8 July 2019. A lab in the U.K. applying machine learning and robotics to agriculture developed and field-tested an autonomous device that harvests iceberg lettuce. Researchers from University of Cambridge describe their system and test results in yesterday’s issue of the Journal of Field Robotics.
Iceberg lettuce is one of a number of crops, like soft fruit and vegetables, that so far resisted automated and mechanized harvesting. Applying robotics to agriculture, in particular, is difficult, because of the unstructured environment in which crops are grown. And with agriculture, robotic systems need to work in harsh weather as well as soil, water, and chemicals that can befoul electronic components.
A team from the Biologically Inspired Robotics Laboratory at Cambridge, led by robotics-engineering professor Fumiya Iida aims to meet these challenges with a robotics system designed to harvest iceberg lettuce, a popular vegetable used in salads and other dishes worldwide. Moreover, iceberg lettuce farms often need to hire part-time harvesters on short notice, to meet on-demand production and and shipping requirements of large retail customers in growers’ supply chains. In addition, iceberg lettuce heads grow on stalks of varying lengths, surrounded by other vegetation on uneven terrain, and must be cleanly cut from a stalk without damaging the leaves.
The team led by research associate Josie Hughes developed a system they call Vegebot to meet these requirements. Vegebot uses camera images evaluated with computer vision and machine learning for training the system to identify lettuce in the field and find the heads meeting the standards for harvest. A system called a convolutional neural network combines image analysis and machine learning with an algorithm that dissects an image by layers to understand the features in the image. Different aspects of each layer discovered and analyzed by the algorithm are translated into data that the algorithm then uses to train its understanding of the problem being solved, with that understanding enhanced and refined as more images and data are encountered.
While the convolutional neural network is identifying and evaluating crops for harvest, Vegebot is also measuring the length of the stalk connecting to the head, and calculating the best spot to cut the stalk. The device includes a gripper and cutting mechanism, where the gripper needs to apply enough force to lift the head off the ground for cutting, yet not so much force that it harms the lettuce. And after testing several cutting methods, the team chose a pneumatic cutting mechanism that offers the fastest and cleanest cuts.
The Cambridge team partnered with G’s Growers, a local farming cooperative, to test a prototype Vegebot in the field, and compared the results with manual harvesters. “For a human,” says Hughes in a university statement, “the entire process takes a couple of seconds, but it’s a really challenging problem for a robot.”
The tests results bore out the extent of this challenge. The Vegebot assessed 69 lettuce heads in the field. The system successfully identified 91 percent of lettuce heads from surrounding vegetation and accurately classified 82 percent of the heads as ready or not fit for harvest. And while the Vegebot successfully collected 88 percent of heads ready to harvest, the device damaged 38 percent of the heads, with each head taking about 32 seconds to locate, identify, and harvest, far longer than human harvesters.
While these mixed results show room for improvement, the researchers are taking steps to boost Vegebot’s performance, and sharing their data to enhance farm operations in general. “We’re also collecting lots of data about lettuce, which could be used to improve efficiency, such as which fields have the highest yields,” says Hughes. “We’ve still got to speed our Vegebot up to the point where it could compete with a human, but we think robots have lots of potential in agri-tech.”
More from Science & Enterprise:
- Patch Performs Fast DNA Collection for Plant Diseases
- Flying, Driving Drone Robot Unveiled
- Genomic Analysis Finds Lost Tomato Taste Gene
- NSF Funding Natural Agri-Chemical Technology
- Genetic Engineering, Silencing Boost Plant Output
* * *
You must be logged in to post a comment.