Professional Documents
Culture Documents
Robust Navigation System Based On Rfid Transponder Barriers For The Interactive Behavior-Operated Shopping Trolley (Inbot)
Robust Navigation System Based On Rfid Transponder Barriers For The Interactive Behavior-Operated Shopping Trolley (Inbot)
Abstract
Purpose – The purpose of this paper is to present a navigation system designed for highly dynamic environments which is independent from a
metrically exact global map.
Design/methodology/approach – A navigation system is developed to cope with highly dynamic environments. Here, this refers especially to
changes in the environment itself, like the daily deployment or removal of advertisements or special offers in a supermarket. The navigation system is
split into a global part, relying on non-concealable artificial landmarks and a local part containing a behavior-based control using a dynamic potential
field approach. The required information are the definitively static structures and the actual sensor information only.
Findings – The system proved to be useful in environments that change frequently and where the presence of many people complicates the perception
of landmarks.
Practical implications – The presented navigation system is robust against changes in the environment and provides reliable collision avoidance
capabilities.
Originality/value – It is a useful navigation system for autonomous robots dedicated to frequently changing and populated environments.
1. The application of InBOT 2009a), it can follow or lead the user, and finally it can be
commanded to act independently.
The interactive behavior-operated shopping trolley (InBOT, The primary challenge InBOT has to cope with is highly
Figure 1) is an approach to transfer state of the art robotics dynamic environments in the described scenario. This
technology into human everyday environments. InBOT includes on the one hand the presence of lots of dynamic
addresses several everyday problems. Among other objects, which constrain local navigation, but on the other
possibilities this means helping the customer to find the hand that the spatial arrangement of the environment changes
desired products without extensive search in big frequently. Advertisements or special offers are placed or
supermarkets, or relieving the customer from the burden of removed daily. This is a great challenge to mobile robots
pushing the shopping cart using his own force all the time, because as result they cannot rely on a global metrical map or
especially if the cart is heavily loaded or the customer is have to build this map several times a day even while the shop
elderly or handicapped. Especially, for these groups of is open and customers are around, in the worst case. InBOT
customers it could be very interesting that InBOT is able to aims to solve this problem by being independent from a global
metrical map. A global topological map based on the
avoid collisions on its own, even with objects that are moving
definitively static structures in combination with a local map
themselves. InBOT’s ability to perform special local
of the visible area has to be sufficient. The different
maneuvers and a flexible task-planner round off the
topological areas are discriminated by occlusion-free
capabilities. It provides four different modes of operation to radio-frequency identification (RFID) transponder barriers.
assist the user in the best way: InBOT can be steered like an A rough overview over the main components of InBOT is
ordinary shopping cart by the haptic handle (Göller et al., shown in Figure 2.
The navigation system consists of two main parts (Figure 3),
The current issue and full text archive of this journal is available at the task planner and the local navigation. First of all global
www.emeraldinsight.com/0143-991X.htm planning and topological navigation take place. Here, the
commands of the user are analyzed and the task planner
377
Robust navigation system based on RFID transponder barriers Industrial Robot: An International Journal
Michael Göller et al. Volume 36 · Number 4 · 2009 · 377 –388
Figure 1 The InBOT is equipped with a sensorized haptic handle, a generates a global plan based on the topological navigation.
touch screen and a remote control. The lower casing was removed to The local navigation contains a behavior network that fulfills
show the mecanum drive of InBOT the given tasks inside a local area. Here, the basic and
advanced behavior repertoires (Figure 4) are set up. First, the
actual scene is observed and sub-targets are generated with
respect to static and dynamic objects. Then the behavior
network drives InBOT along this route and is responsible for
the micro-management of the obstacle avoidance and the
velocity control.
Figure 2 Main components of InBOT’s navigation system. The task planner generates a plan based on the commands given by the user. The behaviors
of the different behavior repertoires execute the plan based on information obtained from sensors or the local world model
ata
Pr
dd
ep r inf
e
n
se
erg
io
ro
ns
act
/m
ce orm
o
d
sse
se Ex
xt
trac
aly
Ne
d tion
n ted
/a tasks
ed
a
ess
r oc
P
Sensors
Behaviours 2× S300 Laser scanner
Preprocessed sensor information
Advanced behaviour repertoire 2× 2D-Measurement wheels
Basic behaviour repertoire Gyroscope
Safety behaviours Haptic handle
Remote control
data
Movement en sor
commands
Ra ws
378
Robust navigation system based on RFID transponder barriers Industrial Robot: An International Journal
Michael Göller et al. Volume 36 · Number 4 · 2009 · 377 –388
Figure 3 Overview of InBOT’s navigation system. The command received from the user is analyzed and a global plan is created based on the
topological map of the shop. This plan is handed down to the local navigation step by step. The different behaviors for predictive and reactive obstacle
avoidance execute the plan and drive InBOT through the environment safely
User interface
Move to product (milk); Area 26; (5.5, 1.2)m
Follow person
Task
Follow robot
Deive to (Topological node,
target inside node)
Topolgical
navigation
01 02 03 04 05
06 07 08
29
34
Actual: 12
09 10 11 12 13 14
30
Drive to 16
35
15 16 17 18 31 Drive to 22
32 Drive to 26
19 20 21 22 23 24 36
33 Drive to (5.5, 1.2) in 26
25 26 27 28
Follow person
Task
Follow robot
Deive to target inside node
Local navigation
Move autonomously
Follow user
Lead user
Haptic steering mode
Straight left
Very slow
Careful
No restriction 0°
Straight right
exact measurement. When moving inside such a node the using a behavior network as its control mechanism. It is based
robot is moved by the behavior-based control. on a highly dynamic potential-field approach.
Three main fields of duty can be identified here. These are
the task and topological route planning, the handling of 2.1 Local world model
dynamic objects and the interaction with humans, and finally The local world model is a layer-independent model of the
safe and reliable local navigation. A three-layered navigation robot’s surroundings and inner state. It is divided into several
concept was developed to cover the mentioned three levels corresponding to the degree of abstraction of the
identified tasks (Figure 5). The first layer, called strategic contained data.
layer, is responsible for deliberative long-term planning. It 2.1.1 Data acquisition level
houses the task-planner and the topological navigation. The The reactive behaviors responsible for the obstacle avoidance
tactical layer contains a behavior-based control to deal with work based on an occupancy grid (Elfes, 1989; Thrun, 2002).
dynamic objects in a local area and with the interaction with These grids merge information from various sources like the
nearby humans. The behaviors are based on different two laser-scanners, information received from other robots or
modules for scene analysis like object and user tracking. memorized at an earlier point in time. This level additionally
And finally, at the bottom of the hierarchy, the reactive layer’s houses the inner state of the robot like its position and
goal is to fulfill movement tasks inside a local topological area velocity, the actual task and the target.
379
Robust navigation system based on RFID transponder barriers Industrial Robot: An International Journal
Michael Göller et al. Volume 36 · Number 4 · 2009 · 377 –388
Figure 4 Overview of InBOT’s control hierarchy. The task planner generates a plan based on the command given by the user and the actual mode to
operate in. This plan is processed by the behaviors of the different behavior repertoires step by step
Tasks Modes
Commands
Move to product Guide user
from user Move to user Follow user
interaction Stop Haptic steering
Wait Autonomous
Task
planning Task planner
topological navigation
Advanced
Park left Move to target Turn around …
behaviour
repertoire
Basic
behaviour Avoid obstacles Find gap Straight to target …
repertoire
Fusion
Hardware
Hardware
abstraction
abstraction
layer
Tactical layer information needed for a certain area is the relative positions
of the contained entry points. The semantic map is a semantic
Local navigation - Tactical part extension of the topological map. Each node is enhanced with
Action generation
systems.
Hard ware abraction layer 2.1.3.1 RFID barriers. A crucial question for the topological
navigation is: how does the robot know when it is entering a
new topological area? Usually the robot looks for certain
landmarks. Many of the landmark-based systems for
localization share a common disadvantage: the landmarks
can easily be occluded. Wall- or ceiling-mounted long-range
Reactive Fast RFID-based systems can easily be blocked as well, especially
380
Robust navigation system based on RFID transponder barriers Industrial Robot: An International Journal
Michael Göller et al. Volume 36 · Number 4 · 2009 · 377 –388
Figure 6 The shop (left) is divided into several areas (middle). Each of the areas forms one topological area of the topological map (right)
01 02 03 04 05
06 07 08
34
09 10 11 12 13 14 29
30
35
15 16 17 18
31
32
19 20 21 22 23 24 36
33
25 26 27 28
01 02 03 04 05
06 07 08
29
34
09 10 11 12 13 14
30
35
15 16 17 18 31
32
19 20 21 22 23 24 36
33
25 26 27 28
381
Robust navigation system based on RFID transponder barriers Industrial Robot: An International Journal
Michael Göller et al. Volume 36 · Number 4 · 2009 · 377 –388
Figure 8 Placement of RFID barriers to divide a corridor in different 2.2.1 Task planning
areas The primary job of the task planner is to collect and arrange
the tasks received from the user interface. Usually, the tasks
are given in a certain order. In this case the task planner just
queues up the tasks. But the task planner can sort them by
different policies as well. After that the task planner breaks
down the tasks in several sub-tasks that can be processed by
the lower components. For example, the task “Finish
Shopping” would be divided into “GotoCashier”,
“QueueUp”, and “Wait”.
2.2.2 Topological navigation
If a sub-task is encountered that contains a “Goto” command,
topological navigation takes place. The robot plans a route
from its current topological node to the node that contains the
target. This planning is done by an A * algorithm. It is
enhanced to be able to utilize the semantic information
Figure 9 The largest diameter of the coupling area is found at a provided by the map. Because the topological map is a very
distance of 2 cm from the reader high-level discretization, the navigation takes place in a small
state space, so the processing costs are very low.
20 2.2.3 Route planning
Route planning takes place based on the plan provided by the
15
topological navigation. If the next command is a movement to
an adjacent topological area the relative coordinates of the
10
Z [cm]
–10
–20 –15 –15
–10 –5 0 5 10 –20 2.2.4 Flexible program tree
15 20
X [cm] In each step the generated (sub-)plan is stored in the program
tree (Figure 11). Here, two kinds of nodes are defined: action
nodes and planning nodes. When the plan is executed it is
(Figure 10). The reading performance with a single tag is analyzed from left to right in a depth-first manner. Each time
significantly inferior (Göller et al., 2008b). a planning node is encountered the needed planning step is
performed by the responsible component described above.
2.2 Strategic layer The whole planning process is not performed at once, but step
Tasks coming from the user interface first of all show up at the by step on demand. This way the robot is able to start with
strategic layer. Here, the task planner decomposes the tasks. the first action without delay and no computational effort is
First, the incoming tasks are arranged and concatenated. In wasted if the plan is interrupted.
the topological navigation step the active task is analyzed and Owing to its tree structure the plan is very flexible and can
if necessary a sequence of topological areas is calculated to be changed at runtime without effort (Figure 12). In the
reach the target area. Finally, target points are calculated for depicted case the actual action would be postponed and new
the actual topological area. actions would be inserted. After completing the inserted
actions the robot continues with the postponed action.
Each time an action node is encountered while processing
Figure 10 If using double tags in the barriers the barriers are detected the program tree, the corresponding action is determined and
very reliably at velocities of up to 2 m s2 1 activated. The active action is handed down to an action
140 selection module. The action selection selects the needed
1 RFID tag behaviors to fulfill the active action and motivates them.
120 2 RFID tags
2.3 Tactical layer
Read RFID tags [%]
382
Robust navigation system based on RFID transponder barriers Industrial Robot: An International Journal
Michael Göller et al. Volume 36 · Number 4 · 2009 · 377 –388
Figure 11 Sketch of a plan in a tree structure. On the right side the corresponding steps of the planning process are indicated
Task
planner
Go to
Go to …
<Product 2>
<Product 1>
Move to Move to
Move to Change Move to Change location location
connection area connection area of product 1 of product 2
Route
planner
Figure 12 Plan after the user interrupted the robot and commanded it to stop and to move to a local point
–
Stop!
go over there. Task
Ok, wait here. Planner
Go to Go to ...
<Product 1> <Product 2>
Move to Move to
Move to Change – Change location location
connection area area of product 1 of product 2 Route
planner
Move ot
location Move to
Stop Wait
(X,Y) connection
here for the sake of completeness only. Large parts of the Figure 13 The behavior-module is the basic unit of a behavior-based
interactive and cooperative functionalities were under control. Data flow, via the interfaces (e) and (u), is distinguished from
development while this document was written. control flow, via the interfaces rating (r), motivation (m), activity (a),
and inhibition (i)
2.4 Reactive layer
The main objective of this layer is to provide the repertoire of
Control
basic behaviors (Figure 4), which offers the basic skills of rm a
InBOT. To fulfill the demanded fast response to a : a(f(e,i,m)) : activity
r : r(e) : rating (in terms of satisfaction)
environmental changes, these behaviors act on impulse
m : (external) motivation
based on sensory input only. To acquire a goal-oriented e u = f(e,i,m) U i : (external) inhibition
overall behavior they need to be arbitrated by higher
behaviors. The reactive layer receives a goal from higher Data
behaviors and transforms it into a set-point velocity vector. To e : data input vector
respect additional constraints several other velocity vectors are i
u : data output vector
generated by different behaviors. All of them are finally
merged to a resulting velocity vector which is handed down to
the hardware abstraction layer.
Owing to this loose coupling of the different modules the
2.4.1 Behavior networks present functionalities can easily be recombined. They even
The behavior modules (Figure 13) are the basic units of the can be motivated halfway only, to support some other
behavior networks. They are small software modules that are functionality. This way for example a slight obstacle avoidance
dedicated to a single well-defined task and work augmentation for the haptic steering mode can be
independently and in parallel at all times (Göller et al., implemented (Göller et al., 2009b). Another advantage is
2008b). Individual behaviors can be woven into behavior the easy expandability of the network. New functionalities can
networks. In such a network a whole set of behaviors combine simply be applied by hooking in their corresponding behavior
their strictly dedicated work to fulfill a higher task. modules.
383
Robust navigation system based on RFID transponder barriers Industrial Robot: An International Journal
Michael Göller et al. Volume 36 · Number 4 · 2009 · 377 –388
2.4.2 Reactive layer of the behavior network velocity vector. All these vectors are merged with the vector
The reactive layer is split into four levels itself (Figure 14). from one of the task-oriented behaviors and generate a kind of
The task-oriented level deals with the selection or generation highly dynamic potential field. It depends on many
of (sub-)targets. The behaviors in the object-oriented level are parameters like the status (position, velocity, orientation,
based on a dynamic potential field method using repelling or and task) of the robot and environmental information
attracting vectors as common language. They move the robot (distance, position, velocity of objects, and target). Because
away from obstacles and towards the given (sub-)target. Each of the complexity of the processed information no path
module contributes one vector and all vectors are merged by a planning is performed. The robot calculates the most
fusion behavior. beneficial movement online instead. The data base for this
2.4.3 Task-oriented level level is occupancy maps.
The top level of the behavior network is responsible for
fulfilling a certain task. Possible goal-oriented data sources are 2.4.4.1 Dynamic obstacle handler group. This group consists
found here. They provide either a target location or directly a of two modules. The first one generates reflexes like a vector
set-point velocity vector. In the case of a target location, a that points directly away from a nearby dynamic object to
predictive obstacle handling takes place. Finally, one vector is obtain a safety distance. The second behavior needs a
selected and handed down to the object-oriented level. movement model acquired in the tactical layer. Based on the
2.4.3.1 Predictive obstacle handling. If a target location is movement model a probabilistic temporal repelling field is
known, i.e. the task is a point-to-point movement instead of, calculated and finally merged to a resulting repelling vector.
e.g. the haptic control of the robot, predictive obstacle This group of behaviors is assisted by a more deliberative
handling takes place to overcome the local minima problem of handler for dynamic obstacles. It is found in the tactical layer
potential field approaches (Koren and Borenstein, 1991b) itself. It generates sub-targets to preventively avoid crowded
and to generate more efficient paths. The obstacles are areas or solve blocking situations on a spatiotemporal basis.
scanned for corners, which are then ordered by their quality
depending on their distance and direction compared to the 2.4.4.2 Static obstacle handler group. The structure of the
robot and the target. The best corner is chosen as new sub- corresponding group is shown in Figure 16. Two behaviors
target. This way the robot is driving on a kind of dynamic observe the right-hand and left-hand obstacles. In special
visibility graph. A brief example is given in Figure 15
cases look for gaps is motivated to look for passages and break
(Göller et al., 2008a, 2009a).
tie is activated to force a decision if necessary. The behavior
2.4.4 Object-oriented level for wall following inhibits one of the obstacle avoidance
At this level the task-oriented attracting vector is merged with behaviors. There is a detailed description of these behaviors in
attracting and repelling vectors added by several special Göller et al. (2008a, 2009a). An extract of a Matlab
behaviors. Each of the reactive behaviors generates one simulation of them is shown in Figure 17.
Figure 14 The four levels of the reactive layer containing the behavior network. Behavior groups, or sub-networks, are indicated with G, the special
fusion behaviors with F
Task-oriented level
W W
Mobile (_v,a) (_v,a) (_v,a)
F Obstacle
object Max:
handler (G) select source handler (G)
(_v,a)
F Gew:
Object-oriented level
drive command
(_v,a)
Virtual sensors/ Drive abstraction Drive-oriented level
W
world model
Predictive (_v,a)
safety Safety-oriented level
384
Robust navigation system based on RFID transponder barriers Industrial Robot: An International Journal
Michael Göller et al. Volume 36 · Number 4 · 2009 · 377 –388
Figure 15 The predictive obstacle handler uses corners of visible 3.1 Topological navigation
obstacles as sub-targets which temporarily replace the original target The first focus lies on evaluating the reliability of the detection
of RFID-based artificial landmarks. Tests showed that normal
X barriers (with single RFID tags) where not detected reliably
X enough. At velocities higher than approx. 1 m s2 1 not all
barriers where detected (gray line in Figure 10). When using
pairs of tags instead the robot was able to detect all barriers at
velocities of up to approx. 2 m s2 1, which is sufficient for our
scenario. Additionally the best distance between two pairs of
tags in a barrier (20 cm) and the best distance of the RFID
reader from the ground (2 cm) were determined.
After successfully testing the basic functionality of the global
navigation on different series of movement tasks (Figure 18)
X X the robustness of the system was tested. For example, doors
where closed to simulate impassable obstacles. If a route
existed it was found by the robot. Next the robot was hijacked.
While on its way to a target it was moved away into a
neighboring area. The robot recognized the “mistake” at once
when hitting the wrong barrier and re-planned the route
(Figure 19).
Figure 16 The main behavior modules of the group are responsible for avoiding obstacles. They are assisted by behaviors dedicated to special
situations
Obstacle handler group
W W (a) W (i)
+ Look for ( v–,a) Break ( v–,a)
Follow ( v–,a) gap tie
+
wall
(m) (m)
&
W (i) (a) W (i) (a)
Avoid obstacle –( v ,a) Avoid obstacle ( v–,a)
left right
Fgew:
obstacle handler
385
Robust navigation system based on RFID transponder barriers Industrial Robot: An International Journal
Michael Göller et al. Volume 36 · Number 4 · 2009 · 377 –388
Figure 17 A Matlab simulation illustrates the vectors of the behaviors Figure 20 InBOT solves a U-shaped obstacle. The gray “X”s represent
avoid obstacles (light gray and gray), the task-oriented vector (gray), the corners of the obstacle chosen by the predictive obstacle handler
and the resulting vector (black)
Figure 18 Exemplary topological path in our labs Figure 21 InBOT solving two classical shortcomings of potential field
methods: dead ends and narrow corridors
386
Robust navigation system based on RFID transponder barriers Industrial Robot: An International Journal
Michael Göller et al. Volume 36 · Number 4 · 2009 · 377 –388
The concept presented in this document is designed to be generated for every object. Another specialty of the concept is
independent from a global metrical map. the robot’s predictive obstacle analysis, prior to the application
Robots controlle1d by a behavior-based control approach of the potential field. Additionally the holonomic character of
are hardly found in complex real-world applications like the mecanum drive is taken into account.
museum guides or intelligent shopping carts. They tend to be
very application-specific instead and do not follow the 5. Conclusion and outlook
necessary general-purpose approach required for real-world
applications. InBOT is an approach to close this gap. In this paper, a navigation system for a mobile robot was
There are robots which are indeed used as shopping introduced that was developed especially for highly dynamic
assistants, but are not personalized to one customer. Others environments. A main aspect to deal with was the lack of
do their job as personalized guides in museums. But the metrical maps in such environments. Therefore, the
concept of a combination seems not to be well known. In navigation was divided into two parts.
contrast InBOT will be able to lead a user, to process The strategic part works based on very few static structures
shopping lists or to inform the user about the goods in the assumed to be invariant. The resulting areas are discriminated
basket or to enable self-scanning. It will be fully personalized. by ground-mounted barriers of RFID transponders and build
up a topological map of the location, which is then
4.1 Strategic layer annotated with semantic attributes. Global navigation takes
There are many works focusing on artificial landmarks. For place based on this semantic map. A special focus was the
example, in works by Guo and Xu (2006) and Yoon and Kweon development of a reliable barrier detection. Because of the
(2001), colored visual landmarks are used. They are designed to low number of tags needed and the sparse distribution of
be robust against perspective transformations and easy to barriers in the shop this system is cheap and its installation is
identify. Besides, these camera-based approaches there are not too invasive. In several tests the reliability and robustness
infrared-based landmarks (Tomizawa et al., 2007) or long-range of the system has been proven.
RFID systems (Chae and Han, 2005; Hähnel et al., 2004). By In future work the semantic aspect of this kind of navigation
using these RFID concepts only rough position estimations are will be enhanced. It is still necessary to determine the relative
possible. Therefore, these systems are only used as an coordinates of neighboring barriers. Instead, the navigation
enhancement for other systems. The concept presented in this could be performed on a purely semantic basis. For example,
document proposes ground-mounted RFID transponder the robot could be given “the third door” as waypoint instead
barriers to make occlusions impossible and to strongly increase of actual metric coordinates. Thus, the last remaining
the position accuracy of the system. The disadvantage on the metrical information of the location could be dismissed.
other hand is that position corrections are only possible at the In the reactive part a hierarchical behavior-based control for
very moment the robot passes over an RFID barrier. a holonomic robot was developed. The reactive component is
Topological maps are very compact. They represent large able to avoid collisions and to find a way to the intended
structures like corridors or rooms so they are as far as possible target. The predictive obstacle handler on the other hand
time-invariant (Althaus, 2003). Additionally they are easy to generates efficient paths that are comparable to those
construct even for large places. When considering generated by visibility graph methods. This way the most
human-robot interaction the fact stands out that topological prominent shortcomings of potential field methods are
maps are much easier understood by humans than large negated as well. It should be kept in mind that the robot
metrical ones. Often the topological maps are combined with has no global map knowledge and therefore is only able to
local metrical maps (Kuipers, 2006). They can be constructed plan the path in visibility range of the sensors or within a local
manually, as done for Robox (Siegwart et al., 2003), but often memorized area. The network character of the control makes
the robot learns them automatically (Thrun, 1998). it easy to extend the control with new functionalities either by
hooking in new behavior modules or by recombining already
4.2 Reactive layer present functionalities. This way it is possible to use the
The proposed behavior-based control centers on the very popular obstacle avoidance functionality to augment the steering
potential field methods (Khatib, 1986). These are subject to functionality of the haptic handle and let InBOT move
limitations such as trapping situations or oscillations because of the around obstacles while it is being pushed by its user. Another
close field of view (Koren and Borenstein, 1991), so these methods exemplary augmentation can be achieved by only slightly
are often enhanced (Khatib and Chatila, 1995; Antich, 2005). activating the target-attracting functionality while the robot is
A comparable approach, the virtual force field, is introduced in pushed, so that the robot pulls the user slightly when a
Borenstein and Koren (1990) and later on leads to vector field direction change or a turn is necessary.
histograms (VFH, Borenstein and Koren, 1991; Ulrich and Finally, after developing the basic navigation system, future
Borenstein, 1998). These are based on the occupancy/certainty- work will focus on three main fields of duty: the handling of
grid methods. All relevant cells generate a repelling force. dynamic objects will be finalized, the coordination with other
Generally these methods do not take the robot’s dynamics into robots will be advanced, and the most important aspect, the
account. In contrast the dynamic windows method (Fox et al., interaction with the user, will be elaborated with the help of
1997; Fox et al., 1996) is specially designed to cope with higher user studies and experiments which are now made possible.
velocities where the robot’s dynamics must not be omitted.
The navigation concept provided in this document proposes a
References
potential field method based on occupancy grids encapsulated
in behavior modules. In contrast to usual potential field Althaus, P. (2003), Indoor navigation for mobile robots: control
methods the robot’s dynamics are taken into account. In and representations, PhD thesis, Royal Institute of
contrast to the VFHs only one representative repelling vector is Technology, Stockholm.
387
Robust navigation system based on RFID transponder barriers Industrial Robot: An International Journal
Michael Göller et al. Volume 36 · Number 4 · 2009 · 377 –388
Antich, J. (2005), Extending the Potential Fields Approach to Hähnel, D., Burgard, W., Fox, D., Fishkin, K. and Philipose,
Avoid Trapping Situations, University of the Balearic Islands, M. (2004), “Mapping and localization with RFID
Palma de Mallorca. technology”, Proceedings of the 2004 IEEE International
Böhme, H.-J. and Gross, H.-M. (2006), “Progress in the Conference on Robotics and Automation, pp. S1015-20.
development of an interactive mobile shopping assistant”, Khatib, M. and Chatila, R. (1995), “An extended potential
Proceedings of the Joint Conference on Robotics: 37th field approach for mobile robot sensor-based motions”,
International Symposium on Robotics (ISR 2006) and 4th Proceedings of the International Conference on Intelligent
German Conference on Robotics (Robotik 2006), VDI Verlag, Autonomous Systems (IAS-4).
Munich, Germany, Paper No. 83, p. 20. Khatib, O. (1986), “Real-time obstacle avoidance for robot
Borenstein, J. and Koren, Y. (1990), “Real-time obstacle manipulator and mobile robots”, International Journal of
avoidance for fast mobile robots in cluttered Robotics Research, Vol. 5 No. 1, pp. 90-8.
environments”, Proceedings of the IEEE International Koren, Y. and Borenstein, J. (1991), “Potential field methods
Conference on Robotics and Automation (ICRA), pp. 572-7. and their inherent limitations for mobile robot navigation”,
Borenstein, J. and Koren, Y. (1991), “The vector field histogram Proceedings of the IEEE International Conference Robotics and
– fast obstacle avoidance for mobile robots”, IEEE Journal of Automation.
Robotics and Automation, Vol. 7 No. 3, pp. 278-88. Kuipers, B. (2006), “An intellectual history of the spatial
Chae, H. and Han, K. (2005), “Combination of RFID and semantic hierarchy”, in Jefferies, M. and Yeap, A.W.-K.
vision for mobile robot localization”, Proceedings of the 2005 (Eds), Robot and Cognitive Approaches to Spatial Mapping,
International Conference on Intelligent Sensors, Sensor Networks Springer Verlag, New York, NY, pp. 243-64.
and Information Processing, pp. 75-80. Siegwart, R., Arras, K.O., Bouabdallah, S., Burnier, D.,
Elfes, A. (1989), “Using occupancy grids for mobile robot Froidevaux, G., Greppin, X., Jensen, B., Lorotte, A., Mayor,
perception and navigation”, Computer, Vol. 22 No. 6. L., Meisser, M., Philippsen, R., Piguet, R., Ramel, G.,
Fox, D., Burgard, W. and Thrun, S. (1996), “Controlling Terrien, G. and Tomatis, N. (2003), “Robox at Expo.02: a
synchro-drive robots with the dynamic window approach to large-scale installation of personal robots”, Robotics and
collision avoidance”, Proceedings of the IEEE/RSJ International Autonomous Systems, Vol. 42 Nos 3-4, pp. 203-22.
Conference on Intelligent Robots and Systems (IROS-96). Thrun, S. (1998), “Learning metric-topological maps for
Fox, D., Burgard, W. and Thrun, S. (1997), “The dynamic indoor mobile robot navigation”, Artificial Intelligence,
window approach to collision avoidance”, IEEE Robotics Vol. 99 No. 1, pp. 21-71.
and Automation, Vol. 4 No. 1, pp. 23-33. Thrun, S. (2002), Learning Occupancy Grids with Forward
Glover, B. and Bhatt, H. (2006), RFID Essentials (Theory in Sensor Models, Carnegie Mellon University, Pittsburgh, PA.
Practice), O’Reilly, Andover. Thrun, S., Bennewitz, M., Burgard, W., Cremers, A.B.,
Göller, M., Kerscher, T., Zöllner, J.M. and Dillmann, R. Dellaert, F., Fox, D., Hahnel, D., Lakemeyer, G.,
(2008a), “Behaviour network control for a holonomic mobile
Rosenberg, C., Roy, N., Schulte, J., Schulz, D. and
robot in realistic environments”, paper presented at 11th
Steiner, W. (1999), “Experiences with two deployed
International Conference on Climbing and Walking Robots
interactive tour-guide robots”, Proceedings of the
and the Support Technologies for Mobile Machines, Coimbra.
International Conference on Field and Service Robotics,
Göller, M., Kerscher, T., Zöllner, J.M. and Dillmann, R.
available at: www.cs.washington.edu/al/mobile_Robotics/
(2009a), “Obstacle-handling of the holonomic driven
two-museum-robotics-fsr-99.pdf
interactive behavior operated shopping trolley InBOT”,
Tomizawa, T., Saiki, Y.M., Ohya, A. and Yuya, S. (2007),
Proceedings of the IEEE 7th International Workshop on Robot
“Property modifiable discreet active landmarks”,
Motion and Control (RoMoCo09).
Proceedings of the IEEE International Conference on Robotics
Göller, M., Steinhard, F., Kerscher, T., Zöllner, J.M. and
Dillmann, R. (2008b), “RFID transponder barriers as and Automation, pp. 3420-6.
artificial landmarks for the semantic navigation of Ulrich, I. and Borenstein, J. (1998), “VFH þ : reliable
autonomous robots”, paper presented at 11th International obstacle avoidance for fast mobile robots”, Proceedings of the
Conference on Climbing and Walking Robots and the 1998 IFFF International Conference on Robotics and
Support Technologies for Mobile Machines, Coimbra. Automation Leuven, May, Belgium.
Göller, M., Kerscher, T., Ziegenmeyer, M., Rönnau, A., Viboonchaicheep, P., Shimada, A. and Kosaka, Y. (2003),
Zöllner, J.M. and Dillmann, R. (2009b), “Haptic control “Position rectification control for mecanum wheeled omni-
for the interactive behavior operated shopping trolley directional vehicles”, Proceedings of the Industrial Electronics
InBOT”, Proceedings of the New Frontiers in Human-Robot Society, IECON 2003, The 29th Annual Conference of The
Interaction Symposium at the Artificial Intelligence and IEEE, November, Vol. 1, pp. 854-9.
Simulation of Behaviour (AISB) 2009 Convention. Yoon, K. and Kweon, I. (2001), “Artificial landmark tracking
Graf, B., Hans, M. and Schraft, R.D. (2004), “Care-O-bot II based on the color histogram”, paper presented at IEEE/
– development of a next generation robotic home assistant”, RSJ International Conference on Intelligent Robots and
Autonomous Robots, Vol. 16 No. 2, pp. 193-205. Systems 2001.
Guo, Y. and Xu, X. (2006), “Color landmark design for
mobile robot localization”, paper presented at IMACS
Corresponding author
Multiconference on Computational Engineering in Systems
Applications, Vol. 2, pp. 1868-74. Michael Göller can be contacted at: goeller@fzi.de
388