IEEE Robotics & Automation Magazine - June 2020 - 161

How does the operator's perception of usability and workload change over levels of swarm autonomy?
● How do objective measures of user cognitive workload
change over levels of autonomy?
To answer these questions, we created two control modes
for our swarm that differ by degree of robotic autonomy. We
monitored the operator's cognitive load via pupillometry during task performance and subsequently assessed the user's
subjective experience. This article presents the main components and contributions of our experimental field setup and
discusses the HSI results obtained at PANGAEA-X.
●

Decentralized Control
In safety and mission-critical multirobot applications, it is unacceptable to end a mission because of a single unit failure. For
example, when a group of rovers explores a lava tube on the
moon, if the leader robot-the one coordinating the mission
and allocating the tasks-gets stuck, the team's exploration
potential will be greatly reduced. In an emergency response scenario where a team of aircraft is searching for victims, losing the
link to the control station responsible for trajectory planning
could result in the loss of life. Decentralized paradigms, which
use only local information for control and communication, can
mitigate these issues while rendering a team more adaptable to

dynamic environments. For instance, failure of a single robot or
a broken communication link would not compromise the mission, as the remaining robots can collaboratively reorganize their
activities to cover the search region.
For these reasons, we consider decentralized control to be
an ideal solution for space applications, as previously shown
for the formation control of multiple collaborating spacecraft
[11] or to synchronize the actuators of a Martian ground
robot [12].
Figure 2 shows the details of our implementation for these
experiments. For interrobot communication, we rely on Xbee
mesh modules and, for localization, on GPS. Because the
robots' coordination relies solely on interrobot distance and
bearing, localization is required only for user inputs. GPS can
be substituted with other available means (for example, ultrawide-band, landmarks, or camera-based mapping).
Swarm Behavior Design
Developing sophisticated, fully decentralized behaviors is very
challenging, as they can rely only on limited information and
local interactions. To simplify our implementation, we use
Buzz, a domain-specific programming language for robot
swarms [13]. Buzz provides special constructs for robots
to share data with the swarm (a technique called virtual

GPS,
GLONASS
Ubuntu 16.04 LTS
DJI N1
(PixRacer)

XBee <1 GHz
RF Module

NVIDIA Jetson TX1 (TX2)

Ground Station via
802.11 × 5-GHz Wi-Fi
and XBee RF Module
2.4-GHz Controller
Figure 2. The control architecture (from left to right): the localization system and (backup) remote controllers interface with the UAVs'
onboard computers and flight controllers (FCUs) to execute the decentralized behavioral Buzz script. The entire fleet executes the same
script, interfacing with the FCU and communication device (Xbee) through the ROS and Mavlink protocol. RF: radio frequency. (Source:
Freepik from Flaticon.)

JUNE 2020

IEEE ROBOTICS & AUTOMATION MAGAZINE

161

IEEE Robotics & Automation Magazine - June 2020

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - June 2020