Using Geographic Information Systems (GIS) for

UAV Landings and UGV Navigation

Nathan Rackliffe, Holly A. Yanco, and Jennifer Casper
The MITRE Corporation, Bedford, Massachusetts 01730-1420
nrackliffe@mitre.org, hyanco@mitre.org, jcasper@mitre.org
AbstractUnmanned air and ground platforms are not cur-
rently designed to process contextual information (e.g., land
use, building types and time of day) to assist in landing and
navigation. Instead, a human operator often provides manually-
gathered and synthesized contextual input through control com-
mands. This paper describes research that explores the integra-
tion of geographic information system (GIS) data with sensor
data to enable (1) unmanned aerial vehicles (UAV) to locate safe
emergency landing locations without operator intervention, and
(2) unmanned ground vehicles (UGV) to incorporate contextual
GIS information for navigation. Informal evaluation of the
system revealed the potential to impact navigation and emergency
landings through the additional contextual information provided.
I. INTRODUCTION
The use of contextual information such as the location and
uses of buildings in the area can improve the autonomous
capabilities of unmanned systems. However, at this time,
unmanned air and ground vehicles are not currently designed
to process contextual information to aid in landing and nav-
igation. For unmanned aerial vehicles (UAV) operating in
the national airspace (NAS), it is critical to be able to nd
safe emergency landing zones that avoid people, houses, and
populated areas, particularly in cases where the UAV has lost
its link with the operator on the ground. For unmanned ground
vehicles (UGV), sensors and processing provide information
on physical objects, terrain and mapping, but not contextual
information about the area (i.e., whether a building is a school,
hospital, shopping mall, etc.); this contextual information can
be used to improve the performance of a UGV and increase
its autonomous capabilities.
Autonomous systems used by the military, including UAVs
and UGVs, require frequent monitoring and intervention from
human operators. Operators take into account any contex-
tual information they can gather. In addition, missions are
planned for these systems at varying levels of granularity.
Some systems are given an end goal, some are given way-
points, and others require additional detail which might include
teleoperation. Situations may arise during the execution of
these plans that require dynamic replanning. An example of
such an event for a UAV would be a failure requiring an
emergency landing; an example of such an event for a UGV
would be encountering a hazard. Contextual information such
as the time of day and type of environment could inform
Approved for Public Release: 11-0082. Distribution Unlimited.
navigation and emergency landing plans. Additionally, the use
of GIS data together with sensor data gathered by autonomous
systems could potentially improve the systems reaction to
anomalous situations that arise and improve navigation routes.
The objective of this effort is to explore the integration of
GIS data with sensor data and time of day information to
enable (1) UAVs to locate safe emergency landing locations
without operator intervention, and (2) UGVs to incorporate
contextual GIS information for navigation. Ultimately, the
combination of GIS maps that include precomputed proba-
bilities for preferred landing sites or navigation routes with
current sensor data will help inform and improve the operation
of UGVs and UAVs.
Preliminary evaluation of our system, which integrates GIS
and sensor data, reveals potential impact through more efcient
navigation and the ability to locate safe landing zones. This
effort builds the foundation for employing GIS data streams in
UAV landing and UGV navigation. The platform agnostic na-
ture of our effort, which focuses on the fusion and processing
of multiple layers of GIS data according to a specied mission
prole, could be used for any type of autonomous system. This
effort not only impacts the UAV and UGV elds, but can also
apply to other elds such as data analysis interfaces.
This paper presents an approach to incorporating GIS that
revolves around building probability and cost surface maps.
The architecture includes Environmental Systems Research
Institute (ESRI) ArcGIS components for GIS data processing.
1
Our system was designed to take in the GIS data, process it
according to mission parameters utilizing ArcGIS, and output
probability maps which can either indicate where to land or
navigate. ArcGIS Server supports various output formats. In
this paper, the maps are presented as heat-map style images.
An overview of the architecture is shown in Figure 1.
We have taken a platform-agnostic approach as the maps
produced may be used for UGVs, UAVs and other types of
existing systems. Three phases are needed to create the maps:
(1) data collection, (2) data conversion, and (3) geo-processing.
In our current system, steps one and two are manual; the
architecture executes step three.
II. RELATED WORK
A wide variety of GIS data, covering many different types
of features, is publicly available. The level of detail, age
1
ESRI ArcGIS Information: http://www.esri.com/software/arcgis/index.
html
978-1-61284-481-7/11/$26.00 2011 IEEE 145
Fig. 1. Overview of architecture incorporating components of ESRI ArcGIS.
of the data, and accuracy varies for each data set. Despite
the potential imperfections in the GIS data, the information
provides an additional level of detail to inform autonomous
systems. Although GIS data is typically two dimensional (2D),
three dimensional (3D) GIS data is being produced through
highly accurate 3D environment models. The authors in [1]
explored the use of 3D GIS data for robot localization. The
work described in the following paragraphs, as well as our
research, makes use of 2D GIS data.
The use of GIS data within the robotics eld has been
explored over the last decade and grown in recent years as
GISs have evolved.
2
One of the earliest uses of GIS for
robots was in the design of a mechanical guide machine
for the visually impaired [2]. Although the machine did not
have the traversing capabilities of a dog, it offered additional
navigation through dynamic map creation that incorporated
GIS for contextual information. The authors in [3] highlight
the importance of contextual information for mobile robot
navigation. Satellite and mapping imagery provided the means
to repair spatial inaccuracies for waypoint navigation.
A common use of GIS data is for robot localization in
outdoor environments. Advanced sensors enable perception
of the immediate surroundings, although context is unknown.
Correlating features or landmarks extracted from sensors with
GIS data allows aligning. GIS is useful in verifying robot
position in the world, in conjunction with GPS in some
2
An annotated bibliography may be viewed here: http://dusk.geo.orst.edu/
gis/student bibs/nehmer/
cases. The use of GIS for landmark localization to aid vehicle
navigation has been used by many researchers (e.g., [4][10]).
In [7], an approach for tracking a specic feature (roads) for
autonomous vehicle navigation is described. A road model is
used to determine if sensor data is picking up road edges; GIS
and GPS data are employed for verication and localization.
GISs provide an inherent means to share geospatial data.
The research in [11] takes an alternative approach to incorpo-
rating GIS. Instead of consuming the GIS data, GIS was used
as a means of sharing information on the status of a disaster
environment. The data retrieved from the robots employed
to explore the disaster area was transmitted to a GIS via a
designated data format over wireless communications link.
This information may then be consumed by other systems or
robots. This approach results in a view of the environment that
is dynamically updated.
GIS data can expand to large areas and encompass an entire
space in which a team of robots are operating. The GIS data
covering the operating area provides a shared view of the
environment. The authors in [9], [12] highlight the use of GIS
data for cooperative map building and navigation. Localization
is coordinating through use of the GIS maps. Each robot
updates their position and ndings to the common space for
sharing with the other systems.
A benet of incorporating GIS with robotics is that the
systems support human readable formats: robotics ultimately
involves human interaction at some level and GIS data sup-
ports human-robot interactions; a point stressed in [9].
146
Our effort is focused on exploring the use of GIS with
sensor data for constructing probability maps for UGV nav-
igation and UAV landings. The authors in [13] presented a
method to build a probability map using local sensors and
known environment statistics intended for UGVs and UAVs.
Our effort expands on this concept through the addition of
GIS data. This contextual information supports more detailed
probability map construction.
III. ARCHITECTURE DESIGN
A system that leverages GIS data, including ours, involves
several processing steps. First, data needs to be collected from
available sources, including federal, state and local agencies.
Next, the data needs to be converted from the source formats
into common projections and extents so it can be compared.
Finally, the data is geoprocessed into the usable products. In
our work, the products are either probability maps or cost
surfaces to be used by unmanned autonomous systems to make
navigation decisions. The following subsections walk through
the three steps in processing GIS data for use by unmanned
systems.
A. Data Collection
A wide variety of GIS data is publicly available, covering
many different types of features. The level of detail for
each data set varies as well as the age of the data and its
accuracy. We can take advantage of this meta-data to inform
our probability modeling process; for example, more current
data can be weighted more heavily than old data. GIS data
can be roughly categorized into the following levels: global,
federal, state, county-city, and organization; the advantages
and disadvantages of each type are described below. For this
effort, we used data from Texas A&M University (TAMU).
Their data is specic to their campus, but is of high enough
granularity to be useful for unmanned vehicle navigation.
3
1) Global Data: Data at the global level is usually very
coarse grained and is generally only able to resolve features
larger than 1 square kilometer. This granularity limits the
utility of these layers for UGV operations, where the robot
only rarely will cross from one tile to the next. It can be
useful for UAV operations where a course-grained map can
still provide some information. An advantage of these data
sets is that you can achieve full global coverage with relatively
small le sizes.
2) Federal Data: Data from federal agencies gives more
detail about the features found within their boundaries. Many
of the federal layers are released openly to the public and con-
tain a wide variety of information such as terrain information
and land use. While this data is ner grained than that in the
global level, it is still mostly too coarse for UGV use, but can
be useful for UAVs.
3
The data was obtained from the following server: http://fcor.tamu.edu/
?pageid=77.
3) State Data: Most states have a GIS ofce that coordi-
nates the use of GIS data within their boundaries. Because
they cover a much smaller area, the resolution of the data is
usually superior to that obtained from the federal level. They
also have data from state agencies that is tailored to the local
area.
4) County-City Data: Generally, county-city data is the
nest granularity level available. Most cities now use GIS
systems to track tax parcels and for planning purposes. In some
cases, they have the location of every permanent structure
within their boundaries. While some of this data is freely avail-
able over the web, most GIS departments are not connected to
the internet and will deliver the data on DVD for a reasonable
fee.
5) Organization Data: For organizations of sufcient size
(e.g., large corporations, universities, etc.), they usually will
have their own GIS ofce that tracks their campus and its
details. This data is almost always at a higher delity than
available from any of the above sources. It generally contains
at least the structure information and may include details as
ne grained as individual tree and shrub placement.
B. Data Conversion
Agencies employ different data standards. It is important to
convert the diverse data sources to a common format; ArcGIS
provides tools to do this. This process generally needs to be
done only once per data layer, as further updates to the layer
will be able to leverage the same projection transformation.
For UGVs, the area of operation is small enough that few
problems arise. A UAV operating across multiple states would
need to correlate many more data sources and require addi-
tional conversion before use.
For the examples shown in this paper, we used data covering
the TAMU campus. The data from TAMU is in a North
American Datum (NAD) 1983 Texas Central State Plane pro-
jection. The other layers used were projected to this common
projection so that all layers could be compared directly. This
process, while optional, speeds the geoprocessing step since
you can then avoid the need to project each layer on-the-y.
C. Geoprocessing
With the data in a common projection, it can be geo-
processed into the components that we want to include in
the probability map. The rst stage is to create a rasterized
representation of each data layer. The ArcGIS tools allow
this step to be scripted based on a set of parameters. Once
the layers are rasterized, each layer of interest is thresholded
into a binary raster with good and bad areas marked. The
thresholding parameters are determined by the mission prole.
For example, on the structures layer, which includes all of
the buildings in the area, the buildings are marked as non-
traversable, while everything else is marked as traversable.
Future work could expand this classication beyond a simple
binary threshold; our initial work was focused on the devel-
opment of the architecture and a proof of concept.
147
Fig. 2. Geoprocessing for this effort involves three phases: rasterization, thresholding, and computing a weighted sum. The result is a composite map
representing probability of traversal or probability of minimized impact (depending upon a UGV or UAV application, respectively).
With all of the interesting layers rasterized and thresholded,
the nal geoprocessing steps are executed. A weighted sum is
calculated from each of the component layers. The weights
are selected according to the mission prole of the robot.
For the UGV case, roads may be preferred for navigation, in
which case the contribution of the traversable roadways will be
increased. However, if the mission prole stated that the UGV
should try to be stealthy by avoiding roads, the contribution
of the roadways would be reduced. Buildings are generally to
be avoided, for both UGVs and UAVs, and so are strongly
weighted as non-traversable. However, for a patrol task for a
UGV, the area immediately around buildings could receive a
higher weighting, resulting in the UGV system traveling near
the buildings as it patrols.
The nal composite layer is built up from these summed
components and, if needed, clipped to the area of inuence
of the robot. This composite layer can be thought of as
a probability of traversability (for UGVs) or probability of
minimized impact (for UAVs) map for the area in question.
The composite layer is added to the GIS server and is available
to be served up in a wide variety of formats to clients. Figure 2
represents the geoprocessing workow for this effort.
IV. PROBABILITY MODELS
The probability models used for the geoprocessing step
previously described are dened by the user through ArcGIS.
The models may be adjusted on the y as conditions change
or additional information is gathered. Figure 3 represents
the probability of minimized impact for an emergency UAV
landing. The map on the left represents safe landing zones
(green), unknown (yellow), and not-safe (red) for a small UAV
during the weekday. The map on the right shows the safe,
unknown, and not-safe zones for a small UAV during the
weekend. Note the larger amounts of red due to likelihood
of the area being used for sporting events on the weekends.
The cone-shaped colored area in both images represents the
projected landing area given the current position and heading
of the aircraft.
Figure 4 represents a similar situation to that in Figure 3
except with a large UAV. The probability model was adjusted
to account for the large UAV for the different size and
operating parameters of a larger UAV. Notice how the not-safe
zones (red) increased in size due to the potential for impact
reach.
V. INTEGRATION OF PROBABILITY MAPS WITH SENSOR
DATA
By treating the probability map as another sensor, we
can use it in sensor fusion on the autonomous systems. By
combining the information we receive from the GIS data (e.g.,
The XYZ building is in front of me) with that of the other
sensors (e.g., Radar says theres something in front of me; the
camera shows a window), we can build an enhanced picture
of the world. Because the extent of the GIS data is much
greater than any of the robots direct sensors, we can make
148
Fig. 3. Minimized impact probability maps for an emergency UAV landing at TAMU based on current location and possible trajectories: (a) small UAV
during the weekday, (b) small UAV during the weekend.
Fig. 4. Minimized impact probability maps for an emergency landing of a large UAV: (a) weekday, (b) weekend.
plans that are not limited to the current line of sight of the
robot, but instead extend further into the world.
Additionally, once we have a probability map of our area
of interest, we can use this information to facilitate robot
navigation. By feeding the robots position into the probability
map, we can calculate a least-cost surface from that starting
point. This use of the data allows the robot to determine least-
cost paths from its current position to any other point on
the map. By providing this capability on ArcGIS Server, the
paths can be calculated dynamically and updated as the robot
moves across the world, allowing us to ofoad some of the
navigation tasks from the robot to the GIS service. Figure 5
presents the two images representing cost-surface maps for
UGV navigation at TAMU. The image on the left is the least-
cost path. The image on the right shows the adjusted path
after a hazard has been detected. The yellow areas permit
navigation according to the GIS information; the coloring
fades to blue representing non-traversal. The robots current
position is marked by the red circle, along with the planned
path (red line).
VI. DISCUSSION
The architecture, incorporating ArcGIS components for geo-
processing, proved useful in constructing the minimal impact
and cost-surface probability maps. Informal feedback from
professionals in the military UAV and UGV elds conrmed
the potential benet of this information being shared with
operators and mission planners. Additionally, it is relatively
straightforward to modify weights and parameters to represent
differently sized UAV platforms.
It is important to note that these maps should be considered
just another type of sensor and one that is out of date. Because
of the time required for collection of GIS data, the layers
used do not reect current conditions, but rather the state of
the world as it was during the last survey. However, related
efforts have demonstrated how robot system information may
be used to update GIS data. For this effort, we make use of
the GIS data available without updating; future work would
involve GIS updating.
In addition, data obtained from different sources will likely
have different ages. Generally, we would weight more recent
information as being more valuable, but also need to consider
the resolution of the data: perhaps it makes more sense to use
a higher resolution data set that may be older.
While for the purposes of this effort, the composition
function used was a simple weighted sum, it is by no means
the only function that could be used. Because of the strong
geoprocessing tools available, any function could be applied
to the available map layers to produce results best suited to
the particular scenario.
149
Fig. 5. Cost-surface maps for UGV navigation at TAMU based on the current location and intended destination: (a) least-cost path for the UGV, (b) least-cost
path adjusted when hazard is marked.
VII. CONCLUSIONS AND FUTURE WORK
This effort provided a proof of concept in how GIS data
is useful for UGV navigation and UAV landing through pro-
ducing cost surface and minimal impact probability maps. In-
formal feedback supported the benet this information would
bring to the users of UGV and UAV systems. Lessons learned
from this effort, discussed in Section VI, dictate the future
work.
Follow-on efforts will need to address: a) core architecture,
b) user requirements, and c) interface for data interaction.
We should consider the architecture design when evolving the
ability to overlay and fuse GIS data for computing probabilis-
tic maps. The design currently employs ArcGIS. Alternative
designs may provide a more seamless means of building the
probability maps. It is important to also consider integration
within existing command and control environments.
The target users of this effort include the robot operators
as well as mission planners. Mission planners are responsible
for plotting resource utilization to meet mission objectives. A
mission planner would need to identify an area of interest,
requirements for the mission, and desirable or undesirable
landing zones. These types of contextual information should
be incorporated in the probability map construction.
Presenting the integrated GIS data to a user is also crucial.
Another avenue for future work includes the design of a user-
centered interface for command and control incorporating the
map probability views using alternative interactions, such as
multi-touch displays [14]. The richness of the map probability
views may demand multi-touch to enhance data presentation.
ACKNOWLEDGMENT
The authors would like to thank the organizations who made
their GIS data available and Dr. Richard Weatherly for his
support. This effort was funded by The MITRE Corporation
through the MITRE Innovation Program.
REFERENCES
[1] J. Peng, M. El Najjar, C. Cappelle, D. Pomorski, F. Charpillet, and
A. Deeb, A novel geo-localisation method using GPS, 3D-GIS and laser
scanner for intelligent vehicle navigation in urban areas, in Advanced
Robotics, 2009. ICAR 2009. International Conference on. IEEE, 2009,
pp. 16.
[2] H. Mori and S. Kotani, Robotic travel aid for the blind: HARUNOBU-
6, in European Conference on Disability, Virtual Reality, and Assistive
Technology, 1998.
[3] S. Shair, J. Chandler, V. Gonzalez-Villela, R. Parkin, and M. Jackson,
The use of aerial images and gps for mobile robot waypoint navigation,
Mechatronics, IEEE/ASME Transactions on, vol. 13, no. 6, pp. 692699,
2008.
[4] C. Brenner and B. Elias, Extracting landmarks for car navigation
systems using existing GIS databases and laser scanning, International
archives of photogrammetry remote sensing and spatial information
sciences, vol. 34, no. 3/W8, pp. 131138, 2003.
[5] P. Bonnifait, M. Jabbour, and V. Cherfaoui, Autonomous navigation in
urban areas using GIS-managed information, International Journal of
Vehicle Autonomous Systems, vol. 6, no. 1, pp. 83103, 2008.
[6] C. Cappelle, M. Najjar, D. Pomorski, and F. Charpillet, Intelligent Ge-
olocalization in Urban Areas Using Global Positioning Systems, Three-
Dimensional Geographic Information Systems, and Vision, Journal of
Intelligent Transportation Systems, vol. 14, no. 1, pp. 312, 2010.
[7] L. Bai and Y. Wang, Fusing image, GPS and GIS for road tracking
using multiple condensation particle lters, in Intelligent Vehicles
Symposium, 2008 IEEE. IEEE, 2008, pp. 162167.
[8] S. Renault, A. Le Meur, and D. Meizel, GPS/GIS localization for
management of vision referenced navigation in urban environments,
in Intelligent Transportation Systems, 2005. Proceedings. 2005 IEEE.
IEEE, 2005, pp. 608613.
[9] J. Mirats Tur, C. Zinggerling, and A. Corominas Murtra, Geographical
information systems for map based navigation in urban environments,
Robotics and Autonomous Systems, vol. 57, no. 9, pp. 922930, 2009.
[10] M. Kais, S. Dauvillier, A. De La Fortelle, I. Masaki, and C. Laugier,
Towards Outdoor Localization Using GIS, Vision System andStochastic
Error Propagation, 2004.
[11] J. Meguro, K. Ishikawa, T. Hashizume, J. Takiguchi, I. Noda, and
M. Hatayama, Disaster information integration into geographic infor-
mation system using rescue robots, in Proc IEEE/RSJ Int Conf on
Intelligent Robots and Systems (IROS2006), 2006.
[12] A. Sanfeliu and J. Andrade-Cetto, Ubiquitous networking robotics in
urban settings, in Proceedings of the IEEE/RSJ IROS Workshop on
Network Robot Systems, 2006, pp. 1418.
[13] M. Jun and R. DAndrea, Probability map building of uncertain
dynamic environments with indistinguishable obstacles, in American
Control Conference, 2003. Proceedings of the 2003, vol. 4. IEEE,
2003, pp. 34173422.
[14] M. Micire, Multi-touch interaction for robot command and control,
Ph.D. dissertation, University of Massachusetts Lowell, 2010.
[15] H. Yanco, M. Micire, J. Drury, and J. Casper, Improving command
and control: how new interaction methods will help, University of
Massachusetts Lowell, Tech. Rep., 2008.
150