Brad Hamner of Carnegie Mellon University walks alongside an autonomous vehicle as it moves down the rows in Washington State University’s research orchard near Wenatchee.
The question is not whether scientists can develop robotic pruning, tree training, and fruit picking systems. The question is how soon growers can adapt their orchards to accommodate them, a robotics expert says.
Scientists across the country are working on a $6 million, four-year project to develop automated equipment for apple production. A team at Carnegie Mellon University in Pennsylvania has developed a prototype of an autonomous vehicle designed to be used for a variety of orchard applications, such as insect monitoring, weed management, and plant stress and disease detection. There are also efforts under way to improve crop forecasting and harvesting efficiency with new technology.
But any kind of automated orchard equipment will work best in modern orchards where the trees are trained to a vertical axe system or fruiting wall, said Dr. Sanjiv Singh, robotics professor at Carnegie Mellon.
"The question is how soon is the grower going to adopt that," he said.
The $6 million project, funded through the federal Specialty Crop Research Initiative, began last year and involves scientists and extension educators from Washington State University, Pennsylvania State University, Purdue University in Indiana, Oregon State University, and several private technology companies.
The Carnegie Mellon team began working on the autonomous vehicle last October. The scientists purchased an electric utility vehicle and modified it so that it can be programmed to drive up and down specified orchard rows without needing a driver or operator. The vehicle, which can function day or night, can tow a mower, sprayer, or other piece of equipment while carrying a variety of sensors to monitor the tree canopy.
The basic Toro vehicle cost $10,000. Making it robotic involved adding a laptop computer and laser scanners to keep the vehicle on course and detect obstacles. The laser scanners are expensive in small quantities, Singh said, but would be cheaper if the vehicle were produced commercially.
"The good news is it’s low technology," he said. "If you were to do it at high volume, that’s very easy to do. I’m a great believer in trying to do things as simply as possible, using the least possible technology to solve the problem, not technology for technology’s sake."
The vehicle was tested first in Pennsylvania and then at Washington State University’s research orchard near Wenatchee and in more mature plantings at Valley Fruit Company’s Hilltop Orchard on the Royal Slope in Washington.
Brad Hamner, research programmer at Carnegie Mellon, who designed the software for the vehicle, said his greatest challenge was to try to make it work in as many different kinds of canopies as possible, because they look quite different to the laser scanners. For example, in a newly planted orchard with small trees, the scanners can see multiple rows, not just the row the vehicle is in.
The vehicle can be commanded to drive down a row and turn at the end into the next row, or follow any desired path, and continue to do that as long as necessary, Hamner said. "The idea is eventually you would start it at one end of the block and have the robot drive down the entire block."
In this summer’s tests, the vehicle was used to tow a robotic orchard scout that is being developed by Vision Robotics of San Diego, California, with financial support from the Washington Tree Fruit Research Commission. The objective is to locate the position and size of each piece of fruit on the trees. The scout has eight pairs of stereo cameras that take pictures of the canopy as it moves down the row, creating a three-dimensional map of the fruit.
Tony Koselka, vice president of engineering at Vision Robotics, said this information can be used to estimate the crop load and is a step towards the more difficult task of automating harvest. There’s also potential for the scout to go through the orchard during bloom to assess the need for thinning.
Since launching the project three years ago, the company has been continually enhancing the design of the scout to improve its ability to detect both red and green apples. The latest version has flash lighting to address problems with harsh shadows in strong sunlight and cameras that are tilted upwards so that their view of the fruit is less likely to be obscured by leaves. The company will do further tests in Washington this fall.
Singh said the same kind of components used to make the utility vehicle autonomous could be used to convert an existing harvest platform into an autonomous platform to carry workers through the orchard doing pruning, tree training, or picking.
Research is under way on technology that would improve harvest efficiency, but developing a totally automated harvester would be expensive and time consuming, Singh said. He sees greater commercial potential for new technology that can be used over a longer season for multiple tasks.