טכניון מכון טכנולוגי לישראל
הטכניון מכון טכנולוגי לישראל - בית הספר ללימודי מוסמכים  
M.Sc Thesis
M.Sc StudentNir Omer
SubjectModeling of Tree Structure Using Robotic Arm for Pruning
DepartmentDepartment of Civil and Environmental Engineering
Supervisor Professor Raphael Linker
Full Thesis textFull thesis text - English Version


Abstract

Pruning orchard trees is a labor intensive operation performed at least annually in most orchards and requires skilled manpower. There is a high motivation for the development of an automated pruning system to reduce the manpower needed for the task.
An automated pruning system includes the following sub systems and tasks: 3-D scan system, an algorithm for building an accurate model of the tree from the scan data, path planning for moving the robotic arm from one pruning point to the next, and a pruning device mounted on the end-effector of the robotic arm.

This work deals with the development of a semi-automatic robotic pruning system. The system is to perform automated scanning, modeling, and pruning of a tree, while a human operator is still needed to decide on agronomical aspects of the pruning operation. The system consists of a color camera, laser stripe projector, and electric shears mounted on the end-effector of a robotic arm. The system makes use of close range laser telemetry for the generation of a point cloud of the tree skeleton. Using an algorithm developed in this study, we built an accurate model of the tree to be pruned: In order to aggregate the point cloud into clusters corresponding to branches, the original point cloud was transformed into a "vector point cloud" that contained not only the position of each point but also the direction of the cloud in its vicinity. This direction was estimated by minimizing the distances between the neighboring points and a straight line passing through the point under investigation. In cases where the simple least square procedure did not produce reliable results (such as near branching-off areas), a custom procedure based on RANdom Sample Consensus (RANSAC) was used. The final skeleton of the tree was obtained by clustering the "branch points" into branches, connecting the branches and determining the branches' hierarchy, i.e. which branch would fall as the result of removing another branch. The obtained tree model is then shown graphically to a human operator which makes the decision regarding which branches must be pruned based on agronomist considerations. Finally a simple path planning algorithm is used to guide the robotic arm with the electric pruning shears mounted on its end-effector.

Small-scale tests with seven tree segments were conducted in a laboratory environment and demonstrated the ability of the system to perform robotic pruning.