2011 IEEE International Conference on
Automation Science and Engineering
Trieste, Italy - August 24-27, 2011
FrC4.6
Development of Robotic Laboratory Automation Platform with
Intelligent Mobile Agents for Clinical Chemistry
Byung June Choi 1 , Won Suk You1 , Seung Hoon Shin 1 , Hyungpil Moon1 ,
Ja Choon Koo 1 , Wankyun Chung 2 and Hyouk Ryeol Choi ∗1
Abstract— In this research, we propose an innovative robotic
platform for clinical tests suitable for small or medium sized
laboratories using small-sized multiple mobile robots and a
robotic arm. The proposed robotic platform is referred to
as “BioRobot platform”. The BioRobot platform not only
provides flexibility in test process by carrying out various
clinical tests simultaneously through multiple mobile agents, but
also increases productivity by having controllable throughput
according to amount of tests. Therefore, the various algorithms
which are related to robotic technologies have been applied
in this platform to operate the entire hardware and several
mobile agents simultaneously. To evaluate the performance of
the BioRobot platform, various control methods are implemented, which provides parallel processing and scalability. The
feasibility of the BioRobot platform with three mobile robotic
agents is validated through preliminary experiments.
I. INTRODUCTION
Automated machines are becoming indispensable in today’s laboratory environments, and new technologies employing robotic automation are rapidly incorporated in the
clinical laboratory. However, typical Total Laboratory Automation(TLA) systems used in large hospitals have been
optimized for rapid tests of a large number of samples
by adopting conveyor belts. The conveyor belt systems are
difficult to setup in small or medium sized hospitals since
the size of system is enormous, and a large amount of investment is needed to introduce related facilities. Therefore, in
small or medium sized hospitals clinical tests are performed
manually, or the samples are sent to the institutions with
dedicated facilities for clinical tests. Moreover, the existing
systems still take time to complete the ordered tests because
most of tests are conducted sequentially. In order to realize
the medical diagnosis for patients, an innovative laboratory
system is required to support high flexibility.
Up to now, many researches on analytical techniques or innovative devices have been reported [1]- [18]. Among them,
personalized clinical tests with robotic automation appears
to be attractive as an alternative to the TLA system [4].
Robotic automation is defined as a dedicated robotic system
capable of performing selective laboratory tasks. In general,
robotic automation is flexible and requires less footprint and
financial investment than the TLA system. Shortcomings
of throughput compared to the TLA are compensated with
flexibility of operation under minimum overhead. Robotic
automation systems designed for selective analytical tasks
may better meet the needs of small- or medium-sized laboratories because these systems can either operate stand alone,
or be integrated to form work cells [5]- [7]. Also, the robotic
system can be customized or personalized based on needs
and budgets of individual laboratory [8]- [10].
In this paper, a robotic laboratory automation system
called the “BioRobot platform” is proposed that can be
used for clinical chemistry tests. In past researches, the
issue of hardware implementation and the characteristic of
platform were presented [11] [12] [13]. In this work, the
operation method of the platform with various machines that
is based on parallel processing, scalability and fault tolerance
is introduced. The platform can carry out various tests simultaneously via parallel processing. Also, the mobile agents can
be easily added or removed because the Immediate PlugIn/Out (IPIO) functionality using Bluetooth communication.
In addition, the platform has the advantage of fault tolerance
because the entire system remains unaffected, even if one
of the mobile agents break down [13]. Also, the various
algorithms which is related to robotic technology has been
applied in this “BioRobot platform”, such as localization
method for mobile agents, path planning algorithm for each
mobile agent, and job scheduling algorithm.
This work is financially supported by the Ministry of Knowledge Economy (MKE) and Korea Institute for Advancement in Technology (KIAT)
through the Workforce Development Program in Strategic Technology.
H. R. Choi is with Professor of School of Mechanical Engineering, Sungkyunkwan University, Suwon, The Republic of Korea.
hrchoi@me.skku.ac.kr
W. K. Chung is with Professor of Department of Mechanical Engineering,
Pohang University of Science and Technology, Pohang, The Republic of
Korea. wkchung@postech.ac.kr
978-1-4577-1732-1/11/$26.00 ©2011 IEEE
708
For the automation robotic platform and realization parallel processing…
Mobile agent
Microplate
loading module
Incubator
tor
(with
hotometer)
ter))
spectrophotometer)
Localization
problem of
mobile agents
Docking module
Reagent tray
Miniature arm
m
Path planning
Sample tray
Loading of
sample blood
Identification
of sample
Test planning
& scheduling
Movement of mobile agents
Results & removing
of microplate on the
mobile agent
Microplate loading module
Fig. 1.
Loading
ading of
of
croplate oon
n
microplate
he mobile age
h
geent
g
nt
the
agent
Microplate loading module
ule
Detection using
spectrophotometer
Incubator
Incubation
Dispensing
of sample and
reagent
Job scheduling
ure arm
Miniature
Requirement of algorithmic solution for BioRobot platform
This paper is organized as follows. In section II, the
TABLE I
operating procedure of BioRobot platform is presented. Also,
the algorithmic problems such as localization method, and
path planning algorithm for multiple mobile robots, job
scheduling algorithm for platform are presented in section
III. In section IV, experimental procedures for evaluating
the performance of the Mobile Agents are discussed and the
paper is concluded with summary in section V.
S PECIFICATION OF B IO R OBOT P LATFORM
Hardware components
Miniature arm
Sample tray
II. OPERATING PROCEDURES OF BIOROBOT
PLATFORM
As depicted in Fig. 1, the BioRobot platform is composed
of mobile agents, miniature arm for handling samples and
reagents, microplate loading module to supply microplates
on the mobile agents, sample tray module, reagent chamber
module, incubator module, and photometry scanner [13].
The basic functions of the BioRobot platform are enlisted
as loading of the microplate on the mobile agent, test
scheduling, dispensing of the sample and reagent, planning
and control of mobile agents, incubation, and detection. First,
the samples and reagents are loaded into individual tubes.
Then, the samples and reagents are identified at each tray using RFID tags. The user determines the test sequences using
the provided scheduler software. Now, the BioRobot platform
is automatically operated according to the job schedules. The
microplates are loaded on the mobile agents in the microplate
loading module. The mobile agents transfer the microplates
from the loading module to the miniature robotic arm which
is equipped with the liquid handler. The miniature robotic
arm takes a pipette tip to absorb the samples or reagents
from each of the sample or reagent tubes, and the liquid
handler dispenses each sample or reagent into the well of
the microplate. Then, the mobile agents with the microplate
that contains the mixtures of samples and reagents navigate
to the inside of the incubator module. After the scheduled
incubation time, the spectrophotometer in the mobile agents
scans the sample and provides the test results to the platform
controller via wireless communication. Finally, the mobile
agents with the microplate return to its initial position for
reloading (in the microplate loading module). In addition,
the absorbance sensor integrated multiple mobile agents can
reduce the clinical test procedure and the test time owing to
their parallel processing capability. Therefore, the clinical
test process in the proposed system achieves a parallel
procedure rather than a sequential one. The efficiency and
the speed of a test are to be optimized depending on the
number of operated mobile agents, and flexible scheduling
and robust operation of the entire clinical test are achievable.
III. ALGORITHMIC SOLUTION FOR BIOROBOT
PLATFORM
Whenever several mobile agents are deployed in the same
environment on the platform, there is the need for coordinating their movements to prevent the undesirable situations
such as collision, congestions or deadlocks. Also, if the
initial pose of the mobile agent is unknown or the mobile
agents have been kidnapped, the mobile agent is difficult to
estimate its current position directly from the general position
Reagent tray
Accuracy of
liquid handler
Expected throughput
Dimension(W×D×H)
2×Microplate loading module,
8×Recharging and docking module,
1×Miniature arm, 1×Dispensing channel,
1×Sample tray, 1×Reagent tray,
1×Incubator, 2×Spectrophotometer,
Mobile Agents
4-DOF SCARA type
7mL tubes, 21 sample tubes
(with exclusive pipette tips)
40mL tubes, 70 reagent tubes
(with exclusive pipette tips)
1.0mL pipette,
1.0µL resolution
240 photometric tests/hour
1,560×1,100×1,200(mm)
(from the ground to the base plate
(workspace of Mobile Agents))
tracking method. Therefore, in order to solve these problems,
unique algorithmic approach for BioRobot platform is necessarily required. In this section, a new strategy for global
localization and path planning algorithm of each mobile
agent is presented. In addition, a scheduling algorithm called
RIT (Reduced Idle Time) algorithm is presented, which can
minimize the delay time of the job and provide the parallel
processing.
A. Localization method for multiple mobile agents in
BioRobot platform
Coordinate of mobile robot
Coordinate of magnetic landmark
Y
y3
y3
y3
y2
x1
y2
x2
y2
x3
x1
x2
y1
x3
x2
x3
y1
x1
y1
X
Global coordinate of the platform
Fig. 2. Concept of proposed localization method using magnetic landmarks
A new strategy for global localization using several patterns of magnetic landmarks is presented in this section. The
basic concept of the global localization based on magnetic
landmarks is to measure the magnetic field between the
coordination of the hall sensors of the mobile agent and the
coordinate of the magnet set, as shown in Fig. 2 [14]. For
one set of landmarks is composed of four magnetic bars in
rectangular configuration as shown in Fig. 3 where total six
patterns can be obtained. When the mobile agent is located
at the center of the four-magnets landmark, it can easily read
709
Current position
of mobile robot
: (xi,yi,θ)
y
3
yi
2
4
2
3
2
4
2
2
5
3
5
3
3
3
4
2
4
5
1
3
3
2
1
2
1
1
4
5
4
1
4
O
(a) Arrangement of four magnetic bars in rectangular
configuration
xi
2
1
4
2
x
Equations for pose compensation
Patterns
Landmark 1
Landmark 2
Landmark 3
Landmark 4
Landmark 5
Landmark 6
t l +t l
t l +t l
t l −t l
−t l + t l
xe = 2 2 2 4 ye = 1 1 3 3 θ e = sin −1 1 1 3 3 = sin −1 2 2 4 4
d
2
2
d
−l + l
−l + l
−l − l
l +l
ye = 1 3 θ e = sin −1 1 3 = sin −1 2 4
xe = 2 4
2
2
d
d
(t1 = −1, t2 = −1, t3 = 1, t4 = 1)
Fig. 4. Global localization method for finding current position information
of mobile agent on the platform (Distribution of magnetic landmarks,
generation of virtual map, and Sequence of localization method)
−l + l
l2 + l4
−l − l
−l + l
ye = 1 3 θ e = sin −1 1 3 = sin −1 2 4
2
2
d
d
(t1 = −1, t2 = 1, t3 = 1, t4 = 1)
xe =
−l − l
−l2 + l4
−l + l
l +l
ye = 1 3 θ e = sin −1 1 3 = sin −1 2 4
2
2
d
d
(t1 = −1, t2 = −1, t3 = −1, t4 = 1)
xe =
l +l
l2 + l 4
ye = 1 3
2
2
(t1 = 1, t2 = 1, t3 = 1, t4 = 1)
xe =
l1 − l3
−1 −l2 + l4
= sin
d
d
θ e = sin −1
−l + l
l2 − l 4
−l − l
−l − l
ye = 1 3 θ e = sin −1 1 3 = sin −1 2 4
2
2
d
d
(t1 = −1, t2 = 1, t3 = 1, t4 = −1)
xe =
l −l
−l2 + l4
l +l
l +l
θ e = sin −1 1 3 = sin −1 2 4
ye = 1 3
2
2
d
d
(t1 = 1, t2 = −1, t3 = −1, t4 = 1)
xe =
(b) Equations for compensation of pose errors about six magnetic
landmarks
Fig. 3. Generation of six magnetic landmarks, (a) Arrangement of four
magnetic bars in rectangular configuration, (b) Equations for compensation
of pose errors about six magnetic landmarks
one of four hall sensors is placed in the center region of the
magnetic field which is generated by four magnetic bars [14].
Once this is done, the mobile agent can be controlled to a
location where four hall sensors are properly located on top
of magnetic bars.
The mobile agent identifies the pattern of magnetic landmark by reading the sequence of the magnetic polarities
while rotating, and it obtains the current global position
information. At this point, there still is ambiguity in current
location of the robot because the number of patterns is not
large enough to cover the entire grid-like map. However,
once the mobile agent recognizes the pattern of the current
landmark, position tracking can be done afterward, thus the
mobile agent can move to the one of the neighboring grid
with very small tolerable pose error [14].
B. Path Planning Algorithm
the polarity of the magnets while making a complete turn in
one direction.
In this paper, the patterns of magnetic landmarks are
distributed in such a way of forming a grid-like global map
as shown in Fig. 4 where the number inside of the grid
stands for the type of the landmark pattern. It should be noted
that once the pattern is identified, position tracking is easily
done in this proposed framework. The first step for global
localization is to control the mobile agent so to be situated on
the range of magnetic landmarks. When fairly large size of
magnets is used, the range of magnetic field is large enough.
In this case, a simple random walk algorithm may situate the
mobile agent onto a proper location where the mobile agent
can read the four magnetic fields and begin the procedure of
identifying the pattern of the landmark. Another approach is
to use a gradient descent algorithm for one hall sensor so that
Since the BioRobot platform consisted of the several
mobile robots that are called mobile agents, the moving
trajectories for the individual robots have to be computed
such that collisions between the robots and static obstacles
as well as between the robots among themselves are avoided.
To obtain the moving paths of each mobile agent, The D ∗
lite search algorithm is applied in the proposed BioRobot
platform. It produces an optimal path from the start position
to the goal in the sense of minimizing a pre-defined cost
function.
D∗ lite algorithm can guarantee the least cost paths between a start point and any number of goal points as the cost
of edges between change of point [15]. The basic principle
of D∗ lite algorithm calculates a least cost path from a
start point Sstart ∈S to a goal point S goal ∈S, where S is the
set of point(node) in some finite state space, as shown in
Fig. 5 [16]. To do this, it stores an estimate g(s) of the cost
710
from each state s to the goal [17]. It also stores a one-step
lookahead cost rhs(s i ),
sstart
(Start point)
10
9
f (si ) =
=
8
rhs(si ) + h(sstart , si )
mins′ ∈Succ(s) (c(si , s′ ) + g(s′ )) + h(sstart , si )
7
(1)
rhs(s) =
0
i f s = sgoal
mins′ ∈Succ(s) (c(s, s′ ) + g(s′))
otherwise,
(2)
6
h1
5
h2
4
h3
s1
s2
s3
3
2
where Succ(s) ∈ S denotes the set of successors of s, f (s i )
denotes evaluation function to determine the shortest path,
and c(si , s′ ) denotes the cost of moving from s i to s′ (the
edge cost), g(s′ ) denotes the operating cost function that
actual operating cost have been already traversed, h(s start , si )
denotes the heuristic function that estimates the cost of an
optimal path from state s start to si , and rhs(si ) is the onestep lookahead cost. As with D ∗ Lite uses a heuristic and
a priority queue from f (s i ) to focus its search and to order
its cost updates efficiently. As shown in Fig. 5, starting with
the goal point (node), it maintains a priority queue of nodes
to be traversed. The lower f (s i ) for a given node s i has the
higher its priority. At each step of the algorithm, the node
with the lowest f (si ) value is removed from the queue, the f
and rhs values of its neighbors are updated accordingly, and
these neighbors are added to the queue [16] [17]. The D ∗
algorithm continues until a start point (node) has a lower f
value than any node in the queue. The set of lowest f values
is then the shortest path from start point (node) to goal point
(node).
Therefore, D ∗ Lite expands nodes from the queue in
increasing priority, updating their g-values and rhs-values of
their predecessors, until there is no node in the queue with
a key value less than that of the start point.
1
0
0
1
2
3
4
5
6
7
8
9
sgoal
(Goal point)
10
hi(sstart,si) : Heuristic function
Basic principle of D∗ lite algorithm
Fig. 5.
Machine
number
1
211
2
122
231
223
3
4
131
114
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
Time
14
15
Time
(a) Gantt chart of the SPT algorithm
Machine
number
1
211
C. Job Scheduling algorithm
2
122
Independent n jobs should be processed according to a predetermined procedure using m machines for the minimization
of total processing time. Solutions of this problem have been
studied extensively in the operational research field and this
problem is normally referred to as a “job shop” problem [18].
The popular methods for the job shop scheduling problem
are loading rules, heuristic rules, integer linear programming,
complete enumeration, sampling methods, and learning techniques. The loading rule and the heuristic rule are briefly
reviewed here because those algorithms are used for the
present work. The loading rule progresses by selecting the
next process of the individual machine. The selected processes have to be feasible for the fixed sequence of tests and
have to be finished with the least amount of idle time for each
machine. If many processes are to be done, the job sequence
is to be determined by the so-called shortest processing time
(SPT), which is one of the heuristic rules. The SPT is usually
regarded as the basic algorithm [19].
3
4
131
231
223
114
1
2
3
4
5
6
7
8
9
10
11
12
13
(b) Gantt chart of the proposed RIT algorithm
Fig. 6. Gantt chart of scheduling algorithm by the SPT and RIT algorithm,
(a) Gantt chart of the SPT algorithm, (b) Gantt chart of the proposed RIT
algorithm
Although the SPT may provide the shortest job schedule
for simple cases, this method cannot be applied to a case
such that a single job uses a particular machine more than
twice [20]. An exemplary case shown in Fig. 6 explains the
limitations of the SPT algorithm. The algorithm distributes
jobs to corresponding machines using information stored
in array M. When a situation happens at the middle of
711
(1)
(a) The part of Gantt chart, with which 10 clinical tests are
scheduled by algorithm using one mobile agent. The processing time
is 2,670 seconds
(b) The part of Gantt chart, with which 10 clinical tests are
scheduled by algorithm using three mobile agents. The processing
time is 1,091 seconds
Recognition of
magnetic landmark 1
(3)
a scheduling process such that job 131 is to be inserted,
the only option for the SPT is locating the job at the end
of sequence. Regarding job number 131, 1 stands for job
number 1, 3 is for process number 3 in job 1, and 1 represents
machine number 1 that is used for process 3 of job 1.
Although job 131 can be allocated between job 211 and job
231 to minimize the idling time, the SPT algorithm routinely
assigns the job at the end of sequence because a job uses
a particular machine more than once. The basic principle
of RIT algorithm is to be improved the SPT algorithm
by left-shifting, as shown in Fig. 6. If 131 operation is
assigned at machine 1, the 131 operation is rearranged at
the feasible range of the idle time by searching ranges of the
entire idle time, after the completing time of the previous
operation of the same job as the operation (for example, 122
operation) [20]. Therefore, the RIT algorithm can reduce the
total processing time and tardiness.
IV. EXPERIMENTS AND DISCUSSIONS
In previous research work, the BioRobot platform was
fabricated, and it was tested with the frequently applied
clinical chemistry test items to confirm the merits of the
parallel processing and the scalability using multiple mobile agents. Through further research, the various methods
that were related to robotic technology had been applied
and verified in this BioRobot platform, such as the global
localization technique, the path planning algorithm and job
scheduling algorithm. In the first experiments, the proposed
RIT algorithm was simulated in the operation conditions
of 10 chemistry tests for job schedule. Figure 7 shows the
Gantt charts of simulation results. When one mobile agent
on the BioRobot platform carried out 10 chemistry tests
(ALT, total cholesterol, AST, glucose, TG, total protein,
lactate dehydrogenase, BUN, Creatine, IP), the scheduled
processing time had taken 2,670 seconds, as shown in
Recognition of
magnetic landmark 2
(4)
Recognition of
magnetic landmark 3
Fig. 7. Comparison of the Gantt chart, with which 10 clinical tests are
scheduled by algorithm using multiple mobile agents
(2)
Determination of current
global position and
moving direction
Fig. 8. The experiments to prove the performance of global localization
method using various patterns of magnetic landmarks
Fig. 7. However, when three mobile agents on the BioRobot
platform simultaneously carried out 10 chemistry tests, the
scheduled time had taken 1,091 seconds. It was possible to
confirm that the total processing time could be minimized in
clinical chemistry tests. In the second experiments, in order
to evaluate the performance of global localization technique
of mobile agents, the experiments were carried out as shown
in Fig. 8. The mobile agent was placed in an arbitrary
position on the platform, and then four hall sensors of mobile
agent looked for the changes of magnetic poles during the
rotation. They then identified the magnetic landmark. The
mobile agent moved to the two neighboring landmarks,
and identified the pattern of the neighboring landmarks
using the four hall sensors. Based on these information,
the robot compares the measured three neighboring patterns
with the map information that is saved in advance, the robot
then estimates its current position and moving direction, as
shown in Fig. 8. Finally, the overall BioRobot platform was
operated by reproducing the chemistry tests, as shown in
Fig. 9. First, the mobile agents moved towards the microplate
loading module. Then, the microplates were loaded into the
each mobile agent. Next, the mobile agents moved to the
pneumatic pipettor of miniature arm, and then each sample
or reagent was dispensed to the well of microplate on the
mobile agent. The microplate containing the mixtures of the
sample and the reagents was transferred to the incubator
module by the mobile agent, and the results of tests were read
with a spectrophotometer inside of incubator module. Finally,
the mobile agents that had completed the work returned
to the initial position for the next clinical test. According
to the experiments, the mobile agents could perform the
712
ordered clinical tests by means of distributed robotic system.
Therefore, it was possible to confirm that the BioRobot
platform can be used in chemistry tests.
(1)
Preparation of the tests
(waiting time, localization)
(3)
(2)
Loading of the microplate
on the mobile agent 1
(4)
Movement of the
mobile agent 1, and addition
of mobile agent 3
(5)
Completion of tests of
mobile agent 2, and movement of
mobile agent 2 from incubator
to microplate loading module
(7)
Movement of mobile agent 1 and 3,
and movement of mobile agents
from miniature arm to incubator
Dispensing of samples
and reagents on the
mobile agent 1 and 2
(6)
Movement of mobile agent 1, and
removal of mobile agent 2 that
is caused by abnormal operation
(8)
Incubation and detection
Fig. 9. Operation of BioRobot platform that reproduces the sequence of
the real clinical tests
V. C ONCLUSIONS
In this research, we propose an innovative robotic platform
for clinical tests suitable for small or medium sized laboratories using small-sized multiple mobile robots and a robotic
arm. To evaluate the performance of the BioRobot platform,
RIT(Reduced Idle Time) scheduling is implemented. Also,
the feasibility of the BioRobot platform with three mobile
robotic agents is validated through preliminary experiments
by applying the robotic technologies such as motion control,
localization, and path planning algorithm, etc. As ongoing
researches, the reliability of the system on the real circumstances will be carried, and as the results, we will add the
other functionalities of tests such as immunoassay analysis
etc. and the complete BioRobot system for clinical tests will
be built up.
VI. ACKNOWLEDGMENTS
This work is financially supported by the Ministry of
Knowledge Economy (MKE) and Korea Institute for Ad-
vancement in Technology (KIAT) through the Workforce
Development Program in Strategic Technology.
R EFERENCES
[1] M. Sasaki, T. Kageoka, K. Ogura, H. Kataoka, T. Ueta, S. Sugihara,
“Total laboratory automation in Japan past, present and the future”,
Clinica Chimica Acta, vol. 278, 1998, pp. 217-227.
[2] J. W. Holman, T. E. Mifflin, R. A. Felder, L. M. Demers, “Evaluation
of an Automated Preanalytical Robotic Workstation at Two Academic
Health Centers”, Clinical Chemistry, vol. 48, no. 3, 2002, pp. 540-548.
[3] J. Zakowski, D. Powell, “The Future of Automation in Clinical
Laboratories”, IVD Technology Magazine, July 1999, pp. 48-59.
[4] R. A. Felder, G. J. Kost, “Modular Stepwise Automation”, Medical
Laboratory Observer, vol. 30, 1998,pp. 22-27.
[5] S. Graves, B. Holman, R. A. Felder, “Modular Robotic Workcell for
Coagulation Analysis”, Clinical Chemistry, vol. 45, no. 5, 2000, pp.
772-777.
[6] R. A. Felder, “Modular workcells: modern methods for laboratory
automation”, Clinica Chimica Acta, vol. 278, 1998, pp. 257-267.
[7] S. Graves, H. Holman, M. Rossetti, C. Estey, R. A. Felder, “Robotic
automation of coagulation analysis”, Clinica Chimica Acta, vol. 278,
1998, pp. 269-279.
[8] J. Boyd, “Robotic Laboratory Automation”, Science, Washington, D.
C.,vol. 295m, 2002, pp. 517-518.
[9] R. Petersen, M. G. Bissell, “Leveling the playing field: the economics
of robotics in the hospital clinical lab”, Medical Laboratory Observer,
vol. 30, 1998,pp. 42-45.
[10] D. R. Meldrum, C. H. Fisher, M. P. Moore, M. Saini, M. R. Holl, W. H.
Pence, S. E. Moody, D. L. Cunningham, P. J. Wiktor, “ACAPELLA5K, A High-Throughput Automated Genome and Chemical Analysis
System”, Intl, Conference on Intelligent Robots and Systems, 2003,
pp. 2321-2328.
[11] B. J. Choi, S. M. Jin, S. H. Shin, J. C. Koo, S. M. Ryew, J. H. Kim,
W. H. Son, K. T. Ahn, W. K. Chung, H. R. Choi, “Development of
Flexible BioRobot Platform for Integrated Clinical Test”, Journal of
the Association for Laboratory Automation, 2008, vol. 13, no. 2, pp.
90-96.
[12] B. J. Choi, S. M. Jin, S. H. Shin, J. C. Koo, S. M. Ryew, J. Kim,
W. H. Shon, K. T. Ahn, W. K. Chung, H. R. Choi, “Development of
Flexible Laboratory Automation Platform using Mobile Agents in the
Clinical Laboratory ”, In Proc. IEEE Intl, Conference on Automation
Science and Engineering, 2008, pp. 918-923.
[13] B. J. Choi, W. S. You, S. H. Shin, H. Moon, J. C. Koo, W. K. Chung
and H. R. Choi, “Robotic laboratory automation platform based on
mobile agents for flexible clinical tests”, In 2010 IEEE International
Conference on Automation Science and Engineering (CASE), 2010,
pp. 186-191.
[14] W. S. You, B. J. Choi, B. Kim, H. Moon, J. C. Koo, W. K.
Chung and H. R. Choi, “Global localization for a small mobile robot
using magnetic patterns”, In 2010 IEEE International Conference on
Robotics and Automation (ICRA), 2010, pp. 2618-2623.
[15] S. Koenig and M. Likhachev, “Improved fast replanning for robot
navigation in unknown terrain”, Proceedings of the IEEE International
Conference on Robotics and Automation (ICRA), 2002, pp. 968-975.
[16] S. Koenig and M. Likhachev, “D* Lite”, In Proceedings of the
National Conference on Artificial Intelligence (AAAI), 2002.
[17] M. Likhachev, D. Ferguson, G. Gordon, A. Stentz and S. Thrun,
“Anytime Dynamic A*: An Anytime, Replanning Algorithm”, In
Proceedings of the International Conference on Automated Planning
and Scheduling (ICAPS), 2005.
[18] R. Schafer, “Concepts for Dynamic Scheduling in the Laboratory”,
Journal of the Association for Laboratory Automation, vol. 9, no. 6,
2004, pp. 382-397.
[19] J. W. Kim, B. J. Choi, K. H. Noh, H. R. Choi, J. C. Koo, S. M. Ryew,
J. H. Kim, W. H. Son, “Automatic Scheduling Algorithm for Personalized Clinical Test”, SICE-ICASE International Joint Conference, 2006,
5823-5827.
[20] S. H. Shin, B. J. Choi, S. M. Ryew, J. W. Kim, D. S. Kim, W. K.
Chung, H. R. Choi and J. C. Koo, “Development of an Improved
Scheduling Algorithm for Lab Test Operations on a Small-Size Bio
Robot Platform”, Journal of the Association for Laboratory Automation, 2010, vol. 15, no. 1, 15-24.
713