Aerospace and Electronic Systems Magazine August 2017 - 23

a quadratic O(N2) complexity as in EKF SLAM. The equivalence
of the partial sampling and the full sampling methods is formally
proved by the authors in [22].
The computational cost of SLAM solutions is significantly
affected by the number of landmark features, which typically
grows as the vehicle explores more regions. One of the effective approaches to dealing with this is the compressed form of
SLAM [15], [22] in which a large map is partitioned into a local
and global map, and only the local map is updated with a quadratic complexity of O N L2 , with NL being the size of the local
map, which is typically NL << N. At the same time, the local-toglobal correlation information is accumulated, or compressed, and
propagated to the global map only when the vehicle crosses the
local boundary.
Most SLAM work has been for land vehicles or stabilized
drones in which the motion can be approximately planar. If the
vehicle undergoes full 6DOF motion, inertial sensors are necessary
to measure the high dynamics of the vehicle with high data rates.
Because vision systems can also provide the full pose (the orientation and position with scale ambiguity in the case of a monocular
system), their integration within the SLAM framework has also
been actively investigated [23]-[29]. However, the fusion of full
SLAM with the GNSS and inertial system has not been studied in
depth [30]. Among the reasons is that the robotics community is
more interested in the standalone capability of the SLAM system
without relying on the GNSS information, as well as the high computational cost hindering the integration. However, the mapping
capability in the GPS coordinates (for example, latitude and longitude) becomes increasingly important to using the geographical
infrastructure system (GIS) database and Google Map.
This article presents tightly coupled integration of SLAM and
the GNSS and inertial system, extending previous work on loosely
coupled integration [30] and compressed and unscented SLAM for
3 degrees-of-freedom vehicles [22]. Figure 1 illustrates the simulation output of the fusion system showing the observations of the
GNSS pseudorange and pseudorange rates, as well as the downlooking visual feature measurements. The key benefits of this approach are the continuous and reliable calibration of the receiver
clock and IMU errors even with only one satellite in view, as well
as offering the mapping in the GPS coordinates. The main contributions of this article are as follows:

( )

AUGUST 2017

Integration of the GNSS pseudorange and pseudorange rate
with the inertial SLAM with a reduced number of satellite
vehicles
Compressed implementation of unscented GNSS and inertial
SLAM in a full 6DOF scenario, achieving O N L2 complexity

( )

The overall architecture of the fusion system is shown in Figure 2, in which a compressed-unscented filter is used with 17 inertial and receiver states and 3N map states, with N being the number
of features. The raw GNSS pseudorange, the GNSS pseudorange
rate, and the IMU measurement are directly fused in the filter, thus
forming a tightly coupled architecture. A vision system provides
the (triangulated) range, bearing, and elevation measurements to

Figure 1.

Tightly coupled GNSS and SLAM navigation in a simulation environment. The vehicle undergoes 6DOF motion following two racehorse
tracks after taking off and fuses GNSS pseudoranges (shown as three
magenta lines toward the satellite vehicles) and ground features (shown
as a red line down to a feature). The compressed filter partitions the
area into a local region (a rectangular box beneath the vehicle, with map
uncertainty ellipses in blue) and a global one (outside the box, with map
uncertainty ellipses in red). The local region is redefined whenever the
vehicle crosses the boundary, and because of the compressed approach,
the computational time can be managed in a way suitable for real-time
applications.

IEEE A&E SYSTEMS MAGAZINE

Table of Contents for the Digital Edition of Aerospace and Electronic Systems Magazine August 2017