Skip to main content
ARS Home » Midwest Area » East Lansing, Michigan » Sugarbeet and Bean Research » Research » Research Project #438383

Research Project: Development of a Control and Planning System for an Apple Robotic Harvester

Location: Sugarbeet and Bean Research

Project Number: 5050-43640-003-004-S
Project Type: Non-Assistance Cooperative Agreement

Start Date: Jul 16, 2020
End Date: Jul 15, 2025

Objective:
The overall objective of this cooperative agreement is to develop a control and planning system to coordinate with the computer vision system and the robot manipulator to pick apples from trees. The specific objectives are to: 1. Develop and validate control algorithms for fast picking of apples by a single-arm robot. 2. Develop a systems control and algorithms for planning and coordinating multiple arms of an apple robotic harvester to pick apples.

Approach:
1. Validate the single-arm control algorithms in lab. The robot arm control performance will be evaluated in terms of picking speed and end-effector reaching accuracy on a set of predefined locations in the workspace. Measures of success include speed of picking-and-return within 3 seconds and the maximum end-effector positioning error within 2 cm. 2. Improve the existing computer-vision based apple detection and localization algorithm. Building on the existing computer vision-based perception system, algorithms will be developed to increase the computational efficiency of the deep neural network-based detection, e.g., by simplifying the existing network structure without losing much accuracy for higher processing rate in frames per second. Methods will be explored further to enhance fruit localization accuracy. Measures of success include a detection rate of >95% in lab and >90% in orchard, and a localization error within 2 cm. 3. Integrate the entire system and evaluate its performance in orchard. Full integration of the perception system and the robotic arm, will be performed, and its performance in orchard will be evaluated in terms of detection and detaching rates. Missed detection and failed detachment scenarios will be recorded and analyzed to improve the system. Measures of success include >90% detection rate and >85 picking rate. 4. Develop planning and control algorithms to enable the operation of multi-arms. Based on the framework for single arm, corresponding planning and control algorithms will be developed to enable cooperative operations between multiple arms. Planning and control algorithms will be developed and validated in simulation and two robotic arms will be assembled and integrated. Measurements of success include planning and control algorithms for efficient and collision-free apple picking in simulation, as well as an assembled two-arm system for further lab and orchard validations in the following year.