Author ORCID Identifier

https://orcid.org/0009-0002-7609-8086

Semester

Summer

Date of Graduation

2024

Document Type

Thesis

Degree Type

MS

College

Statler College of Engineering and Mineral Resources

Department

Lane Department of Computer Science and Electrical Engineering

Committee Chair

Yu Gu

Committee Co-Chair

Jason Gross

Committee Member

Jason Gross

Committee Member

Gianfranco Doretto

Abstract

The global decline in pollinator populations poses a significant threat to agriculture, motivating the development of robotic pollination systems. Previous works demonstrated successful robotic pollination of bramble flowers using visual servoing; however, pollination was limited to specific flower orientations. As such, the objective of this work is to develop a robotic pollination system that is capable of pollinating a wider range of orientations.

This research introduces an imitation learning-based framework for robotic pollination that positions the manipulator to view chosen flowers in specific orientations. The developed model leverages object detection (YOLOv8) to identify individual flowers and a convolutional neural network (CNN) to produce manipulator actions. To generate training data, this work proposes a novel methodology where expert trajectories are recorded in reverse. This method involves sampling random points around a known flower pose and recording demonstrations from the goal configuration back to these points. This approach frees the expert from manually managing the camera's view to ensure it sees the flower, while also allowing the data collection process to be autonomous.

In the experimental evaluation, the model successfully positioned the end-effector within an acceptable range in 88% of the test cases (44 out of 50 trials), achieving an average positioning time only 38.7% slower than the theoretical optimum. Notably, due to the method's use of only relative and current state information, this approach exhibits the unintended property of effective object tracking. This property is very useful for continuous monitoring and adjustment during pollination tasks. These results demonstrate the potential of using an imitation learning-based approach for robotic pollination that can contribute to the future sustainability of global agriculture.

Included in

Robotics Commons

Share

COinS