Source: OREGON STATE UNIVERSITY submitted to
ROBOTIC PRUNING IN MODERN ORCHARDS
Sponsoring Institution
National Institute of Food and Agriculture
Project Status
NEW
Funding Source
Reporting Frequency
Annual
Accession No.
1023398
Grant No.
2020-67021-31958
Project No.
OREPRU20
Proposal No.
2019-06458
Multistate No.
(N/A)
Program Code
A1521
Project Start Date
Jul 15, 2020
Project End Date
Jul 14, 2023
Grant Year
2020
Project Director
Davidson, J. R.
Recipient Organization
OREGON STATE UNIVERSITY
(N/A)
CORVALLIS,OR 97331
Performing Department
CE Indust/Mnfctr Engineering
Non Technical Summary
Pruning - a critical perennial operation required to maintain tree health and produce high yields of quality fruit - is one of the most labor-intensive orchard activities in the production of high-value tree fruit crops. As the fresh market tree fruit industry continues to face the challenge of an uncertain labor force, the development of robotic technologies that are able to perform labor-intensive field operations - like pruning - will play a crucial role in its long-term sustainability. Automating selective pruning is a complex problem requiring high-resolution sensing, complex manipulation, and advanced decision making on determining which branches to prune. The specific research objective of this project is to develop a robotic system for autonomous, dormant pruning of fruit trees. To accomplish our objective, we have formed an experienced, interdisciplinary project team with expertise in horticulture, agricultural mechanization, computer graphics, and robotics.Our approach builds on preliminary investigations, which indicate that machine vision systems can be used to analyze the manual pruning process to i) formulate pruning rules, and ii) identify pruning locations for autonomous pruning of fruit trees. Specifically, our approach is to first build high-quality digital models of the trees offline using a combination of artificial intelligence techniques and human intervention. Second, we use the knowledge gained from studying the manual pruning process to formulate pruning rules, which can be 'practiced' on the digital models and evaluated for their effectiveness. Third, we will use state-of-the-art machine learning algorithms to map pruning plans back to the real world in order to reduce onsite computational demands. Fourth, we will design a custom end-effector that can localize itself in real time in order to safely and reliably perform the pruning cuts. We expect that the proposed robotic pruning system will help sustain the global competitiveness of the US tree fruit industry. In addition, consumers will benefit through increased access to premium quality fruit.
Animal Health Component
0%
Research Effort Categories
Basic
25%
Applied
50%
Developmental
25%
Classification

Knowledge Area (KA)Subject of Investigation (SOI)Field of Science (FOS)Percent
40211192020100%
Goals / Objectives
The long-term goal of our research team is to increase yields of premium quality fruit, improve production efficiency, and reduce dependency on human physical labor through automation and robotics. The specific goal of this research project is to establish the feasibility of using a robotic system for dormant pruning of fruit trees, substantially reducing the human labor involved. This project has the following three objectives:Create a collaborative human-robot training method for intelligent pruningCreate digital tree growth models that can be used both for developing pruning rules and offline path planningCreate an integrated perception and manipulation framework for fruit tree pruning
Project Methods
We will analyze manual pruning activities performed by domain experts (e.g. experienced workers, growers, and horticulturalists) to understand pruning patterns. We will then create an interactive, digital traning evnironment where automated rule sets can be iteratively trained and evaluated, using the manual data for training.We will use empirical data gathered throughout the growing season - via high-resolution, labeled scans of the trees - as input to the development of a synthetic digital tree model. After learning parameters that yield correct tree growth models, we will use the digital models to improve the pruning rules by predicting the resulting growth patterns.Using the digital models and pruning rules, we will plan efficient pruning paths (offline). We will develop a perception system that maps this plan back to the real world, grounding pruning points in the robot's coordinate frame. We will also design a custom end-effector that is capable of performing the final localization and cut using a lightweight perception system that is mounted on the end-effector itself.After simulations, functionality tests, and laboratory experiments, we will conduct field evaluations of the integrated robotic system at both WSU's Roza Research Orchard and commerical orchards. We will assess the system's accuracy by comparing the results of robotic pruning with that of trained workers. Performance measures to be used for the comparison include the percentage of successful pruning occurrences, Pruning Branch Proportion, and resulting branch spacing and branch length. We will use overall cycle time to evaluate pruning throughput. Finally, the horticultural effects of pruning cuts (e.g. vegetative growth and distribution of fruiting sites) will also be assessed after robotic pruning.