Hand-Drawn Maps for Robot Navigation

Abstract

The goal of this work is to create a robot interface that allows a novice user to guide, control, and/or program a robot to perform some task. The assumption is that, although the user may be a domain expert in how the task should be done, he is not an expert in robotics. During the actual robot use, he should focus on the task to be done rather than worrying about the robot or the interaction modality. To address this goal, we have been investigating the use of hand-drawn route maps to transfer navigation tasks to robots. In the paper, we provide an overview and current status of ongoing work with sketches. We discuss what type of information would be useful for directing and controlling a robot and then show how this information can be extracted from a sketched route map, in the form of spatial relationships. An analysis example of a PDA-generated sketch is included. Also, preliminary results are presented which compare the analysis of a sketched map with that of a real map.

Full text

2002 AAAI Spring Symposium, March, 2002, Technical Report SS-02-08
Hand-Drawn Maps for Robot Navigation
Marjorie Skubic, Sam Blisard, Andy Carle, and Pascal Matsakis
Dept. of Computer Engineering and Computer Science
University of Missouri-Columbia
Columbia, MO 65211
skubicm@missouri.edu
Abstract
The goal of this work is to create a robot interface that allows
a novice user to guide, control, and/or program a robot to perform
some task. The assumption is that, although the user may be a
domain expert in how the task should be done, he is not an expert
in robotics. During the actual robot use, he should focus on the
task to be done rather than worrying about the robot or the
interaction modality. To address this goal, we have been
investigating the use of hand-drawn route maps to transfer
navigation tasks to robots. In the paper, we provide an overview
and current status of ongoing work with sketches. We discuss
what type of information would be useful for directing and
controlling a robot and then show how this information can be
extracted from a sketched route map, in the form of spatial
relationships. An analysis example of a PDA-generated sketch is
included. Also, preliminary results are presented which compare
the analysis of a sketched map with that of a real map.
Introduction
Being able to interact and communicate with robots in
the same way we interact with people has long been a goal
of AI and robotics researchers. However, much of the
robotics research in the past has emphasized the goal of
achieving autonomous agents. In our research, we are less
concerned with creating autonomous robots that can plan
and reason about tasks, and instead we view them as semi-
autonomous tools that can assist a human user. The robot
may have some perception capabilities, reactive behaviors,
and perhaps limited reasoning abilities that allow it to
handle an unstructured and dynamic environment. But the
user supplies the high-level and difficult reasoning and
strategic planning capabilities.
In this scenario, the interaction and communication
between the robot and the human user becomes very
important. The user must be able to easily communicate
what needs to be done, perhaps at different levels of task
abstraction. In particular, we would like to provide an
intuitive method of communicating with robots that is easy
for users that are not expert robotics engineers. We want
domain experts to define their own task use of robots,
which may involve controlling them, guiding them, or even
programming them.
As one strategy for addressing this goal, we have been
investigating the use of hand-drawn route maps, in which
the user sketches an approximate representation of the
environment and then sketches the desired robot trajectory
with respect to that environment. The objective in the
sketch interface is to extract spatial information about the
map and a qualitative path through the landmarks drawn on
the sketch. This information is used to build a task
representation for the robot, which operates as a semi-
autonomous vehicle. Note that the task representation is
based on sensing and relative position, not absolute
position.
Although qualitative navigation presents problems for
autonomous robots (e.g., due to perception difficulties; see
a discussion in Murphy 2000), we believe that it is a good
idea for semi-autonomous robots, where the user
interactively observes and directs the robot. Qualitative
navigation more closely mimics the human navigation
process, and in the context of sketch interfaces, also allows
for a more intuitive interface with the human user.
Possible applications include the following:
1. Military applications. The user looks at a scene and
sketches a route through landmarks. Also,
programming strategic behaviors, such as how to
search or how to escape.
2. Programming large construction or mining
equipment.
3. Guiding planetary rovers.
4. Controlling personal robots.
In the remaining sections of the paper, we first discuss
background material on human navigation and the use of
sketched route maps. We illustrate how spatial relations
can be used to analyze sketched maps and briefly describe
our methodology for modeling spatial relationships based
on the histogram of forces (Matsakis 1998). In addition,
we provide a framework for the robot control architecture,
and discuss what kind of information must be extracted
from the sketched maps to facilitate the necessary robot
control. Finally, sketch interpretation is illustrated with a
map sketched on a PDA. We also present preliminary
results, which compare the analysis of a sketched map with
that of a real map. The conclusion includes a brief
discussion on the current status and future directions.
Human Navigation and Sketched Route Maps
A sketched route map is drawn to help someone
navigate along a path for the purpose of reaching a goal.
An example is shown in Fig. 1a. Although route maps do
not generally contain complete map information about a
region, they do provide relevant information for the
navigation task. People sketch route maps to include
landmarks at key points along the path and use spatial
relationships to help depict the route, often adding arrows

2002 AAAI Spring Symposium, March, 2002, Technical Report SS-02-08
and other notation for clarity (Tversky and Lee 1998). In a
study of 29 sketched route maps, each contained the
information necessary to complete a complicated
navigation task (Tversky and Lee 1998).
Figure 1. (a) A map sketched on paper, describing a route
through the MU campus. (b) A view of the actual map.
Buildings 13, 9, 29 and 81 correspond to major landmarks
included on the sketched map.
Research by Michon and Denis (2001) provides
insights into how landmarks are used for human navigation
and what are considered to be key route points. In
studying route directions, they found that landmarks were
used more frequently at four types of critical nodes: (1) the
starting point, (2) the ending point, (3) at a change in
orientation, and (4) at places along the route where errors
could easily occur, such as major intersections (Michon
and Denis 2001). Thus, people use the relative position of
landmarks as cues to keep on track and to determine when
to turn left or right.
The work of the researchers noted above and others
(e.g., Previc 1998, Schunn and Harrison 2001) indicate the
importance of environment landmarks and spatial
relationships in human navigation. The work suggests that
spatial relationships of landmarks with respect to the
desired route may be useful not only for robot control but
also as a link between a robot and its human user. In the
next section, we describe a tool for modeling spatial
relationships that is fast, robust, and handles all object
contours using either raster data or vector data (i.e., a
boundary representation).
Modeling Spatial Relationships
Freeman (1975) proposed that the relative position of
two objects be described in terms of spatial relationships
(such as “above,” “surrounds,” “includes,” etc.). He also
proposed that fuzzy relations be used, because “all-or-
nothing” standard mathematical relations are clearly not
suited to models of spatial relationships. By introducing
the histogram of angles, Miyajima and Ralescu (1994)
developed the idea that the relative position between two
objects can have a representation of its own and can thus
be described in terms other than spatial relationships.
However, the representation proposed shows several
weaknesses (e.g., requirement for raster data, long
processing times, anisotropy).
In the context of image analysis, Matsakis and
Wendling (1999) introduced the histogram of forces.
Contrary to the angle histogram, it ensures processing of
both raster data and vector data. Moreover, it offers solid
theoretical guarantees, allows explicit and variable
accounting of metric information, and lends itself, with
great flexibility, to the definition of fuzzy directional
spatial relations (such as “to the right of,” “in front of,”
etc.). For our purposes, the histogram of forces also allows
for a low-computational handling of heading changes in
the robot’s orientation and makes it easy to switch between
an allocentric (world) view and an egocentric (robot) view.
The Histogram of Forces
The relative position of a 2D object A with regard to
another object B is represented by a function FAB from IR
into IR +. For any direction θ, the value FAB(θ) is the total
weight of the arguments that can be found in order to
support the proposition “A is in direction θ of B”. More
precisely, it is the scalar resultant of elementary forces.
These forces are exerted by the points of A on those of B,
and each tends to move B in direction θ (Fig. 2). FAB is
called the histogram of forces associated with (A,B) via F,
or the F−histogram associated with (A,B). The object A is
the argument, and the object B the referent. Actually, the
letter F denotes a numerical function. Let r be a real
number. If the elementary forces are in inverse ratio to dr,
where d represents the distance between the points
considered, then F is denoted by Fr . The F0 –histogram
(histogram of constant forces) and F2 –histogram
(histogram of gravitational forces) have very different and
very interesting characteristics. The former coincides with
the angle histogram—without its weaknesses—and
provides a global view of the situation. It considers the
closest parts and the farthest parts of the objects equally,
whereas the F2 –histogram focuses on the closest parts.
Figure 2. Computation of FAB(θ). It is the scalar
resultant of forces (black arrows). Each one tends
to move B in direction θ.
Throughout this paper, the referent B is the robot. The
F-histogram associated with (A,B) is represented by a
(a) (b)
A
B
θ

2002 AAAI Spring Symposium, March, 2002, Technical Report SS-02-08
limited number of values (i.e., the set of directions θ is
made discrete), and the objects A and B are assimilated to
polygons using vector data. The computation of FAB is of
complexity O(n log(n)), where n denotes the total number
of vertices (Matsakis and Wendling 1999). Details on the
handling of vector data can also be found in Skubic et. al.
(2001a, 2002a).
Linguistic Description of Relative Positions
The histogram of forces provides a tool for modeling
spatial relationships; the model can also be used to build
qualitative spatial descriptions that provide a linguistic link
to the user. Matsakis et al. (2001) present such a system
that produces linguistic spatial descriptions of images.
The description of the relative position between any 2D
objects A and B relies on the sole primitive directional
relationships: “to the right of,” “above,” “to the left of” and
“below” (imagine that the objects are drawn on a vertical
surface). It is generated from F0
AB (the histogram of
constant forces associated with (A,B)) and F2
AB (the
histogram of gravitational forces). First, eight values are
extracted from the analysis of each histogram: ar (RIGHT),
br (RIGHT), ar (ABOVE), br (ABOVE), ar (LEFT),
br (LEFT), ar (BELOW) and br (BELOW). They represent
the “opinion” given by the considered histogram (i.e., F0
AB
if r=0, and F2
AB if r=2).
For instance, according to F0
AB the degree of truth of
the proposition “A is to the right of B” is a0(RIGHT). This
value is a real number greater than or equal to 0
(proposition completely false) and less than or equal to 1
(proposition completely true). Moreover, according to F0
AB
the maximum degree of truth that can reasonably be
attached to the proposition (say, by another source of
information) is b0(RIGHT) (which belongs to the interval
[a0(RIGHT),1]).
F0
AB and F2
AB’s opinions (i.e., the sixteen values) are
then combined. Four numeric and two symbolic features
result from this combination. They feed a system of fuzzy
rules and meta-rules that outputs the expected linguistic
description. The system handles a set of adverbs (like
“mostly,” “perfectly,” etc.), which are stored in a
dictionary, with other terms, and can be tailored to
individual users.
A description is generally composed of three parts. The
first part involves the primary direction (e.g., “A is mostly
to the right of B”). The second part supplements the
description and involves a secondary direction (e.g., “but
somewhat above”). The third part indicates to what extent
the four primitive directional relationships are suited to
describing the relative position of the objects (e.g., “the
description is satisfactory”). In other words, it indicates to
what extent it is necessary to utilize other spatial relations
such as “surrounds”. When range information is available,
a fourth part can also be generated to describe distance
(e.g., “A is close to B”) (Skubic et. al. 2001a).
Framework for Human-Robot Interaction
Robot navigation is modeled as a procedural task (i.e.,
a sequence of steps) to mimic the human navigation
process. The framework for the robot control architecture
is shown in Fig. 3. Task structure is represented as a Finite
State Automaton (FSA) in the Supervisory Controller,
following the formalism of the Discrete Event System
(Ramadge and Wonham 1989). The FSA models a
sequence of moves, each of which is governed by a robot
behavior (a local control strategy). The complete sequence
comprises a task. The sensor-based qualitative state (QS) is
used for task segmentation. A change in QS is an event that
corresponds to a change in the type of movement.
Figure 3. The Robot Control Architecture
For navigation tasks, the QS is formed by the spatial
relationships of environment landmarks with respect to the
robot. Thus, the robot uses landmarks in the same way that
a person would use landmarks. Through the State
Classifier, the robot is provided with the ability to
recognize a set of qualitative states, which are extracted
from sensory information, thus reflecting the current
environmental condition. For navigation, robot-centered
spatial relations provide context (e.g., “there is an object to
the left front”). Adding the ability to recognize classes of
objects provides additional perception (e.g., “there is a
person to the left front”).
The robot is also equipped with a set of (reactive)
behaviors that are managed by the Behavioral Controller.
Reactive behaviors allow the robot to respond quickly and
safely to dynamic and unexpected conditions, such as
avoiding moving obstacles. Output from the behavioral
controller is merged with discrete commands issued from
the Supervisory Controller. Note that this combination of
discrete event control in the Supervisory Controller and the
“signal processing” in the Behavioral Controller is
consistent with Brockett’s (1993) framework of hybrid
control systems and has similarities to other approaches
used for qualitative robot navigation (e.g., Kuipers 1998).
As shown in Fig. 3, the interface between the robot and
the human user relies on the qualitative state for two-way
communications. In robot-to-human communications, the
QS allows the user to monitor the current state of the robot,
ideally in terms easily understood (e.g., “there is an object
on the right”). In human-to-robot communications,
Behavioral Controller
Supervisory
Controller
State
Classifier
discrete
commands
User Interface
qualitative
state
sensor
signals
actuator
commands
qualitative
instructions
Behavioral Controller
Supervisory
Controller
State
Classifier
discrete
commands
User Interface
qualitative
state
sensor
signals
actuator
commands
qualitative
instructions

2002 AAAI Spring Symposium, March, 2002, Technical Report SS-02-08
commands are segmented by the QS, termed qualitative
instructions in the figure (e.g., “while there is an object on
the right, move forward”). This illustrates the type of
information that must be extracted from the sketched route
maps, in order to direct the robot along the intended path.
We extract a sequence of qualitative states in the form of
spatial relationships of environment landmarks with
respect to the robot. And we extract the corresponding
robot movements for each QS which make up the set of
qualitative instructions for the desired task.
Interpreting a PDA-Sketched Map
We now have all of the pieces for analyzing a sketched
route map, and in this section we illustrate how navigation
information is extracted from a map sketched on a PDA
such as a PalmPilot. The stylus interface of the PDA
allows the user to sketch a map much as she would on
paper for a human colleague. The PDA captures the string
of (x,y) coordinates sketched on the screen which forms a
digital representation suitable for processing.
The user first draws a representation of the
environment by sketching the approximate boundary of
each object. During the sketching process, a delimiter is
included to separate the string of coordinates for each
object in the environment. After all of the environment
objects have been drawn, another delimiter is included to
indicate the start of the robot trajectory, and the user
sketches the desired path of the robot, relative to the
sketched environment. An example of a sketch is shown in
Fig. 4, where each point represents a captured screen pixel.
The extraction of spatial information from the sketch is
summarized in Fig. 5. For each point along the trajectory, a
view of the environment is built, using the radius of the
sensor range. For each object within the sensory radius, a
polygonal region is built using the boundary coordinates of
the object as vertices.
Figure 4. A route map sketched on a PDA.
We have used different strategies for building the
polygonal representations of the objects, for example,
using only the points of the object that fall within the
sensory radius (Skubic et. al. 2001b). Here, if any of the
object points lie within the sensory radius, we use the
entire object boundary. This approach coincides more
closely with our recent work using occupancy grid cells
(Skubic et. al. 2002b).
Once the polygonal region of an object is built, the
histograms of constant forces and gravitational forces are
computed as described previously. The referent is always
the robot, which is modeled as a square for the histogram
computations. To capture robot-centered spatial
relationships, the robot orientation must also be
considered. The robot’s heading is computed using
adjacent points along the sketched path to determine an
instantaneous orientation. We compensate for the discrete
pixels by averaging 5 adjacent points (centered on the
considered trajectory point), thereby filtering small
perturbations and computing a smooth transition as the
orientation changes. After the heading is calculated, it is
used to shift the histograms along the horizontal axis to
produce an egocentric (robot) view.
Figure 5. Synoptic diagram showing how spatial information is extracted from the sketch.
robot
path
“left-front”
SYSTEM of
FUZZY RULES
constant
forces
gravitational
forces
-
π
-
π
π
π
(a) (b)
(c)
(c)
(d)
+
m
a
i
n
d
i
r
e
c
t
i
o
n

2002 AAAI Spring Symposium, March, 2002, Technical Report SS-02-08
The histograms of constant forces and gravitational
forces associated with the robot and the polygonal region
of each object are used to generate a linguistic description
of the relative position. In addition, features can be
extracted from the histograms to further represent the
spatial relationship. In processing the sketch, we extract
what is considered to be the “main direction” of the object
with respect to the robot and discretize it into one of 16
possible directions, as shown in Fig. 6. Examples of
corresponding linguistic descriptions are also shown in Fig.
6 for a sampling of directions.
Figure 6. Sixteen directions are situated around the robot
(the small circles). The “main direction” of each object is
discretized into one of these 16 directions. Examples are
included of corresponding linguistic descriptions.
In addition to extracting spatial information on the
environment landmarks, we also extract the movement of
the robot along the sketched path. The computation of the
robot heading, described above, provides an instantaneous
orientation. However, we also want to track the change in
orientation over time and compute what would correspond
to robot commands, e.g., move forward, turn left, make a
“hard” left. The turning rate is determined by computing
the change in instantaneous heading between two adjacent
route points and dividing by the distance between the
points to normalize the rate. A positive rate means a turn
to the left, and a negative rate means a turn to the right.
The spatial information and robot movement extracted
from the sketched map in Fig. 4 is summarized in Fig. 7.
In Fig. 7b, the main direction of each object is plotted for
the route steps in which the object is “in view”; labels of
the corresponding directions are displayed on the graph to
show the symbolic connection. The normalized turning
rate which tracks the robot movement along the trajectory
is also shown in Fig. 7b. For reference, we have included a
sampling of linguistic descriptions generated along the
route (Fig. 7c), which correspond to the positions shown in
Fig. 7a. Note that the descriptions have an egocentric
(robot) perspective.
1
6
5
11
12
16
#1
#2
#3
#4
#5
1
6
5
11
12
16
#1
#2
#3
#4
#5
1. Object #1 is behind the robot but extends to the right relative to the robot
Object #2 is to the right of the robot but extends forward relative to the robot
5. Object #2 is to the right of the robot but extends to the rear relative to the robot
Object #3 is in front of the robot but extends to the right relative to the robot
6. Object #2 is to the right of the robot
Object #3 is in front of the robot but extends to the left relative to the robot
11. Object #2 is to the right of the robot
Object #3 is behind-left of the robot
12. Object #2 is nearly to the right of the robot but extends to the rear relative to the robot
Object #3 is mostly to the left of the robot but somewhat to the rear
Object #4 is loosely to the right-front of the robot
16. Object #3 is to the left of the robot but extends to the rear relative to the robot
Object #4 is to the right of the robot but extends to the rear relative to the robot
Object #5 is in front of the robot
The object is very close to the robot.
(a) (b)
(c)
Object is mostly in front
but somewhat to the left
Object is mostly to the left
but somewhat forward
front
robot
Object is to the right
Object is to the right front
Object is mostly in front
but somewhat to the left
Object is mostly to the left
but somewhat forward
front
robot
Object is to the right
Object is to the right front
PDA Ske tch
-4
-2
0
2
4
6
8
10
12
14
16
0 2 4 6 8 10 12 14 16 18
steps
Main direction
Turning ra te
object #1
object #2
object #3
object #4
object #5
Front
Rear
Left
Right
turning right
tur ning lef t
Figure 7. The PDA sketch.
(a) The original sketch with an
overlay of the robot’s sensory
radius for several points along
the route.
(b) Normalized turning rate of
the robot along the sketched
route with the corresponding
discrete main directions of the
objects.
(c) Generated egocentric
linguistic descriptions for the
route points shown in (a).

2002 AAAI Spring Symposium, March, 2002, Technical Report SS-02-08
The turning rate in Fig. 7b, although not translated into
discrete robot commands, shows the general trend in the
robot movement along the route and the correlation with
relative positions of the environment landmarks. At the
beginning of the route, when object #1 is behind the robot,
the robot’s movement is generally straight ahead (slightly
to the left). When object #3 is in view, the robot turns to
the right until the object is mostly on the left. When object
#4 is in view to the front, the robot turns left and stops
when object #5 is in front and very close. In this way, we
can extract the key points along the route where a change
in direction is made, by capturing the relative positions of
the landmarks with respect to the route.
Comparing a Sketched Map to a Real Map
To further investigate the use of sketches, we also
compared a sketched (qualitative) map to a real
(quantitative) map. In particular, we wanted to test the
hypothesis that, although users may not sketch a map to an
accurate scale, nor even use accurate shapes, they do tend
to use accurate spatial relationships. And we wanted to
test how consistent these relationships are, as measured by
our tool. In this section, we present preliminary results
showing an analysis of the maps in Fig. 1.
Sketches were collected by asking students to draw a
map showing the route on the MU campus, from Brady
Commons to the Memorial Union, two well-known
landmarks. Maps were sketched on paper. For analysis,
the sketched maps, as well as the real map, were digitally
scanned. A few key boundary points were manually
extracted from the scanned maps for 4 building landmarks,
to produce a digitized representation as shown in Figures
8a and 9a. The digitized version of the sketch is similar to
the type of representation captured from a PDA-sketched
map.
The maps have been processed as described in the
previous section; the results are shown in Figures 8b and
9b. The one change from the previous analysis is that the
main direction of each environment landmark is plotted for
the complete route. This eliminates the need for setting a
sensory radius that corresponds to both the sketched map
and the real map. (We did not ask that the sketches be
drawn to an accurate scale and we did not expect that they
would be.) Also, note that a main direction of 16 in the
graph is equivalent to a direction of 0 (front) because of a
circular wrap. The higher numbers are used for
convenience to show the continuous trend over the entire
route.
In comparing the results, we can see a pattern that is
similar although not completely identical. For example, in
the first right turn, consider the two closest landmarks. For
both maps, the A&S building is slightly left of the rear
(main direction = 7) just before the turn and changes to
exactly rear (main direction = 8) at the first step in the turn.
During these same two steps, Ellis Library is slightly to the
rear of right (main direction = 11) in the real map and
exactly right (main direction = 12) in the sketched map. At
other key steps where the route turns, one can observe
similar results.
To further analyze the sketches, we also computed the
spatial relationships between pairs of landmarks and
compared the results of the real map with those of the
sketched maps. This is less important for extracting a
qualitative task representation of the navigation, but of
interest nonetheless in analyzing the sketches. If the spatial
relations between corresponding landmark pairs agree,
then the spatial relations might be used as a basis for
relating landmarks in a sketched map with those in an
accurate map.
Six sketched maps were compared to the real map in
Figure 1b. Again, each sketch was scanned and key
boundary points were manually extracted as above for the
same 4 building landmarks. At this time, we present
preliminary results for the sketches, comparing only the
main directions of each landmark pair. The sketched route
maps represented a broad range in terms of accuracy, scale,
and shape of the buildings used as landmarks. Also, the
routes sketched did not always follow the same path with
respect to the landmarks. However, in spite of these
differences, the discrete main directions of two landmarks
did not vary by more than +/- 2 values. In most cases, the
values agreed or were within +/- 1, especially in the cases
where the same route was taken in the sketch. These
results are not comprehensive but they do show promise in
using spatial relationships to analyze sketched route maps
and compare them to accurate maps.
Concluding Remarks
The work on sketched route maps for robot navigation
is by no means complete. The work presented here is
merely a snapshot of an initial approach in extracting
qualitative information from the sketch. We believe the
real potential lies in a more interactive approach with the
sketch interface. For example, the spatial states and
behaviors could be displayed as the route is being sketched
so that the user could change them if necessary. Also,
editing gestures could be added, as in Landay and Myers
(2001), allowing the user to delete landmarks, add labels to
landmarks, and to specify qualitative distances (such as
how close should the robot get to the landmarks).
Acknowledgements
Several MU colleagues and students have contributed
to this overall effort, including Professor Jim Keller and
students George Chronis and Ben Forrester. This research
has been supported by ONR, the Naval Research
Laboratory and the MU Discovery Fellowship Program.

2002 AAAI Spring Symposium, March, 2002, Technical Report SS-02-08
Figure 8. The sketched route map corresponding to Figure 1a. (a) The digitized representation. (b) Normalized turning rate
of the sketched path with the corresponding discrete main directions of the environment landmarks.
Figure 9. The route map drawn on the real map in Figure 1b. (a) The digitized representation. (b) Normalized turning rate
of the path with the corresponding discrete main directions of the environment landmarks.
Re al M ap
-10
-8
-6
-4
-2
0
2
4
6
8
10
12
14
16
18
0 5 10 15 20 25 30 35 40 45 50
steps
Main Direction
Turning rate
Br ady Commo ns
A&S
Mem. Unio n
Ellis Libr ary
Right
Rea r
Left
Fro nt
Fro nt
turning right
turning right
turning left
Sketched Map
-12
-10
-8
-6
-4
-2
0
2
4
6
8
10
12
14
16
18
0 5 10 15 20 25 30
steps
Main DI rectio n
Turning rate
Brady Commons
A&S
Mem. Unio n
Ellis Lib rar y
Fron t
Fron t
Left
Rea r
Rig ht
turning right
turning right
tur ning lef t
(a)
(a)
(b)
(b)

2002 AAAI Spring Symposium, March, 2002, Technical Report SS-02-08
References
Brockett, R.W. 1993. “Hybrid models for motion control
systems,” in Essays on Control: Perspectives in the
Theory and Its Applications, H.L. Trentelman and J.C.
Willems (ed.), pp.29-53. Boston, MA: Birkhauser.
Freeman, J. 1975. “The Modelling of Spatial Relations,”
Computer Graphics and Image Processing (4):156-
171.
Kuipers, B. 1998. “A Hierarchiy of Qualitative
Representations for Space,” in Spatial Cognition: An
Interdisciplinary Approach to Representing and
Processing Spatial Knowledge, C. Freksa, C. Habel, K.
Wender (ed.), Berlin: Springer, pp. 337-350.
Landay, J. and Myers, B. 2001. “Sketching Interfaces:
Toward More Human Interface Design,” IEEE
Computer, 34(3):56-64
Matsakis, P. 1998. Relations spatiales structurelles et
interprétation d’images, Ph. D. Thesis, Institut de
Recherche en Informatique de Toulouse, France.
Matsakis, P. and Wendling, L. 1999. “A New Way to
Represent the Relative Position between Areal
Objects”, IEEE Trans. on Pattern Analysis and
Machine Intelligence, 21(7):634-643.
Matsakis, P., Keller, J., Wendling, L., Marjamaa, J. and
Sjahputera, O. 2001. “Linguistic Description of
Relative Positions in Images”, IEEE Trans. on Systems,
Man and Cybernetics, part B, 31(4):573-588.
Michon, P.-E. and Denis, M. 2001. “When and Why Are
Visual Landmarks Used in Giving Directions?” in
Spatial Information Theory: Foundations of
Geographic Information Science, D. Montello (ed.),
Berlin: Springer, pp. 292-305.
Miyajima, K. and Ralescu, A. 1994. “Spatial Organization
in 2D Segmented Images: Representation and
Recognition of Primitive Spatial Relations,” Fuzzy Sets
and Systems, 65(2/3):225-236.
Murphy, R. 2000. Introduction to AI Robotics.
Cambridge, MA: MIT Press.
Perzanowski, D., Schultz, A., Adams, W., Marsh, E. and
Bugajska, M. 2001. “Building a Multimodal Human-
Robot Interface,” IEEE Intelligent Systems, Jan./Feb,
pp. 16-20.
Previc, F.H. 1998. “The Neuropsychology of 3-D Space,”
Psychological Review, 124(2):123-164.
Ramadge, P. and Wonham, W. 1989. “The control of
discrete event systems,” Proceedings of the IEEE,
77(1):81-97.
Schunn, C. and Harrison, T. 2001. Personal
communication.
Skubic, M., Chronis, G., Matsakis, P. and Keller, J.
2001a. “Generating Linguistic Spatial Descriptions
from Sonar Readings Using the Histogram of Forces,”
in Proceedings of the IEEE 2001 International
Conference on Robotics and Automation, May, Seoul,
Korea, pp. 485-490.
Skubic, M., Matsakis, P., Forrester, B and Chronis, G.
2001b. “Extracting Navigation States from a Hand-
Drawn Map,” in Proceedings of the IEEE 2001
International Conference on Robotics and Automation,
May, Seoul, Korea, pp. 259-264.
Skubic, M., Chronis, G., Matsakis, P. and Keller, J.
2001c. “Spatial Relations for Tactical Robot
Navigation,” in Proceedings of the SPIE, Unmanned
Ground Vehicle Technology III, April, Orlando, FL.
Skubic, M., Matsakis, P., Chronis, G. and Keller, J.
2002a. “Generating Multi-Level Linguistic Spatial
Descriptions from Range Sensor Readings Using the
Histogram of Forces,” Submitted to Autonomous
Robots.
Skubic, M., Perzanowski, D., Schultz, A. and Adams, W.
2002b. “Using Spatial Language in a Human-Robot
Dialog,” accepted for the IEEE 2002 International
Conference on Robotics and Automation.
Tversky, B. and Lee, P. 1998. “How Space Structures
Language,” in Spatial Cognition: An Interdisciplinary
Approach to Representing and Processing Spatial
Knowledge, C. Freksa, C. Habel, K. Wender (ed.),
Berlin: Springer, pp. 157-176.