Illinois researchers use soft-armed robots to reduce labor shortage

In 2017, fruits and vegetables worth tens of millions of dollars rotted in California fields because of an ongoing labor shortage. Low-paying manual labor jobs are often filled by immigrants, but recent changes to migrant labor laws and the increasing complexity of immigration policy may have played a part in a severe shortage of people to fill these jobs.

A diverse group of researchers at the University of Illinois at Urbana-Champaign are working together to develop robots capable of reducing the agricultural industry’s reliance on manual labor.

"Labor shortage is a huge problem in agriculture," said ECE ILLINOIS affiliate faculty member Girish Chowdhary, assistant professor of Agricultural and Biological Engineering and principal investigator on the project. "Today’s large farm equipment enables a specific type of agriculture, a monoculture of soybeans or corn. We had this idea to use soft-armed robots to assist with the most labor-intensive crops."

Traditional robots, such as those used in car production, are designed to do one task or a set of tasks and do them quickly and accurately. They are generally very large, and their success is measured by how many of certain product they can make in a set amount of time in a controlled environment. The same model wouldn’t work in agriculture, where each task is different, the environment is unstructured, and speed isn’t as important as dexterity.

"The behavior of conventional robots is very accurate and precise, but soft robots exhibit adaptable and dexterous behavior," said Girish Krishnan, assistant professor of Industrial and Enterprise Systems Engineering, whose expertise is in nontraditional, soft robotics. "When (we’re) talking about berry harvesting, each berry shrub is different, and thus different strategies may be needed to reach intricate positions. Soft robots are inherently adaptable to these variations and can be a game changer in the agriculture industry."

Berries are among the crops whose effective growth requires the most human effort. Berry shrubs are similar to monoculture crops in that they are susceptible to disease, weed, and pest pressures, but they are different in that they don’t grow in a standard shape and the crop doesn’t ripen uniformly across the plant. These attributes, combined with the delicate makeup of a berry shrub’s branches, make it an excellent example of a crop for which future soft arm robots could be useful.

While potentially a far better choice for the agriculture industry, soft robots come with their own challenges. The field of soft robotics itself is very new, with no practical applications yet commercialized. Most soft robots currently built can’t support an onboard power supply—which would be a necessity for robots expected to roam a field for hours at a time. In addition, some machine learning must take place for the robots to complete their assigned tasks.

"The biggest problem is control," said Chowdhary, who is also affiliated with the Departments of Computer Science and Aerospace Engineering as well as the Coordinated Science Lab. "Traditional control methods don’t work for soft robots because of their soft and compliant nature."

A major challenge the team is working on is that of making the robots learn difficult tasks from experience. One technique they are applying is reinforcement learning. The goal is for the robots to learn tasks through a few demonstrations from humans or other robots. The team believes that tackling the control problem for soft robots will drive advances in reinforcement learning and control.

Predicting how the robots will achieve their tasks and interact together is the job of Mattia Gazzola, the team’s computational modeling expert. Whereas traditional robots are normally tested through trial and error, that wouldn’t be an ideal approach when a robot is dealing with valuable growing plants. Unlike hard-bodied robots, soft robots have the additional challenge of body fluidity; when one part of the robot moves, it may affect another pagudert.

Gazzola will run hundreds of simulations in order to predict accurately how a robot’s movements will impact the robot itself and the other robots around it.

"One of the challenges is to acquire data to determine how we should activate the robots and get them to move in the field," said Gazzola, assistant professor in Mechanical Science and Engineering. "The goal is to do all of this in the computer. This way we can enhance the capabilities of these robots and develop them faster."

One way the team will determine what movements are necessary will be to videotape the actions of crop sciences researchers as they perform maintenance tasks with berry shrubs. Sarah Lovell is the team’s crop scientist and polyculture expert. Her research group manages the berry plots that will eventually be used to test the robots. Their movements while pruning, weeding, inspecting, and harvesting the berries will be recorded for the group’s roboticists to mimic. When the robots get to the testing stage, Lovell will be there to oversee implementation and help troubleshoot potential issues.

"I think field testing will help overcome the challenges we’ll face when developing the robots," said Lovell, an associate professor in Crop Sciences. "We’ll learn a lot every time we take the robots into the field and see how they are working. It will be an iterative process in that way."

This project is funded in part by the National Institute of Food and Agriculture through a joint program with the National Science Foundation called the National Robotics Initiative.

Read the original article on the CSL website.