Features
The Thorvald robot developed by Norway-based Saga Robotics does UV light treatment on strawberries at night.
to automate everything. Combining humans with auto- mation is how we succeed, but we can’t just do a like-for- like replacement.” One example Roberts cites is a Cambridge Consultants project where a client asked the team to develop an apple picker. After studying the working environment, Roberts says it became apparent that the greatest need was in transporting the apples back to the packinghouse rath- er than the task of picking. “Rather than developing an apple-picking robot, we looked at how we could improve the situation for the workers,” he explains. “So instead of replacing people with robots, we’re making the same number of workers more productive.” However, not all projects work in the field. Roberts recalls the example of a startup that developed a straw- berry picker and asked Cambridge Consultants to test it in field conditions. “Although the company achieved what it set out to do, it was a long way from successful,” he says. “The project’s goal was to pick a certain percentage of strawberries that were clearly visible and easy to reach, but of course the farmer doesn’t care about that — the farmer cares about picking all the strawberries. This meant the machine was never going to be profitable for the client.” Addressing the Challenges Based at the epicenter of agricultural robotics start-up activity in the United States, a team at the University of California, Davis led by Professor Stavros Vougioukas, vice-chair of its Department of Biological and Agricultural Engineering is looking at ways of optimizing robotic har- vesting designs. In collaboration with two other univer- sities — Carnegie Mellon in Pittsburgh, PA and Montana
State in Bozeman — UC Davis has been awarded a federal grant for the development of a multi-armed harvesting robot by the National Institute of Food and Agriculture, the research arm of USDA. The aim, according to Vougioukas, is to overcome two obstacles that robotic harvesters are facing: one is the low visibility of the fruit; the other is the inadequate speed of the robots. “Everyone is talking about machine vision and AI, and they are great technologies, but unless something is visi- ble from a camera you can’t really identify if it’s a fruit and if it’s ripe,” he explains. “So one of the biggest problems is, depending on the crop, some of the fruits are not even visible to the robot. “For instance, if you go to a strawberry field when the plants are smaller, a robot with one or two cameras can find the fruit, but come June, July, or August when the plants are much bigger and there is more foliage, then a lot of the fruit is hidden under the leaves.” This means, Vougioukas says, that it is challenging for a robotic solution to achieve harvesting of 95% or 100% of a fruit crop, especially in the case of low-lying produce such as strawberries that can be obscured by leaves. “If you look at apples, if you miss a few apples it’s not as much of a problem, but it’s more of a financial issue,” he continues. “If the robot can only pick 70%, either you leave the remaining 30% on the trees and you lose that money, or you will have to bring people in and that fruit may not be the easiest to reach.” UC Davis’ project is addressing the challenge in two ways: one is through multiple cameras providing a range of views of the canopy. By combining these different
42 Vision Magazine
October/November 2023
Powered by FlippingBook