IEEE Robotics & Automation Magazine - September 2021 - 44

and status is a reliable way to ensure that a message sent by
one robot is indeed received by the other.
The mission framework used on both robots is based on
ros_task_manager [37], a task scheduler developed for ROS
that is particularly easy to use and allows for combining multiple
behaviors with elements running in sequence or in parallel,
eventually interrupting each other. This framework is
based on tasks implemented in C++ that are combined into
complex missions implemented in basic Python. Figure 15 is
an example of a coordinated mission.
In-Field Intervention: The Collaborative
Weeding Use Case
The main use case addressed in the Flourish project is the
collaborative weeding application (Figure 1). The UAV flies
over the field, running the navigation and planning algorithms
discussed in the " UAV Localization and Mapping "
and " UAV Mission Planning and Navigation " sections while
analyzing the weed pressure using the classification algorithms
presented in the " Crop and Weed Detection " section.
The UGV is alerted to high weed-pressure areas using the
coordination framework described in the " UAV-UGV Mission
Coordination " section. Thus, the UGV starts to move
toward the selected areas, running the algorithms discussed
in the " UGV Global Positioning and Mapping " and " UGV
Position Tracking and Navigation " sections. In the following,
we describe the tools (see the " UAV Localization and Mapping "
section) and methods (see the " Weed Tracking " section)
used for actual weed treatment with the possible
agronomic impacts reported in the " Agronomic Impacts in
Sugar Beet Crops " section. We successfully tested the whole
pipeline in a public demonstration during a dissemination
event held near Ancona (Italy) in May 2018.
Selective Weed Removal
The weed-intervention module [Figure 3(a)], whose perception
system was introduced in the " Ground Vehicle " section,
includes further tools designed to address the targeted weed
treatment: a weed stamping tool and a selective spraying tool
[Figure 3(b)]. The stamping tool is composed of 18 pneumatic
stamps arranged in two ranks. All of the stamps are individually
controllable, and highly precise positioning is ensured by
allowing only 1 DoF for the positioning across to the driving
direction. The spraying tool is positioned in the back. It is
assembled out of nine nozzles, individually controlled by offthe-shelf
magnetic valves.
Both weeding tools are controlled using a scalable programmable
logic controller. Modules requiring more computational
resources, i.e., weed detection and tracking, are
implemented on a computer dedicated to the weed control
running Linux and ROS.
The bolt of the stamps has a 10-mm diameter, whereas
the footprint of a sprayer is 30 mm when set in the lowest
position, as in our experiments. To actually treat a weed
while the robot is moving is a time-critical part of the process
because a small delay can lead to a position error at the centimeter-level,
which is large enough to miss a small weed. In
our experiments, the decision of which tool to use on which
weed is based only on a size criterion: large weeds are
sprayed while small weeds are stamped.
Weed Tracking
The main challenge in weed tracking with nonoverlapping
multicamera systems [Figure 3(b)] (see the " Weed-Intervention
Module " section) is to deal with the high variance
delay between the instant when the image of the first camera
is acquired and when a target is sensed by the detection
system. To address this issue, a novel tracking system was
developed. The inputs were the images and the coordinates
of the targets given by the classifier (see the " Crop and
Weed Detection " section) in the images of the detection
camera (see Figure 3). The outputs were the trigger time
and position for the actuators. The main steps are illustrated
in Figure 3:
1) The intracamera tracking module estimates the camera
pose and the 3D scene map using VO direct methods.
2) After receiving the delayed classification results and scene
structures, the object initializer and updater module creates
the templates of the received objects, propagates their
updated poses, and accumulates their labels.
UGV
Takes Off
Detects the Areas
of Interest
Completes the
Field Survey
Receives a
Target From
the UAV
Drives to the Target
Row, Enters It,
and Drives to
the Target Area
Figure 15. An example of a coordinated mission.
44 * IEEE ROBOTICS & AUTOMATION MAGAZINE * SEPTEMBER 2021
UAV
Lands at the
Meeting Point
User Interface
Completes the
Area-of-Interest
Treatment
Completes
Field Treatment
Goes to the
Meeting
Point
User Starts the Mission
Mission is Complete
IEEE Robotics & Automation Magazine - September 2021

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - September 2021
