George Fox’s longest running high-level design projects for computer science and engineering students, the Vitibot is getting closer to the goal and a product the wine industry surrounding the university expects every year. looking forward: a rover that can autonomously traverse a vineyard, gathering image data, to accurately predict grape yield months before harvest.

A margin of error of 10-15% is considered a good estimate by most winemakers. Last year, the Vitibot predicted a return with a margin of error of 2-3%.

A Roomba for grapes

Each year, winemakers closely monitor their vines to predict the yield of grapes. Weather, soil, vine diseases and pests like birds and deer affect this yield – but regardless of these elements, vineyard owners need to know if they can meet their commitments to buyers and s they will end up with excess grapes. In the risky business of growing crops, their success depends on accurate prediction.

Iterations of the Vitibot have been around longer than most senior design projects, which typically start in the fall semester and end in the spring.

“I started here in 2015, and it was already a thing,” said computer science professor Brian Snider.

Bob Harder, dean of the College of Engineering, had brainstormed with local entrepreneurs and vineyard and winery owners, developing the technology for a few years before Snider came to the university.

“From the beginning, this was primarily a mechanical engineering project,” Snider said. “But Bob knew there was a need for machine learning and that’s where I got involved.”PHOTO COURTESY: CHRIS LOW - The Vitibot is self-contained and early testing shows it has a very low margin of error when predicting wine grape yields.

Two teams of seniors – new teams each year – worked in parallel with the Vitibot.

The hardware, or data collection, team of mechanical and electrical engineering students designed and built the rover.

The team of computer science and information systems, or data analytics, students worked on machine learning and autonomous navigation – artificial intelligence.

Finally, this year, the two teams merged into one, assembling mechanics and sensor data.

“And that’s when we had the thing to drive itself,” Snider said.

Snider and other College of Engineering professors envision a rover they can sell that has an array of sensors and can be maneuvered by remote control or piloted autonomously (currently the Vitibot uses LiDAR sensors for navigation and GoPro cameras for data collection).

“It’s like a Roomba, almost, but for your crops,” Snider said. “You let it roam the rows of vines on its own and collect data as you go.”

A community effort

To create a rover that could take digital images, feed those images into a machine learning model, and use the trained model to predict performance, the teams needed a proving ground. David and Jeanne Beck, owners of Crawford Beck Vineyard, voluntarily donated their vineyard. Scientists, researchers and educators themselves, they are also interested and invested in technology that can help their vineyard and the process of enabling students to learn.

The Vitibot’s accuracy in predicting grape yield depends on two things: weekly photos of the vines and machine learning built into the rover’s “brain.”

For the past three summers, a student or faculty advisor has walked the rows of Crawford Beck Vineyard, carrying an eight-foot hiking stick with three GoPro cameras attached. Every week, from July until the harvest in mid-September, they took a photo of each vine stock. Eighty-three plants per row, 21 rows, 10,000 pictures every week.

“We have an algorithm, a piece of code, that accepts images as input,” Snider said. “And we have code that learns the relevant numerical features from the images. That information comes in – we call it training data. Then, because we’re doing supervised learning, we tell it, ‘Given these images, here is the correct answer you should predict.'”

The “correct answer” given to the algorithm is the weight of the grapes at harvest. After uploading images all summer, the actual weight is added to the algorithm at the end of the season. This is where the Becks’ contribution figures prominently.

In most vineyards at harvest, workers crisscross the rows, cut the bunches of grapes and drop them into buckets, empty the buckets into bins, earning credit for each bucket they pick. It’s competitive and it’s fast. This saves labor costs.

But to validate the software of the rover, the premium is no longer on time; it’s about the accuracy of the weight of the harvest, not only by the vineyard, but also by the winery – in this case, Winderlea.

How the Vitibot goes from images to final weight of grapes, Snider can’t explain: “You don’t necessarily know what features it’s latching onto. Is he counting the grapes? How does he know? don’t need to know. We will let this algorithm learn what is relevant. ”

The numbers prove it works

“They’re ready to weigh all of these bins of harvested grapes and give us the data,” Snider said. “That’s what ties it all together and validates the success of the software.”

The process is time consuming but absolutely necessary to develop an accurate model. The Becks and Winderlea are ready.

“Their software is better than our actual measurements,” said David Beck. “We count the bunches and we weigh them, and we mark that up by the total number of vines. We make an informed calculation, but we make assumptions about the even distribution of fruit.

Like most winegrowers, Jeanne Beck takes care of forecasting the yield of the grapes herself. It’s tedious, time-consuming, and not as accurate as she would like.

“As for the margin of error that winegrowers want, the answer is zero!” she says. “That not being possible, we strive to both retain the work and achieve maximum precision. Given this inherent inaccuracy, some producers opt for optical assessments, going through the ranks and estimating with the naked eye. This is unsatisfactory, so we are constantly looking for better methods.

“And that’s the beauty of the work that George Fox’s students did. Their estimate was less than 98% of the harvest amount. Mine was only 72% of the harvest amount. Add to that labor savings and you can see how impressive this is.”

The Vitibot is not yet ready for the market, but the software has always produced accurate yield predictions and this year it was able to drive autonomously. “Now that we’ve broken through that wall, the next step is to take him to the vineyard every week and train him to drive in a real, real environment,” Snider said. “And we’re pretty optimistic that next year we’ll get there.”


You rely on us to stay informed and we rely on you to fund our efforts. Quality local journalism takes time and money. Please support us to protect the future of community journalism.

Comments are closed.