[Paper Review] Flower Interaction Subsystem for a Precision Pollination Robot

Jared Strader, Jennifer Nguyen+9 more|arXiv (Cornell University)|Jun 21, 2019

Plant and animal studies32 references4 citations

TL;DR

This paper presents a fully autonomous robotic system with a specialized flower interaction subsystem for precise pollination of individual small flowers. Using a robotic arm with a custom end-effector and depth camera, the system achieves 93.1% flower detection accuracy and 76.9% pollination success rate on high-fidelity artificial flowers, demonstrating a novel integration of perception, pose estimation, and controlled manipulation for precision agriculture robotics.

ABSTRACT

Robotic pollinators not only can aid farmers by providing more cost effective and stable methods for pollinating plants but also benefit crop production in environments not suitable for bees such as greenhouses, growth chambers, and in outer space. Robotic pollination requires a high degree of precision and autonomy but few systems have addressed both of these aspects in practice. In this paper, a fully autonomous robot is presented, capable of precise pollination of individual small flowers. Experimental results show that the proposed system is able to achieve a 93.1% detection accuracy and a 76.9% 'pollination' success rate tested with high-fidelity artificial flowers.

Motivation & Objective

• Address the growing need for precision agricultural robotics to compensate for pollinator decline and labor shortages.
• Develop a system capable of autonomous, precise pollination of individual small flowers in controlled environments like greenhouses.
• Overcome limitations in existing robotic pollination systems by integrating high-precision perception, pose estimation, and controlled manipulation.
• Enable reliable, repeatable pollination without reliance on bees or manual labor, especially in non-traditional environments such as space or growth chambers.
• Establish a foundation for future automation of other plant-specific tasks like harvesting and phenotyping.

Proposed method

• Utilizes a ground robot (BrambleBee) equipped with a KINOVA JACO 2 robotic arm and Intel RealSense D435 depth camera for environmental perception.
• Employs AruCo markers on the end-effector to enable hand-eye calibration and precise pose estimation of the flexible end-effector relative to the camera.
• Applies a lookup table for inverse kinematics to compensate for the flexible nature of the end-effector, enabling accurate pose control.
• Uses image processing and 3D reconstruction to detect flowers, estimate their 3D pose (position and orientation), and map obstacles in the workspace.
• Implements a planning and control pipeline that accounts for reachability constraints and avoids collisions during motion.
• Integrates a custom-designed end-effector with linear actuators to extend and contact the anthers of flowers after precise positioning.

Experimental results

Research questions

• RQ1 Can a fully autonomous robotic system achieve high-precision pollination of individual small flowers in a controlled environment?
• RQ2 What level of accuracy can be achieved in detecting and estimating the 3D pose of small, complex flowers using RGB-D sensing and computer vision?
• RQ3 How does the flexible end-effector's kinematic behavior affect control precision, and can it be compensated using a lookup table?
• RQ4 What are the primary failure modes in autonomous flower contact, and how do pose estimation errors and 'blind driving' impact pollination success?
• RQ5 To what extent can a robotic system replicate the precision of natural pollinators in a non-biological setting?

Key findings

• The system achieved a 93.1% flower detection accuracy, identifying 134 out of 144 artificial flowers correctly with only two false positives.
• The pollination success rate was 76.9%, defined as successful contact with the flower and extension of the linear actuators to touch the anthers.
• The main causes of failure were incorrect flower orientation estimates and 'blind driving' during approach, where the depth camera lost sight of the flower.
• In failed attempts, the end-effector missed the flower center by no more than 2 cm, indicating high positional accuracy when tracking is maintained.
• Pose estimation errors and loss of visual tracking were identified as the primary contributors to missed pollination attempts.
• The system demonstrated the first integration of full autonomy with high-precision manipulation for small-flower pollination, setting a benchmark for future robotic agriculture systems.