Vision-Guided Flexible Robots for Harvesting the Interior of Tree Crops
Motivation: The agricultural industry is currently being impacted by increasing labor costs, a decrease in interest in farm labor among younger generations, and impacts from crises such as the COVID-19 pandemic. With advances being made in robotics, automation can provide a feasible solution to these challenges. Currently, most advancements of these kinds have been more towards simpler, ground-level tasks, leaving more complex tasks lacking innovation. Harvesting tree crops is one of these complex tasks as many groves consist of unstructured canopies where traditional harvesting robots and techniques are not ideal.
Project Highlights:
This project’s aim is to develop a soft continuum robot that is capable of navigating within unstructured environments and harvest fruits inside the canopies of trees under the real-time guidance of computer vision. The progress made thus far is the development of a robot manipulator that has demonstrated dexterity and follow-the-leader motion capabilities.
Figure 1. Macroscale continuum robot developed for agricultural applications: (a) cable-driven, telescoping manipulator mounted onto a mobile base, (b) design of robot manipulator’s major components, and (c) images showcasing actuation systems and interactions of coupled components.
The manipulator design consists of telescoping, continuum tubes that are housed within a framing structure, capable of two variable curvatures. While the tubes are being extended out of or retracted into the frame, cables routed throughout the tubes are displaced via cable-driven actuation systems to achieve the bending of the tubes. A track and roller mechanism is used to allow for the tubes to be telescoping and concentric with each other. Additionally, rigid linear rods mounted within the frame ensures that the retracted tube segments remain straight.
Using a vision-based measurement system, the robot’s motion was characterized to allow for mapping between the manipulator’s task, configuration, and actuator spaces. Experiments built on this characterization are conducted in the lab to test the manipulator’s capability for follow-the-leader motion and achieving a desired end tip position within the task space. The follow-the-leader motion and end tip position average errors were found to be 9.5 mm and 8.8 mm, respectively, which are considered to be within the allowable error range for harvesting produce. By mounting the manipulator onto a mobile base, the robot demonstrated its ability reach targeted fruits by traversing the rough terrain and avoiding obstacles within the canopy.
Figure 2. Characterization approach and experiments/demonstrations performed: (a) vision-based measurement system, (b) meshed surfaces for mapping between different spaces, (c) follow-the-leader experiment results, and (d) demonstration of robot in citrus grove.
Submitted Publications:
R. Deuling, A. Akbari, S. Saini, L. Tian, B. Qi, M. S. Asif, and J. Sheng, “Macroscale Continuum Robot for Manipulation in Confined Agricultural Environment,” (submitted).