Design of Stickbug: a Six-Armed Precision Pollination Robot

TL;DR

This work tackles the decline of natural pollinators by introducing Stickbug, a six-armed, multi-agent precision pollination robot designed for greenhouse environments. The system fuses swarm-like parallelization with centralized refereeing to manage six manipulators on a single drive base, leveraging a Kiwi holonomic drive, a tall mast, and perception via depth cameras and YOLOv8. The authors demonstrate initial viability, achieving over $1.5$ pollination attempts per minute at roughly $50\%$ success on artificial bramble flowers, and they publish the bramble dataset along with design and software files. This design advances scalable, autonomous pollination in controlled agriculture and provides open resources for further development and deployment in greenhouses.

Abstract

This work presents the design of Stickbug, a six-armed, multi-agent, precision pollination robot that combines the accuracy of single-agent systems with swarm parallelization in greenhouses. Precision pollination robots have often been proposed to offset the effects of a decreasing population of natural pollinators, but they frequently lack the required parallelization and scalability. Stickbug achieves this by allowing each arm and drive base to act as an individual agent, significantly reducing planning complexity. Stickbug uses a compact holonomic Kiwi drive to navigate narrow greenhouse rows, a tall mast to support multiple manipulators and reach plant heights, a detection model and classifier to identify Bramble flowers, and a felt-tipped end-effector for contact-based pollination. Initial experimental validation demonstrates that Stickbug can attempt over 1.5 pollinations per minute with a 50% success rate. Additionally, a Bramble flower perception dataset was created and is publicly available alongside Stickbug's software and design files.

Figures (7)

• Figure 1: Displays Stickbug, a six-armed precision pollination robot tasked with autonomously navigating, mapping, and pollinating bramble flowers in a greenhouse. Each arm acts as an independent greedy agent that a referee oversees to resolve inter-agent conflicts
• Figure 2: Displays Stickbug's lightweight 3D-printed planar manipulator on the vertical mast left rail (a), pollination end-effector featuring a Realsense D405 depth camera for flower identification and felt tip for pollination (b), and kiwi drive base with a 3D Velodyne LiDAR for navigation and mapping (c).
• Figure 3: Stickbug's software architecture is distributed across three main agents: the drive base for navigation and mapping, the manipulators for flower identification and pollination, and the referee for inter-agent conflict detection and resolution.
• Figure 4: Shows Stickbug navigating greenhouse rows, leveraging LIO-SAM liosam2020shan for localization and mapping, and Falco zhang2020falco for planning. Green areas are traversable; red indicates untraversable space, and grey areas are outside the scope of local planning. Small spheres mark waypoints, with the current target enlarged in purple. The dark blue line displays Stickbug's path.
• Figure 5: Stickbug employs YOLOv8 Ultralytics_YOLO_2023 for flower detection, illustrated by bounding boxes, and uses a custom binary classifier to assess the visibility of flower centers for pollination (a). Part (b) presents the confusion matrix of the Flower Center Visibility classifier, with part (c) detailing the classifier's performance metrics.