IEEE Robotics & Automation Magazine - September 2017 - 103

the YUV color space. The markers are printed on a canvas
fabric to minimize specular reflections from direct sunlight.
These allow reliable detection of the markers despite nonuniform shadowing and changing illumination across day
and night.
User Interface
With a large fleet of robots, it becomes difficult for a single
human operator to pay close attention to each robot.
Effective multirobot coordination can reduce this workload and provide a more persistent autonomous solution.
Our robot system uses a standard coverage planner that
uses boustrophedon decomposition (lawnmower pattern)
described by Choset [5] and is implemented as described
in Richards et al. [26] with coverage paths constrained to
follow existing straight rows at a known orientation. The
planner allocates each robot a set of rows to cover so that
they each travel approximately the same distance. In this
article, the coverage planner also schedules periodic docking to replenish consumables.
An example screenshot of this user interface is shown in
Figure 7. It displays real-time information about the state and
pose of each robot using icons overlayed on Google Maps. In
online mode, the map images are loaded from a Google
server; in offline mode, from a local cache. When an Internet
connection is unavailable and there are no map images in the
local cache, the interface displays a white background. The
icons also include the remaining power level and spray tank
level. Each robot's completed and future paths are displayed
on the user interface.
The interface allows the user to define the field boundaries and roads between fields and refuelling locations by
either clicking on the map or by entering a latitude and
longitude. Since the fields and roads rarely change, each
farm can be saved and loaded from a server. The interface
sends start, stop, plan, and replan commands to the robots
via the coverage planner over a local area network. The interface was developed using the cross platform Unity3d game
engine as it allows each user to monitor the farm from a
native computer application, web browser, or mobile
device, as shown in Figure 7.
Results
To test each part of the system, we performed numerous
experiments over three months of operation in the field. Most
results in this section are from our project's final integrated
autonomous two-day coverage task in a 59-ha broad-acre
field using the AgBot and SwarmBot and four simulated
robots. This 5-h experiment was split over two days due to
battery capacity and recharge time limitations. A video showing our experiment can be seen at https://www.youtube.com/
watch?v=9K5AiOoFkis.
We used the user interface shown in Figure 7 to configure
the boundary of the field, the location of the road and docking station, and the direction of the crop rows. During the
experiment, the AgBot sprayed water on the field to simulate

1,000 L Tank

Dock Camera
Valve

Dock eBox

2 DoF
Arm

Stereo AR Marker

Nozzle

AR Marker

Funnel

Level Sensor
Tank
Figure 6. The AgBot in the docked position about to be refilled with
liquid. The gravity-fed system has the liquid flow from the 1,000-L
tank through the nozzle into the robot's funnel. The robot's forwardfacing camera uses the stereo AR markers for relative localization
for guidance into the dock. The overhead dock camera uses the AR
marker located on top of the robot to control a 2-DoF robot arm to
position the discharge nozzle over the funnel. When the robot is in
position, it communicates with the dock's electronics box, which
controls the valve. The level sensor determines when the robot's
tank is full.

Road
Boundary
Completed Path
Future Paths
Refill Stations
Robot
Figure 7. A screenshot of the user interface during field trials. The
interface is overlaid on a satellite image of the farm. Each green
circle represents an active robot, while the red shows a robot
that has completed its task. Each bar above the robot shows the
battery (blue) and spray (green) levels. Paths are also drawn for
each robot with completed paths (red) and future paths (blue).
(Image courtesy of Google Maps.)

herbicide application and autonomously docked to refill its
tank five times. The automated docking system was able to
guide the robot into the correct docking location, with the
robot's tank topped up each time with approximately 100 L
SEPTEMBER 2017

IEEE ROBOTICS & AUTOMATION MAGAZINE

103

https://www.youtube.com/

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - September 2017