IEEE - Aerospace and Electronic Systems - June 2021 - 9

Credit: Image licensed by Ingrahm Publishing
operates. For these reasons, Baldini et al. [16] used a learning
approach to reduce drift. To reduce drift, employing
SLAM is possible directly on a UAV as has been demonstrated
in [17]. However, some operating environments
(e.g., a long corridor or tunnel) may affect the reliability of
loop-closures, crucial for SLAM.
To address some of these challenges, the approach presented
in this article proposes to use active sensors on the
UGV to localize the UAV with respect to the UGV in a
manner that is not prone to drift, and with modalities that
will work in a variety of lighting conditions; in particular,
following on the approach considered in our initial work
presented in [12]. With this UAV/UGV teaming configuration,
the UGV has the advantage ofadditional load capacity
and longer endurance to easily carry many sensors along
with the significant computational power needed to use
them in near real-time. In this scenario, the UGV is used to
map the environment as well as for estimating the state of
the UAV within the map, whereas the UAVs mobility is leveraged
to perform the search task. However, when adopting
this approach, due to the nature ofthe UGVs sensors' uncertainty,
as well as the potential for nonoverlapping sensor
fields-of-view, the localization uncertainty of the UAV is
affected by the chosen flight path. As such, the focus ofthis
article is to develop a path planning algorithm for the UAV
that takes into consideration both the exploration needs and
the localization uncertainty ofthe UAV.
Considering the position uncertainty in the planning
algorithm consists of finding a configuration in the belief
space such that motion increases information from sensing,
and reduces pose uncertainty of the robot with respect to
the environment [18]. Planning under uncertainty leads to
the use of a dual-layer architecture; where the first layer
predicts all the possible outcomes and the second layer
determines the best action to take [19]. For mobile robot
path planning, Prentice et al. [20] presented a revised probabilistic
roadmap (PRM) [21], called Belief RoadMap,
where they fuse the predicted position estimated uncertainty
into the planning process. In this article, the trajectory
was designed with respect to the location of
predeployed ranging radio beacons in order to reduce the
JUNE 2021
position uncertainty while traversing from a start to a goal
location. A belief space planning problem is often formulated
as a partially observable Markov decision process and
Van De Berg et al. [22], similar to the approach presented
in this article, used an extended Kalman filter (EKF) to
approximate the robot beliefdynamics.
Different from prior belief space planning research,
our approach has a required exploration goal and seeks to
accomplish it such that the UAVs localization uncertainty
is maintained within an acceptable level during the operation.
This article contributes a new search planning algorithm
that fuses the capabilities of two robots (UAV/
UGV) to balance between environment exploration tasks
while reducing the localization error for the UAV along
the selected trajectory. This article substantively builds
upon our previous work [12], where the initial design and
evaluation of the UAV localization system is presented.
The rest of this article is organized into five sections.
The section " System Description " provides a description
of the system where the assumptions and constraints of the
problem are presented; " Planning Algorithm " explains the
path planning algorithm; " Simulator Overview " presents
the simulation environment; " Results " discusses the results
and " Conclusions " offers conclusions and future work.
SYSTEM DESCRIPTION
ASSUMPTIONS AND CONSTRAINTS
In this article, the UAV and UGV systems are modeled
according to the developed physical hardware system that
is described in [12].
As shown in Figure 1, a subterranean environment
would block a possible positioning system, such as GNSS,
and for this reason, different sensors need to be used for
localization purposes. The UAV is assumed to carry a
downward-facing camera to perform the search, a LIDARLite
altimeter that provides readings at 5 Hz, and an inertial
measurement unit (IMU) with updates at 50 Hz. The
UGV is assumed to be equipped with a 128 channel 3-D
IEEE A&E SYSTEMS MAGAZINE
7
IEEE - Aerospace and Electronic Systems - June 2021

Table of Contents for the Digital Edition of IEEE - Aerospace and Electronic Systems - June 2021
