A robotic harvester being developed by Vision Robotics employs a two-step approach. First, a scout system maps the location of the fruit on the tree. Then the harvester’s eight vision-guided arms pick the fruit. The scout system assesses the size and grad
A robotics company is developing a fruit harvester that might not only solve growers’ labor problems but also enable them to pick just the fruit that buyers want and leave the rest on the tree.
Vision Robotics Corporation of San Diego, California, is developing the harvester for the citrus industry, though it would work equally well for apples, pears, and even cherries—with or without stems, according to Ted Batkin, president of the Citrus Research Board, which is funding the project.
Robots work best in highly preprogrammable, repeatable environments, he said. Unstructured environments, like orchards, are challenging for robots. A robotic harvester must be intelligent, so it can make decisions very rapidly in real time about whether to pick the fruit or not, and it must be able to “see” in order to operate autonomously. Such a machine must also be low cost in order to be competitive with the price of unskilled labor.
When other manufacturers have attempted to develop a robotic harvester, the “end effectors”—the devices that reach out to pick the fruit—have been a stumbling block, Batkin reported at the International Fruit Tree Association’s annual conference. “All other robotic approaches try to locate fruit as they pick it.”
Vision Robotics is trying a different, two-step approach. One robot acts as a scout and another as the picker. The scout scans the orchard in advance and builds a three-dimensional map of each tree with the location of the fruit and interfering branches. It then preplans the fastest picking strategy for each tree.
The picking plan it develops is then sent to the picker, a machine with eight, long, vision-guided arms that can pick fruit simultaneously. Together, the arms can pick four fruit per second with the stems cut. The harvester incorporates 22 camera/computer systems.
As well as locating the fruit, the scout can tell what size and color the fruit is and if it has defects. In the case of oranges, for example, it can tell if the fruit is going to be export grade, choice grade, or used for juice.
Working as a stand-alone device, without the harvester, the scout can be used to predict volume or packout, or develop predictive yield maps.
“It’s going to tell us 30 days before we pick it, so we can make value judgments about what we’re going to do before we get there,” Batkin said. “Currently, 100 percent of the fruit goes to the packing house, then we decide what to do with it. With the scout system, we can send the harvester into the grove and pick only export grade size 88s if that’s what the packing house needs. We don’t spend time and money transporting something else. Later, if the size 138s grew a bit, we can pick those. We can selectively pick whatever we want. It changes the entire dynamics of what we’re doing.”
Batkin said agricultural producers have to respond to what the buyers are asking for. “We used to be able to put everything in a box and sell it to somebody, but we don’t have that luxury any more. With these types of systems, we’ll be able to respond to these changing dynamics.”
He foresees the scout being used as a complete crop management information system, providing information also about trees that are showing nutritional deficiencies or signs of disease, for example.
The harvester, which is six feet wide and compatible with standard tree rows, will also be a versatile piece of equipment with different screw-on attachments that can be fitted on the ends of the arms for jobs such as pruning or thinning. The machine would do selective thinning, to space the remaining fruit a certain distance apart. It will be mounted on a simple platform that’s either towed by a tractor or self-propelled.
The scout and harvester will be georeferenced, but not by a global positioning system alone, Batkin said. When they enter a block, they will start from a reference point, and everything in the orchard will be measured in relation to that georeference point, so they don’t have to take GPS readings. The machines will know where each tree is, to within an inch of their starting point.
A grower’s capital outlay for such equipment might be $350,000, plus $25 an hour for operating and maintenance costs, Batkin estimates. If it were used for 39 weeks, 6 days a week, 12 hours a day, the cost, amortized over five years, would work out at $17 a bin, compared with the $18 a bin it costs to employ people to harvest fruit.
Other issues that the harvester addresses in addition to the rising cost of labor include:
—labor shortages that can make it difficult to get a perishable crop harvested at the right time;
—the fact that people can only work in daylight; and
—the tendency for fruit quality to suffer when a crew is focused on speed.
The harvester is still at the conceptual stage, though Batkin is confident it will soon be a reality. Vision Robotics has been working on it for the past three years and has demonstrated its technical feasibility through computer simulation. This, Batkin explained, eliminates risks. “Boeing, when they build an airplane, doesn’t go out and build parts and tail pieces and bolt them together and see if it’ll fly. They build it in virtual reality on a computer. That’s how we’re doing this project now.”
Ninety percent of the software can be developed in simulation. The company is aiming to have the scout system complete in 18 months and the harvester within three years. The total project will cost $6 million.
“It’s going to come,” Batkin said. “This stuff is going to happen. Whether we’re the ones that make it happen or someone else, it’s just a matter of luck and timing.”