Corn is one of the most essential ingredients in global industry and agriculture. From tortillas and ethanol to starch and alcohol, the plant remains a pillar in many production processes. Measuring the vitality and nutrient density of corn has become crucial to maintaining the efficiency and scale at which we use corn in our lives, but it can be a lengthy and time-consuming process. 

Researchers at Carnegie Mellon University and Iowa State University created an agricultural robot platform to address these challenges. Led by Robotics Institute (RI) Research Professor George Kantor, RI Associate Professor Oliver Kroemer and Ph.D. student Mark Lee, the project enables robots to autonomously insert nitrate sensors into corn stalks and monitor macronutrient levels in the crops. 

Inserting sensors into corn stalks requires a great amount of precision in environments that are often filled with unpredictable obstacles such as natural clutter and uneven terrain. To address these challenges, the team developed a robust perception-action pipeline that employs a deep neural network to detect and identify the corn stalk with the highest likelihood of grasp success rate. The robot then uses a custom gripper that mechanically aligns the sensor with the stalk before inserting it into the plant. 

The team designed the robot platform with easily configurable dimensions so the robot can adjust to the unique row spacing and height of each field. They aimed to insert sensors into the plants during early stages of the plant cycle, since this early stage yields the most valuable data. These mechanisms worked with a nitrate sensor developed by the Iowa State collaborators. 

Researchers deployed the robot at Curtiss Farm in Ames, Iowa, last summer and achieved high success rates in stalk detection, stalk grasping and sensor insertion. They plan to return this year with an improved autonomy stack that includes navigation and a revamped gripper design.

The paper on autonomous crop monitoring was accepted into the IEEE Robotics Automation and Letters journal and the team will present their work at the 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) in October. 

This research is open-source, with all information available on the project homepage.