Agricultural robotics are not necessarily just for the biggest global produce markets, research designing a low-cost robot to plant Jersey Royal potatoes suggests.
University of Cambridge engineers led by Elijah Almanzor set out to devise alternatives to mechanical planters that had been previously used by the potato’s farmers in Jersey. These devices had resulted in errors such as duplicate planting and damaging potato shoots, while their weight put them at risk of compacting soil and posing hazards for users.
With farm labour for manual planting a concern, automation would offer a solution for farmers. But a specific technological challenge was that planting of Jersey Royals is done from standard box crates, which makes visual detection of potatoes by automated machines more difficult.
Further raising the bar for the researchers, they also looked to design a robot to meet the requirements for commercialisation laid down by the Jersey Farmers Union, which was for the robot to cost less than £75,000, have at least a 90% success rate and execute each planting in 3.5 seconds.
The rise of PotatoBot
Over a 10-week period, the team came up with PotatoBot, a robot using a pneumatic spear gripper attached to an arm and operated using the object detection algorithm Mask R-CNN.
In the short time, they brough the robot through from concept to testing, making use of an agile project management approach. This broke the challenge into separate tasks which could be repeated and improved in lab conditions set up to replicate the field, while keeping Jersey farmers informed of their progress. They also used off-the-shelf components to speed up the process.
“Developing a specialised robot for this niche market with limited economic resources incurs high costs and logistical challenges,” explained Almanzor in a paper for the Towards Autonomous Robotics Systems conference.
“To mitigate these challenges, testing in simulated farm conditions as a physical twin is necessary to reduce on-site iterations and developmental costs, especially in the early stages of robot development.”
Need to hone performance
In the lab tests, performance fell a little short of targets needed for commercial viability at this stage, with an 87% success rate and an average task time of 10.38 seconds. However, the robot fell well under the target cost for commercialisation, at only £32,000, and the set-up by the team proved its worth in speedily developing a viable technology for the niche market.
“Through farmer consultation and laboratory simulations, the research demonstrates the potential of robotic automation in addressing labour shortages,” said Almanzor.
There is still much work to be done to have a design ready for Jersey’s fields. More representative training data is needed to hone the algorithm for use in the field, while the use of a visual sensor to provide feedback and optimise robotic arm movements will be needed, said Almanzor. Field testing use full crates of potatoes will also be required before the technology can be considered ready for widespread use.
