Professional Documents
Culture Documents
2, FEBRUARY 2015
I. I NTRODUCTION
Fig. 1. Schematic representing the two stages in nanoimprint lithography:
the pre-defined structures on the template to the substrate. Since its first publication by Chou et al. [1] in 1996, many
researchers have devoted themselves to this area and developed
Manuscript received August 11, 2014; revised August 31, 2014; accepted different variations and implementations. NIL is believed
September 24, 2014. Date of publication September 30, 2014; date of to be a promising technology to replace the conventional
current version November 26, 2014. This work was supported in part by
the National Natural Science Foundation of China under Grant 51275018 photolithography method for semiconductor industry due to
and Grant 51475017, in part by the Specialized Research Fund for the Doc- its low cost, high throughput and high resolution character-
toral Program of Higher Education of China under Grant 20131102110010, istics [2]–[4]. In a general plate-to-plate (P2P) nanoimprinter,
and in part by the Innovation Foundation of BUAA for Ph.D.
Graduates under Grant YWF-14-YJSY-010. The associate editor coordi- parallel alignment between the template and the substrate plays
nating the review of this paper and approving it for publication was a critical role in assuring the uniformity of the imprinted
Dr. Stefan J. Rupitsch. structures [5], [6]. Compared to position feedback control,
J. Jiang, W. Chen, and J. Liu are with the School of Automa-
tion Science and Electrical Engineering, Beihang University, Bei- force control is a better alternative because force signal is
jing 100083, China (e-mail: junjiang@buaa.edu.cn; whchen@buaa.edu.cn; believed to be more direct and is less likely to be influenced by
jmliubuaa@gmail.com). the thickness of the photoresist layer. A force sensor with high
W. Chen is with the Mechatronics Group, Singapore Institute of Manufac-
turing Technology, Singapore 638075 (e-mail: whchen@buaa.edu.cn). performance is therefore required to equip the nanoimprint
J. Zhang is with the School of Mechanical Engineering, Beihang University, machine.
Beijing 100083, China (e-mail: jbzhangbuaa@gmail.com). Unlike conventional applications, NIL has raised a strict
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. standard for developing force sensors. Fig. 1 shows the simpli-
Digital Object Identifier 10.1109/JSEN.2014.2360885 fied schematic of a typical P2P nanoimprinter. The imprinting
1530-437X © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
JIANG et al.: OPTIMUM DESIGN OF A DUAL-RANGE FORCE SENSOR 1115
process mainly includes two phases. The first phase starts from To address this problem, this paper presents a novel dual-
the imprint head moving downward to the template touches the range force sensor that is capable of resolving forces with
substrate. In this phase, the small initial contact force will be two different sensitivities within a wide range, meanwhile
used to fine tune the orientation of the imprint head for parallel uses a multiobjective particle swarm optimization method [15]
alignment. To measure this small contact force, the force to compromise the three characteristics. This optimization
sensor should be made extremely sensitive. The second phase process also allows the designer to make an informed decision
is when the template fully presses the substrate. In this phase, by seeing a wide range of optimal trade-off solutions from an
the imprint force may go up to several hundreds of Newton. overall standpoint.
To monitor this force, the measurable range of the force At the present time, a plate-to-plate (P2P) nanoimprint
sensor should be wide enough. In addition, the nano-imprinter machine is being constructed in our team. A dual-range force
needs to fit into a step-and-repeat imprint process. The step- sensor is therefore required to equip this device to provide
and-repeat nanoimprint lithography is promising technique to real-time force feedback during the entire imprinting process.
replicate nanoscale pattern at low cost across a large area [7]. The proposed force sensor has two unique features. The first is
The repeat frequency ranges from several Hz to several tens the dual-range characteristic. The developed sensor structure
of Hz. In order to adapt itself to the imprint frequency, the possesses two different stiffness values which enables the force
force sensor should include relatively a high bandwidth which sensor to increase its measurement range without sacrificing
is defined as the frequency at which the phase response is its sensitivity, such a device has not been reported previously
30° lagged. Normally, the higher the bandwidth, the quicker in the literature. The second is the optmized performances.
the mechanical response. In order to obtain high sensitivity, broad bandwidth, and large
However, conventional force sensors are usually limited measurement range at the same time, a new optimization
in range by their sensitivities. The sensor either has a high procedure has been developed to optimize the sensor’s
sensitivity with small range or has a large range with low geometric parameters. Very few researchers have tried to
sensitivity. Furthermore, the sensitivity and bandwidth are optimize all the three performances simultaneously.
two conflicting characteristics. Take the case of a common The rest of this paper proceeds as follows. The sensor model
force sensor [8], most of the time, it is made of some is described in Sec. II, where its operational principle and
beam-like structures, each one bending easily around one some general analysis are discussed. Then, two multiobjective
axis. The measurement of the structure displacement or strain design optimization problems formulated to optimize the force
is related to the force in that direction. Thus, in order to sensor are presented and solved in Sec. III. Sec. IV details the
obtain greater sensitivity, the beam-like structures need to optimization results, which successively address the conflicting
be made more compliant. On the contrary, to achieve higher requirement of high sensitivity, broad bandwidth and large
bandwidth, the structures should be made more stiff, resulting measurable range of the force sensor. Afterward, a series of
in a higher resonant frequency. Hence the paradox. Tradi- simulations and prototype tests are implemented in Sec. V.
tional force sensors typically optimize one performance at the The later section sketches the nanoimprint machine embedding
expense of the others [9]–[11], except the work published by the current and future developments. Finally, some concluding
Wood et al. [12]. They optimized the sensitivity and bandwidth remarks are made in Sec. VI.
of a two-dimensional force sensor by maximizing the sensor
quality, Q, which is defined as the product of the sensitivity II. S ENSOR M ODEL OVERVIEW
√ natural frequency of the sensor body (Q ≡ f × 1/k ∝
and
A. Operating Principle and Concept Design
1/ (k · m)). The reasons why the authors can use such
method lie in two aspects: 1) the optimization problem only According to the aforementioned analysis, the different
have two objectives, 2) the two objectives are to maximize and phases in nanoimprint lithography have raised a strict require-
minimize the corresponding performance, respectively. If there ment for the development of force sensors. The developed
are more objective functions or the objectives are both to force sensor should include at least a high sensitivity and
maximize or minimize the performance, this method cannot a wide range to provide real-time force feedback for both
be used. parallel alignment and imprint force maintaining. For this
Regarding the optimization of sensor structures, the litera- reason, we propose the conceptual design of a dual-range
ture can mainly be classified into three categories, i.e. 1) Use force sensor. Its first range is used for small scale and
topology optimization method to optimize the sensor’s shape high sensitivity force measurement, while its second range is
in order to realize a specific function [13]. 2) Use DOE to employed for large scale and low sensitivity force measure-
test the sensor’s performance and therefore find the optimum ment. This function is realized by using the concept of contact-
solution [14]. 3) Relate the sensor model to the sensor perfor- aided compliant mechanisms (CCMs). CCMs are a class of
mance and use a specific algorithm to optimize the geometric compliant mechanisms where the compliant members come
variables [9]. A survey of the literature reveals that most into contact with one another to perform a specific task or to
existing work optimized only one or two performances of a improve the performance of the mechanism itself [16].
force sensor and leaving the other performances unoptimized, Fig. 2 shows the operating principle of the sensor body,
seldomly has anyone tried to optimize the sensor performance where the two springs, which have the effective stiffness
comprehensively. values of kin and kout , represent the two groups of flexures
1116 IEEE SENSORS JOURNAL, VOL. 15, NO. 2, FEBRUARY 2015
Fig. 3. Illustration of a dual-range force sensor: (a) the CAD model and (b) its projection in the x-y plane.
where f 1 is the natrual frequency in the first range. The where σ yield is the yield stress of the material, S f is the safty
parameters A, B, C and D are defined as follows: factor selected according to the designer’s experience.
kin + kout −kin
A= , B= ,
m int + m eq1 m int + m eq1 III. PARAMETERS O PTIMIZATION
−kin kin The goal of developing a high performance force sensor is
C = , D= . (10)
m mov + m eq2 m mov + m eq2 to obtain high sensitivity, broad bandwidth, large measurement
Substituting Eq. (10) into Eq. (9), the natural frequency of range and long fatigue life. There are totally six parameters
sensor body in the first range is obtained. It is obvious that that define these performances, wherein the width of the two
f 1 is lower than f 2 , hence the bandwidth of the force sensor groups of the flexures are set to be equal to simplify the
is determined by f 1 . optimization procedure. In order to determine the optimal cross
section of the flexures, two multiobjective optimization prob-
lems with three objectives each are formulated. The geometric
D. Working Range and Stress Analysis parameters to be optimized are lout , tout , lin , tin , and win . The
Generally, the sensor’s measurable range is associated with optimization problems are defined by the following equations.
the stroke of the compliant mechanism. The longer the stroke, Problem 1:
the larger the measurable range. Considering that the two
motion ranges of the force sensor play different roles during Min. k2 = 4Etin
3
win /lin
3
the measurement. The first range is intended for high 3 w /l 3
4Etin
1 in in
sensitivity and small range force measurement, while the Max. f 2 =
2π 4 × 35 × ρ × lin win tin + m mov
26
second range is specialized for large range force measurement.
Therefore, for the first range, we only calculate the maximum Max. x = σ yield lin
2
/3S f Etin +δ
von-Mises stress generated, and try to minimize this stress Subject to. lblin ≤ lin ≤ ublin
value to get a longer fatigue life in the next section. While for lbwin ≤ win ≤ ubwin
the second range, we analyze the maximum working distance
lbt in ≤ tin ≤ ubt in
of the sensor body.
Problem 2:
In view of the sensor structure in Fig. 3(b), the maximum
von-Mises stress generated in the first range can be calculated Min. k1 = kin kout /(kin + kout )
⎛ ⎞
by [18], [19]
1 A+D A−D 2
Max. f 1 = ⎝ − + BC ⎠
3Etout δ 3kout Etin δ 2π 2 2
σ1 = max
max
2
, 2
(11)
lout kin lin
3Et out δ 3k out Et in δ
where δ is the clearance between the base frame and the Min. σ1max = max 2
, 2
intermediate stage. lout kin lin
When measuring in the second range, only the inner group Subject to. wout = win
of flexures are stretched. To avoid plastic deformation of the S f σ1max < σ yield
flexures, the maximum working distance is delimited by the lblout ≤ lout ≤ ublout
yield stress of the material. It is of the form
lbt out ≤ tout ≤ ubt out
σ yield lin
2
As mentioned earlier, the inner group of flexures solely
x = + δ. (12)
3S f Etin affect the performance of the sensor in the second range.
1118 IEEE SENSORS JOURNAL, VOL. 15, NO. 2, FEBRUARY 2015
TABLE I TABLE II
L OWER AND U PPER B OUNDS ON THE D ESIGN VARIABLES R ESULTS OF G EOMETRIC O PTIMIZATION
A. FEA Simulation
The safety factor of the material S f is set to be 1.5 based on The finite element analysis was performed in the COMSOL
the authors’ experience. software package with Solid Mechanics module. For static
Since there are three objective functions in each design performance evaluation, non-linear analysis with large defor-
optimization problem, the obtained Pareto fronts are 3D in mation and frictionless contact was used. A gradually
nature. However, it could be hard to visualize the results increased force was applied to the movable stage to produce
in 3D. Hence, the 2D plots are utilized here. For each design smaller and larger motions. The contact area was meshed
optimization problem, comparing two objectives at a time with extreme fine tetrahedron elements. Fig. 7 shows an
will generate three plots, so six plots will be given in total. example of the resulting deformation and the position at which
For brevity, we use a specialized 2D plot to demonstrate the the maximum stress occurs during the first motion range.
result of each problem. In the specialized 2D plots, the natural It can be observed from the simulation that both groups of
frequency in the second range and the maximum von-Mises flexures deform when the force is less than 10 N. Afterward,
stress in the first range are represented by the size and color of the intermediate stage came into contact with the stoppers,
the marker. Since the problems are multiobjective optimization only the inner group of flexures deformed. This phenomenon
problems, the results are not one optimal solution but a set confirms the design concept of the dual-range force sensor.
of optimal designs that are part of the Pareto fronts. Fig. 5 The results also reveal that the maximum allowable force
and Fig. 6 show the results of the two optimization problems, is around 210 N. The difference between analytical and
respectively. FEA results is about 5%. The error mainly arises from
When optimizing in the second range, the main objective the assumptions used for the analytical model, which only
is to obtain larger measurement range that is proportional consider the bending deformations of the leaf flexures.
1120 IEEE SENSORS JOURNAL, VOL. 15, NO. 2, FEBRUARY 2015
Fig. 10. Frequency response used to determine the first resonent mode.
Fig. 9. Static calibration curve (output voltage versus reference force).
TABLE III
OVERALL C HARACTERISTICS OF THE F ORCE S ENSOR
voltage profile of the dual-range force sensor, the loading
forces range from 0 to 8 N in 1 N setps, 8 N to 12 N in
0.4 N steps, and 12 N to 110 N in 10 N steps, respectively.
The loading ranges were selected according to the sensitivity
of the sensor considering the simulation results. The forces
were added on top of the sensor and then removed gradually.
Totally four loading-unloading cycles were performed in a
row without resetting zero. Fig. 9 shows the data recorded
during the whole characterization protocol. The slope of the
plot represents the sensitivity in each range. It can be seen TABLE IV
from the plot that the developed force sensor has a sensitivity OVERALL C HARACTERISTICS OF THE F ORCE S ENSOR
of 485 mV·N−1 within 0 to 11.5 N, and a sensitivity of
88.7 mV · N−1 within 11.5 N to 110 N, which confirms
its’ dual-range performance. Fig. 9 also demonstrates that
the sensor is linear and provides high repeatability and low
hysteresis.
For dynamic characterization, the performance was exam- It can be seen from the table that the designed force sensor
ined by the frequency response method. We used an impulse is better than the commercial one. Besides, it is worth noting
signal to excite the movable stage, the resulting vibration that the commercial dual-range sensor has to manully switch
was recorded by the capacitive sensor. The obtained signal between its small range and large range when used for different
was transmitted to a terminal equipment for Fourier analysis. applications. However, the proposed design can automatically
Since the bandwidth of the force sensor is delimited by make this transition, which means the proposed design is more
the first natural frequency which occurs in the first motion practical. In the later section, we give a brief introduction to
range, the impulse excitation was applied with a relatively the nanoimprinter that is currently been developed within our
low magnitude. Fig. 10 depicts the frequency response in the team.
working direction. Although the two ranges of the force sensor
should correspond to two different resonant frequencies, the
C. Presentation of the Targeted Nanoimprint Machine
bandwidth of the force sensor is delimited by its first resonant
mode. It is obvious that the resonant frequency in the second A solid model of the nanoimprint machine where the devel-
range is much higher than that in the first range, so only one oped force sensor will be implemented is shown in Fig. 11.
resonant peak is presented here. We believe it is sufficient The imprinter mainly incorporates the following four units:
to determine the sensor’s bandwidth. As one can see, the • Driving unit for z axis. It consists of an AC servo motor,
first natural frequency is around 290 Hz. This value is 17% a precision ball screw and four guiding rails. This unit is
lower than the analytical model, but is very close to the FEA the main source of the imprint force.
result. After the characterization, the overall performance of • Parallelism error adjustment unit. This unit is built up
the developed force sensor are summarized in Table III, with by the combination of a decoupled XY compliant stage,
the theoretical results included for comparison. Furthermore, a spherical air bearing and two linear motors in order
in order to show the superiority of the proposed design, to adjust the parallelism error once an uneven force
we compare its performance with a commercial dual-range distribution is detected.
force sensor developed by Vernier Software & Technology • Imprint force monitoring unit. In this unit, three circum-
Co., Ltd. [25]. The comparison results are listed in Table IV. ferentially distributed force sensors are arranged behind
1122 IEEE SENSORS JOURNAL, VOL. 15, NO. 2, FEBRUARY 2015
the template holder to record the imprint force and broad bandwidth and large measurable range, this paper
to provide real-time force feedback during the entire used a multiobjective particle swarm optimization algorithm
imprinting process. This unit is of great importance to to optimize the geometric parameters of the force sensor.
the force uniformity on the template which in turn has a Thanks to the optimization, the developed sensor achieves
direct impact on the quality of the imprinting result. a monolithic and compact structure. A prototype sensor has
• Heating unit. This unit is used to heat the substrate to a been manufactured by electrical discharge machining. Both the
certain temperature in order to provide fluidity. static and dynamic performance of the force sensor have been
As mention ealier, the whole imprinting process includes tested. The results indicate that the developed sensor is linear,
two phases. When working in the first phase, each force and shows low stress concentration, high repeatability and low
sensor operates in its small motion range with a relatively high hysteresis. Additionally, the sensor is versatile and very easy
sensitivity. Once an uneven force distribution is detected, the to adapt to specific applications.
signal will be transferred to the imprint force monitoring unit. Experimentally determining the impact of the dual-range
Then, the two linear motors will pull or push the compliant force sensor on the performance of nanoimprint lithography
stage that is connect to a peg to drive the spherical air bearing is part of our future work. This work will include control
for tilting movement, thus adjusting the parallelism error. The algorithm designing, parallel error adjusting and imprint test-
non-contact air bearing is used due to its frictionless and high ing. Although the testing is beyond the scope of the current
motion precision capabilities. During the second phase, the paper, we can comment on the expected impact of the sensor
imprint force goes up to several hundreds of Newton, all the on the performance of nanoimprint lithography. Since the
force sensors step into the large motion ranges to monitor whole imprinting process switches from a non-contact environ-
the holding force. The high sensitivity and large measurable ment to a contact environment, we plan to use the impedance
range of the force sensor guarantees that the sensor can force tracking control algorithm as the main control scheme.
work properly during whole process. At the present time, It is anticipated that the developed dual-range force sensor will
the machine is being built within our team. The next step implement the desired parallel error adjustment process and
is to design an adequate control law based on the force the imprinting process. We will also consider using statistical
measurement, and to test the control law on the device. approach to calibrate the whole system’s uncertainty.
[5] Z. Wen, Z. Dong, P. Liu, and H. Ding, “Design of a fine alignment Jun Jiang was born in Nantong, China, in 1989.
system featuring active orientation adjustment for nano imprint litho- He received the B.Eng. degree from the Univer-
graphy,” Rev. Sci. Instrum., vol. 85, no. 3, pp. 035106-1–035106-11, sity of Science and Technology Beijing, Beijing,
2014. China, in 2010, and the master’s degree from the
[6] K.-B. Choi and J. J. Lee, “Passive compliant wafer stage for single- School of Automation Science and Electrical Engi-
step nano-imprint lithography,” Rev. Sci. Instrum., vol. 76, no. 7, neering, Beihang University, Beijing, where he is
pp. 075106-1–075106-6, 2005. currently pursuing the Ph.D. degree in control engi-
[7] H. Yoon, H. S. Cho, K. Y. Suh, and K. Char, “Step-and-repeat process neering. His research interests include robotics, com-
for thermal nanoimprint lithography,” Nanotechnology, vol. 21, no. 10, pliant mechanism design, sensor development, and
pp. 105302-1–105302-6, 2010. micromanipulation.
[8] M. Turkseven and J. Ueda, “Analysis of an MRI compatible force
sensor for sensitivity and precision,” IEEE Sensors J., vol. 13, no. 2,
pp. 476–486, Feb. 2013.
[9] Z. Wang and H. Hu, “Analysis and optimization of a compliant
mechanism-based digital force/weight sensor,” IEEE Sensors J., vol. 5,
no. 6, pp. 1243–1250, Dec. 2005.
[10] A. Buttafuoco, C. Lenders, R. Clavel, P. Lambert, and M. Kinnaert,
“Design, manufacturing and implementation of a novel 2-axis force Weihai Chen (M’00) received the B.Eng. degree
sensor for haptic applications,” Sens. Actuators A, Phys., vol. 209, from Zhejiang University, Hangzhou, China, in
pp. 107–114, Mar. 2014. 1982, and the M.Eng. and Ph.D. degrees from
[11] P. Baki, G. Székely, and G. Kósa, “Design and characterization Beihang University, Beijing, China, in 1988 and
of a novel, robust, tri-axial force sensor,” Sens. Actuators A, Phys., 1996, respectively. He has been with the School
vol. 192, pp. 101–110, Apr. 2013. of Automation Science and Electrical Engineer-
[12] R. J. Wood, K.-J. Cho, and K. Hoffman, “A novel multi-axis force sensor ing, Beihang University, as an Associate Professor
for microrobotics applications,” Smart Mater. Struct., vol. 18, no. 12, sicne 1998 and as a Professor since 2007. He has
pp. 125002-1–125002-7, 2009. authored over 200 technical papers in referred
[13] B. Zhang, C. Lu, and H.-Z. Huang, “Topology optimization of piezo- journals and conference proceedings. His research
electric force sensor,” in Proc. IEEE Int. Conf. Apperceiving Comput. interests include bioinspired robotics, precision
Intell. Anal., Chengdu, China, Dec. 2008, pp. 132–136. mechanism, micromanipulation, and parallel manipulators.
[14] K. R. Evans, E. Lou, and G. Faulkner, “Optimization of a low-cost force
sensor for spinal orthosis applications,” IEEE Trans. Instrum. Meas.,
vol. 62, no. 12, pp. 3243–3250, Dec. 2013.
[15] C. A. C. Coello, G. T. Pulido, and M. S. Lechuga, “Handling multiple
objectives with particle swarm optimization,” IEEE Trans. Evol. Com-
put., vol. 8, no. 3, pp. 256–279, Jun. 2004.
[16] N. D. Mankame, “Investigations on contact-aided compliant mech-
Jingmeng Liu received the B.Eng. degree in elec-
anisms,” Ph.D. dissertation, Dept. Mech. Eng. Appl. Mech., Univ. trical engineering from Anhui Polytechnic Univer-
Pennsylvania, Philadelphia, PA, USA, 2004. sity, Wuhu, China, in 1991, and the M.Eng. and
[17] Q. Xu, “A novel compliant micropositioning stage with dual ranges and Ph.D. degrees in mechanical-electronic engineering
resolutions,” Sens. Actuators A, Phys., vol. 205, pp. 6–14, Jan. 2014. from Beihang University, Beijing, China, in 2000
[18] H. Tang and Y. Li, “Design, analysis, and test of a novel 2-DOF and 2004, respectively. He was with the School
nanopositioning system driven by dual mode,” IEEE Trans. Robot., of Automation Science and Electrical Engineering,
vol. 29, no. 3, pp. 650–662, Jun. 2013.
Beihang University, from 2006 to 2009, as a Lec-
[19] Q. Xu, “New flexure parallel-kinematic micropositioning system with turer, where he has been an Associate Professor
large workspace,” IEEE Trans. Robot., vol. 28, no. 2, pp. 478–491, since 2010. He has authored over 60 technical
Apr. 2012. papers. His research interests include robotics, motor
[20] C. A. C. Coello, “Evolutionary multi-objective optimization: A historical
control, and CNC.
view of the field,” IEEE Comput. Intell. Mag., vol. 1, no. 1, pp. 28–36,
Feb. 2006.
[21] S. Suresh, P. B. Sujit, and A. K. Rao, “Particle swarm optimiza-
tion approach for multi-objective composite box-beam design,” Com-
pos. Struct., vol. 81, no. 4, pp. 598–605, 2007.
[22] M. K. Gill, Y. H. Kaheil, A. Khalil, M. McKee, and L. Bastidas,
“Multiobjective particle swarm optimization for parameter esti-
mation in hydrology,” Water Resour. Res., vol. 42, no. 7, Wenjie Chen (M’06) is currently a Senior Research Engineer with the Mecha-
pp. W07417-1–W07417-14, 2006. tronics Group, Singapore Institute of Manufacturing Technology, Singapore,
[23] S.-Z. Zhao, M. W. Iruthayarajan, S. Baskar, and P. N. Suganthan, “Multi- where he performs research on micromanipulation.
objective robust PID controller tuning using two lbests multi-objective
particle swarm optimization,” Inf. Sci., vol. 181, no. 16, pp. 3323–3335,
2011.
[24] A. R. Jordehi and J. Jasni, “Parameter selection in particle swarm
optimisation: A survey,” J. Experim. Theoretical Artif. Intell., vol. 25,
no. 4, pp. 527–542, 2013.
[25] Product Information of A Commercial Dual-Range Force Sensor, Jianbin Zhang is currently a Full Professor of Mechanical Engineering
Vernier Software & Technology, Aug. 2014. [Online]. Available: with Beihang University, Beijing, China. His research interest is focused on
http://www.vernier.com/products/sensors/force-sensors/dfs-bta/ engineering manufacturing and robotics.