A new two-fingered robotic arm ‘end-effector’ developed by engineers at Mississippi State University (MSU) is a potential game-changer for global agriculture, according to its designers.
‘The systems technology we’re designing today will give tomorrow’s cotton farmers more ecologically and economically sustainable options for harvesting,’ said Alex Thomasson, the head of MSU’s agricultural and biological engineering department.
Advertisement
Inspired by the distinctive way that a lizard’s tongue reaches for its prey, the mechanism plucks one cotton boll at a time – rather than all at once like current harvesting machines – making it possible to harvest earlier and more often, when seed cotton is at peak quality, and reducing waste; plants are typically machine-harvested when only around 60 per cent of the cotton bolls are open.
The robotic arm is fitted with AI-assisted cameras that enable it to identify whether cotton is ready to be picked or not. It also has a high level of dexterity, enabling it to pick fibre from each individual cotton plant.
Advertisement
‘The cotton plant presents unique challenges to an AI-based camera system because bolls can be oriented in different directions and seed cotton is not solid and contiguous like an apple,’ said the picking device’s designer, Hussein Gharakhani, an MSU ag and bio engineering assistant professor. ‘Our end-effector, which took about a year to develop, works with our camera-based perception system to identify and retrieve the fibre from the boll.’
Xin Zhang, another investigator and department assistant professor, is focused on integrating Gharakhani’s end-effector with a commercial six-degree-of-freedom robotic arm and a four-wheel-drive robotic platform – the ‘Husky’, made by Clearpath Robotics – which operates with a GPS navigation unit and a perception module.
Advertisement
Central to development of the autonomous harvester is its AI perception module – an RGB-depth-based camera (similar to a smartphone camera), a 3D LiDAR sensor for obstacle avoidance and an AI-based processor. Zhang is using Gazebo software, an ROS-based simulator, to replicate the module’s performance in a virtual cotton field. She also is currently testing the robot’s in-field performance onsite at the university’s RR Foil Plant Science Research Center, known as North Farm.
‘For the last few years, we’ve been building and testing these systems individually, and over the next year, we’ll focus on integration and navigation with the goal of building a completely autonomous harvester that can work across unpredictable and uneven terrain,’ she said.
Addressing other unpredictable challenges is driving all areas of agriculture to adopt autonomous technology, according to Thomasson. ‘As the population grows and more people find work in urban areas, fewer people are available for, or even interested in, farm labour, and it’s often difficult to find people who are qualified to operate large agricultural machinery,’ he said.
Advertisement
In addition, economic and environmental effects of conventional harvesters are incentivising robotic solutions. Today’s six-row, round module harvesters, which gather most of the cotton in the USA and weigh about 30 tonnes, can compact soil to the extent that fertiliser and water become less effective in the wheel tracks, leading to possible yield reductions. Also, since these machines harvest at season’s end, fibre from early-blooming bolls is often lost.
The research has been published in Smart Agricultural Technology.