∗

Dynamically tuned design of the MFI thorax

S. Avadhanula, R. J. Wood, D. Campolo, R. S. Fearing

{srinath, rjwood, minmo, ronf}@eecs.berkeley.edu
Department of EECS, University of California, Berkeley, CA 94720

Abstract approach in which the wing motion was separated into

rotations about uncoupled axes.
This paper presents an analysis of the major me- This paper describes a detailed dynamic model of
chanical component (the thorax) of the micromechan- the thorax, a design methodology for the optimal de-
ical flying insect (MFI), a centimeter sized aerial ve- sign of the thorax and the sensors which are currently
hicle currently in development at UC Berkeley. We utilized on the MFI.
present a description of the kinematics of the mecha-
nism which converts piezoelectric actuation into com- 2 Mechanism Description
plex 3D wing motion. A complete non-linear modeling
of the system based on the Lagrangian energy tech- The MFI utilizes piezoelectric actuators which de-
nique is presented. A design methodology is presented flect under the action of an applied electric field [12].
in order to achieve optimal matching conditions. Two The overall transmission from the actuators to the
kinds of sensors which are presently utilized on the wing can be broken up into 2 distinct parts. The first
MFI are described. Experimental results are presented stage consists of a planar fourbar which converts small
which validate some of the modeled non-linear aspects translational input from the piezo into large angular
of the mechanism. motion at the output. (typically about 60◦ for 3 mm).
The next stage of the mechanism consists of the
wing differential mechanism which converts 2 inde-
1 Introduction pendent angular inputs into a coupled flapping and
rotation of the wing.
Micro-aerial flying mechanisms are an attractive
form of locomotion because of their many potential Flapping
applications such as search and rescue, exploration, Axis
sensor distribution, reconnaissance, etc. made possi- θ2
ble by their very small size and high maneuverabil-
ity. Early work on microrobotic flight was done by 4
Shimoyama et al [11], while different aspects and ap-
proaches to micro areal flight have been pursued by 3
various groups ([3]-[8]). Fundamental work by Dick-
φ ψ
Ground

inson et al [4] showed that the complex high speed

motion of the insect wings is responsible for creating 2
unsteady aerodynamic effects which account for the Wing
exceptional lift and agility exhibited by insects. A 1
biomimetic approach was subsequently undertaken by γ
Fearing et al [5] to design a wing transmission which 3
γ1
had the same degrees of freedom as the actual insect. Spherical
γ
Early designs for the mechanism involved a fan-fold 2 joint
wing technique. Recently, the design was modified θ1
to enable the use of rigid wings to achieve the same
motion [14]. This technique avoids some of the aero- Figure 1: Kinematic representation of the wing differ-
dynamic and fabrication problems involved with the ential mechanism.
original design. Initial attempts at modeling the dy-
The wing differential is essentially a spatial five-bar
namics of the differential [14] assumed a lumped model
mechanism (with topology RSRRR) with 2 degrees of
∗ This work was funded by ONR MURI N00014-98-1-0671, freedom. It can be considered a variation of the spher-
DARPA and NSF KDI ECS 9873474. ical fourbar mechanism described in [9], where one of

1
the intersecting revolute joints is replaced by a spher- 80
ical joint. The various links are labeled in Fig. 1. The
60
2.5

Wing Rotation (degrees)

fourbars drive the links 1 and 4 directly. The wing 30
1 40

Spherical Joint
angles (deg)
itself forms link 2 of the mechanism (in practice, link γ1
0 λ 0
γ2
2 is made of a steel beam with the wing glued onto
it.) When the fourbars move in phase, then the whole −30
−40
γ3
differential simply rotates about the global flapping −80
−30 −15 0 15 30
−60
axis. This is defined as pure flapping. When the 2 −60 −30 0 30 60
Phase diff (degrees)
Phase diff (degrees)
fourbars move out of phase, i.e θ1 6= θ2 , then the wing
(a) Forward Kinematics (b) Spherical Joint
also rotates about link 2 through an angle ψ defined
as rotation in addition to some flapping. This is de- Figure 2: Kinematics of the Wing Differential.
scribed precisely in the next section.
Notation
rotations about 3 axes. In order to find out the equiv-
θ1 , θ2 Fourbar actuations alent stiffness of the wing differential we will need to
α ( := θ1 − θ2 ) phase difference between the 2 find out the angles through which each of these flex-
spars ures rotate for a given angle of attack. Fig. 2(b) shows
φ Deviation of wing from stroke plane the variation of the joints in the spherical joint with
ψ Rotation angle of the wing phase difference for λ = 2. It shows that γ3 joint con-
tributes quite a lot more to the overall stiffness of the
3 Kinematics spherical joint than the γ1 and γ2 joints.

In this section, we derive the forward and inverse

4 Dynamics
kinematics of the wing differential mechanism. It can
be shown that the wing rotation ψ, and the stroke
We use the Euler-Lagrange formalization to derive
plane deviation φ, depend on the phase difference α
the dynamics of the system. In order to do this, the
as
various energy terms in the mechanism have to be cal-
ψ = sin−1 (λ sin α) (1) culated in terms of the state variables.
!
1
φ = sin−1 p KE = KEactuators + KEfour-bars
λ2 + cos2 ψ
  +KEspars + KEwing (4)
−1 cos ψ
− tan , (2) PE = P Eactuators + P Edifferential (5)
λ
where The kinematics described in section 3 enable the
l formulation of the various terms above in terms of
λ := (3) θ2 , α, θ˙2 and α̇. Therefore defining {θ2 , α}T , as the
d
In the following discussion, the ratio l/d, denoted by state variables and transforming variables, we get the
λ, is referred to as the differential transmission ratio. following differential equation of motion (details of the
It is evident from eqn 1 that a larger transmission ra- derivation can be found in [1]):
tio means that for a constant phase difference α, we      
obtain a larger rotation ψ. We are interested in trans- θ¨1 θ˙1 θ1
M +b + K +
mission ratios with values ≥ 1, due to factors such θ¨2 θ˙2 θ2
 1 0 2
  
as dynamic balancing (which is explained later) and 2 mw,2 (α̇) τ1
= (6)
also getting the desired kinematic trajectories from − 21 m0w,2 (α̇)2 τ2
the wing. Fig.2(a) shows the variation of ψ with α for
λ between 1 and 2.5. We see that the phase differ- where
ence required to generate the same rotation decreases
with increasing λ. Furthermore, the amount of non- • Mass M =
linearity in the transmission also increases with λ.    
ms,1 + mw,2
mw,12 − mw,2
3.1 Inverse kinematics of the Spherical

 +mf b,1 + mact,1 

 
 ms,2 + mf b,2 
Joint 
 +mact,2 + mw,2

 mw,12 − mw,2  
As shown in Fig. 1, the spherical joint consists of +mw,1 − 2mw,12
3 flexures in parallel in order to achieve independent (7)

2
 The main contribution to the inertia matrix and time-varying [4]. However, for the sake of
comes from the wing and appears as mw,1 , mw,2 analysis, we use a linear damping matrix to ap-
and mw,12 which are the inertia components of proximate them. It should be noted that the lin-
the wing lamina reflected through the differen- ear damping term is calculated so that it overes-
tial transmission. These depend non-linearly on timates the drag force felt on the wing at every
both the differential transmission ratio, λ and the instant except when it achieves the peak velocity
phase difference, α. [14].
   2  
mw,1 cφ 1 − c2ψ · c2φ b2 b12 − b2
0
2 0 0 b = (11)
 mw,2  =  φ φ 2 c2ψ + ψ 2 b12 − b2 b2 + b1 − 2b12
0 0
mw,12 0 φ sψ cφ − ψ sφ
  The most dominant term of the damping ma-
−s2φ cψ Jxx trix is b1 and determines the Q of the wing in
0 0
−2φ ψ sψ   Jzz  , (8) a simple flapping mode. Preliminary measure-
0 0
ψ cφ cψ + φ sφ sψ Jxz ments of the structural b yielded an estimate of
7 × 10−9 N ms/rad.
where the dependence of φ and ψ on λ and α
are given in equations 1-2 and Jxx , Jzz , Jxz are • Stiffness The stiffness matrix of the thorax
the fixed inertia components of the wing. The mechanism is given by
equation above simplifies considerably about the  
nominal position of the differential i.e for α = 0. ks,1 + kd −kd
K = (12)
For α = 0, we have: −kd ks,2 + kd
    
mw,1 1 0 0 Jxx where, kd , called the differential stiffness is de-
 mw,2  =  0 λ2 0   Jzz  (9) fined as
mw,12 0 0 λ Jxz 1 ∂P Edifferential
kd := (13)
The terms ms,1/2 in eqn. 7 are the inertias of links α ∂α
1 and 4 of the differential. Currently, these spars The differential stiffness arises from the flexures
are made out of 12.5 micron thick stainless steel which make up the joints of the mechanism.
folded into square and triangular beams 1 mm Each flexure can be considered a simple rotational
on each side using the methods described in [10]. spring whose stiffness depends on the dimensions
For the current design, the length of the spars (i.e of the flexure such as length, thickness and width
links 1 and 3 in Fig. 1) is fixed at 2 mm and the and also the material of the flexure (see [7] for a
differential transmission ratio is varied by chang- complete treatment of flexures). We presently use
ing the spacing between them. Consequently, the 6.3 micron thick polyester flexures (E = 2.5GP a).
spar inertias are fixed at 2 × 10−12 and 3 × 10−12 Changing the flexure lengths provides the abil-
kg · m2 respectively. ity to change the differential stiffness for a given
Also, the terms mf b,1/2 in eqn. 7 are the inertias transmission ratio. Fig. 3 shows the variation of
of the two fourbars which transmit motion from kd with λ and α for a typical wing differential
the PZTs to the wing differential. Their design is where the major flexures are made of 6.3 micron
also fixed (details in [1]) and consequently, their thick polyester flexures 175 microns long and 1
inertias are fixed at approximately 1.7 × 10−12kg · mm wide. We see that kd increases with the λ
m2 each.
8

Finally the terms mact,i in eqn. 7 is the equivalent 7

Differential Stiffness (x 1e−6 Nm/rad)

rotational inertia of the PZT reflected about the 6

2.5

wing hinge. Assuming a linear fourbar transmis- 5

2 λ
sion, we have
4
1.5
1
mact,i = mP ZT,i /Nf2 , i = 1, 2 (10) 3

where Nf is the fourbar transmission and 1

mP ZT,1/2 is the linear inertia of the PZT mea- 0

−50 −40 −30 −20 −10 0 10 20 30 40 50
sured at the point of actuation. Phase difference (degrees)

• Damping The aerodynamic forces felt by an in- Figure 3: Variation of differential stiffness with λ and
sect wing during its motion are quite non-linear α.

3
 as is to be expected. Also, the non-linearity in resonance at the same frequency. This condition
kd increases sharply with λ. (A linear differen- implies the following condition on the system pa-
tial would have had a constant stiffness for all α.) rameters:
ks,1 and ks,2 in equation 12 represent the actuator
ks,1 + kd ks,2 + kd
stiffness reflected about the wing hinge. = (15)
ms,1 + m2 ms,2 + m2 + m1 − 2m12
kact,i
ks,i = , i = 1, 2 (14) Furthermore, the amplitude of motion at reso-
Nf2
nance should be about the same for both spars,
where Nf is the fourbar transmission ratio and i.e A11 = A22 , which gives the condition that:
kact,i is the linear stiffness of the actuator mea-
sured at the point of actuation. b2 = b2 + b1 − 2b12 (16)

5 Design Synthesis 2. Small cross-coupling It’s desirable to be able to

drive the two spars independently. This will en-
Once the differential equations are set up, we want sure that we are able to generate sufficient phase
to design the differential mechanism to get good be- differences and correspondingly sufficient angles
havior from the mechanism. As a first step, we set of attack. For small cross-coupling at DC, the
up a metric for evaluating the performance of a differ- off-diagonal terms in (12) should be small which
ential equation. We then try to design a mechanism implies:
which will satisfy this differential equation.
kd  ks,1 , ks,2 (17)
5.1 Performance Metric
For small cross-coupling at AC, the off-diagonal
Consider the linearized frequency response of (6) for terms in the mass and damping matrices should
some typical values of kd , Jxx , Jxz , etc shown in Fig. 4. be small, which implies:
A11 represents the amplitude of motion of spar 1 (link
1 in Fig. 1) when only the first actuator is driven with m12 − m2 ≈ 0 (18)
the resonance at f1 . Similarly A22 represents the am- b12 − b2 ≈ 0 (19)
plitude of motion of spar 2 (link 4 in Fig. 1) with
actuator 2 driven. A12 represents the cross-coupling. These matching conditions, (15-19) can be expanded
It represents the motion of spar 2 when actuator 1 is to:
driven. A large A12 means that the actuators begin
to interfere with each other, making it harder to drive kd  ks,1
mw,1 = 2mw,2
them out of phase. The frequency response, while not ks,1 = ks,2
b1 = 2b12 (20)
ms,1 = ms,2
b2 = b12
300
mw,1 = 2mw,12
250

The design effort now involves choosing the various

200
parameters of the mechanism in order to approach the
Magnitude
Magnitude

150 f1 A22 A11 matching conditions given above. The parameters in

100
the construction of the wing mechanism are:
f2
50 A12 1. Fourbar Transmission Ratio, Nf : The four-
0 bar transmission ratio effects the reflected stiff-
0 50 100 150 200 250 300 350

Frequency (Hz) ness of the actuator (10) and also the DC response
of the mechanism. The resonant frequency de-
Figure 4: Performance of a typical 2 input 2 output creases linearly with the transmission ratio while
2nd order system. the DC motion increases linearly with the trans-
mission ratio.
what we desire serves to illustrate some of the charac-
teristics we design for: 2. Differential Transmission Ratio, λ: λ af-
fects how the constant wing inertia terms
1. Symmetry The transfer functions from τ1 to θ1 [Jxx , Jxz , Jzz ]0 are mapped to the wing hinge (8).
should have the same resonant frequency as the The differential transmission ratio also effects the
transfer function from τ2 to θ2 , i.e f1 = f2 . This reflected stiffness of the differential mechanism.
will ensure that we can drive the two spars in (Fig. 3).

4
 3. Flexure Design: The flexures in the wing dif- Fig. 5(b) shows the wing used in the current MFI.
ferential have to be designed for a low parallel This wing consists of a face sheet made of 7 micron
stiffness while avoiding resonant modes due to se- thick polyimide with 200 micron diameter × 10 micron
rial compliance in undesirable directions, which thick polyimide tubes laid across it. (See [1] for fabri-
might occur if the flexures are made too long. cation details). The primary purpose of the tubes is to
provide the necessary stiffness to the wing while also
4. Wing Shape: The wing shape is not a single enhancing the inertial properties of the differential.
parameter. It includes information about how the Table 1 is a brief comparison of the current wing with
mass is distributed over the wing area. In our case a calliphora wing (since the calliphora weighs nearly
the shape can be completely specified by the three as much as the target MFI.) The main advantages of
inertia parameters, Jxx , Jzz and Jxz . this wing were the ease of fabrication and more im-
portantly the highly repeatable and accurate manner
5.2 Fabrication in which the required low inertia could be obtained.

MFI wing Calliphora wing

Fig. 5(a) shows the wing differential fabricated to
Weight 0.48 mg 1.0 mg
final scale. The top spar can be seen to be slightly
Moment of Inertia 21 mg · mm2 22 mg · mm2
leading the bottom spar causing a positive angle of
Moment of Area 1100 mm4 900 mm4
attack. The fabrication, while retaining the kinemat-
(Flapping)
ics previously defined, is significantly different than
the schematic (Fig. 1) due to considerations of flexure Table 1: Comparison of the MFI wing with a Cal-
width optimization, and ease of assembly and align- liphora wing.
ment.
The main difficulties were encountered in the fab-
rication process were firstly, that the flexures cur-
rently employed in the MFI are made of 6.3 micron 6 Experimental Results
polyester. It was found that the lowest differential
stiffness achieved with this material is still more than Linearizing (6) about an operating point yields a
50% of the actuator stiffness. This is not quite low transfer function matrix which relates the two input
enough to get the desired uncoupling. Secondly, the torques τ1 and τ2 to the two spar angles θ1 and θ2 .
PZT actuator used on the structure turned out to
    
have a stiffness of 113 N/m. This was about 50% less θ1 G11 (s) G12 (s) τ1
than the anticipated actuator stiffness. In future ver- = (21)
θ2 G21 (s) G22 (s) τ2
sions, a lower fourbar transmission ratio will be used | {z }
to counter this effect. The lower stiffness resulted in =: G(s)

a decreased resonance and increased coupling at reso-

nance. The effect of non-linearity in the model is to change
the transfer function above for different operating
These factors lead to a differential which exhibited
points.
more coupling than is optimal. From the point of view
of validating the model, however, the present thorax is The experimental results presented here involved
better since a differential with more coupling exhibits measurement of the frequency response of the sys-
rather more “interesting” behavior than a totally un- tem for small amplitudes about two different operating
coupled differential and is much harder to predict. points.
Section 6 describes some of the experiments which
were undertaken to test the validity of the model. 1. α = θ1 − θ2 = 0◦ . The differential stiffness is at
a minimum when the two spars are in phase and
we expect the most uncoupling in this situation.
Fig. 6(a) shows the predicted frequency response
for this operating point.

2. α = θ1 − θ2 = 25◦ . At this phase difference, the

differential stiffness is almost 3 times its nominal
value. Fig. 6(b) shows the predicted frequency re-
(a) Wing Differential (b) Polyimide Wing sponse for this operating point. The major differ-
ence between the two is that the second resonance
Figure 5: Photo of MFI wing differential mechanism. moves from 150 Hz to more than 220 Hz.

5
 250
τ θ1
The signal generated by the strain gages, V0 and
200
{ τ12 θ2 the tip displacement of the actuator, δ is given by:

Magnitude
150
τ θ2
100
{τ21 θ1 δ = C1 Vin + C2 Fr
50
(22)
0
0 50 100 150 200 250 300 350
V0 = C3 Vin + C4 Fr ,
100
τ θ1 where the parameters C1 , C2 , C3 and C4 are 2 × 2 ma-
0 { τ12 θ2 trices which contain the material and physical proper-
τ
Phase

−100
θ2
−200 { τ21 θ1
ties of the actuator and strain sensor. We can readily
−300 measure the frequency response of the strain gage dy-
−400
0 50 100 150 200 250 300 350
namics (V0 /Vin (s)) using a dynamic signal analyzer
Frequency (Hz)
(DSA), HP3562A. We are primarily interested in re-
(a) α = 0◦ lating the thorax dynamics G(s) to the above measure-
250
τ θ1
ment. It can be shown that the measurement dynam-
200
{ τ12 θ2 ics and the structural dynamics are related as (details
Magnitude

150
τ θ2
100
{τ21 θ1
in [13]):
50
0 V0
0 50 100 150 200 250 300 350 (s) = C3 + C4 (G(s) − mact s2 ) ×
100
Vin
τ θ1
{ τ12

−1
0
θ2 I2 − C2 (G(s) − mact s2 ) C1 (23)
Phase

−100
τ θ2
−200 { τ21 θ1 In practice, a state space approach is used to relate
−300
−400
the measurement and thorax dynamics. This avoids
0 50 100 150 200 250 300 350
Frequency (Hz)
numerically unsound calculations such as inversion of
(b) α = 25◦ transfer function matrices. Figs. 8(a) and 8(b) show
the comparison of the actual and predicted strain gage
Figure 6: Predicted frequency response of the wing signals. Fig. 8(a) corresponds to the case when the
differential. sensing actuator is the same as the drive actuator.
Fig. 8(b) represents the cross-coupling, i.e when we
sense the motion of the non-driven actuator. Both
6.1 Strain Gage Sensing / Results these measurements were taken for α = ψ = 0◦ and
show the double peak characteristic of the coupling
The deflection of the PZT actuators depends on between the two sides of the structure.
both the applied electric field and the reaction force on
the actuator from the fourbar [16],[17]. This fact can 6.2 Electrical Sensing
be used to fashion a state estimator which measures
the torques acting on the spars (τ1 , τ2 ) as well as the Piezo-electric actuators link their mechanical and
wing angles (θ1 and θ2 ). The state estimator uses the electrical domains in a bi-directional fashion[6]. This
applied electric field applied to the actuators as the property enables the use of the piezo-electric actuators
input and the moments at the base of the actuator as as sensors. Fig. 9 shows a representation of the actua-
outputs. These moments are measured using 1 DOF tor as a two-port element, where C0 and R0 represent
semiconductor straingages [2],[15]. the electrical capacitance and resistance (losses) of the
actuator (and can be measured before-hand) and Cm
represents the mechanical compliance of the actuator
(Cm = 1/kact ). Tp represents the transmission ratio
δ
of the actuator from the electrical to the mechanical
M0 domain.
The total admittance of the actuator as seen from
F0
the electrical domain can be calculated as

θ Y = Y0 (ω) + Y1 (ω)
V0 = Y0 (ω) + Tp2 Ym (ω)
Vin jωC0
= + Tp2 Ym (ω) (24)
1 + jωR0 C0
We are primarily interested in extracting the mechan-
Figure 7: Strain gage sensing. ical admittance of the structure Ym (ω). To do this, we

6
 0.025 Piezo actuator
0.02
Measured 1:Tp
Magnitude

Mechanical
0.015 Predicted
C0 Cm
0.01
Y Y1 Ym

Load
0.005
R0
0
0 20 40 60 80 100 120 140 160 180 200
Y0
0

Sinusoidal
−50
Vo
Phase

Predicted

Source
−100
Measured Vin
−150 Sensing
Cs
−200
Capacitor
0 20 40 60 80 100 120 140 160 180 200
Frequency (Hz)

(a) Vi,1 → V0,1

0.02

0.015
Measured Figure 9: Using the piezo-electric actuator as a
Magnitude

Predicted mechanical admittance sensor.

0.01

0.005

0
gle resonance (before 200 Hz) which is characteristic
0 20 40 60 80 100 120 140 160 180 200
of high coupling.
100
0
7 Conclusions
Phase

−100 Predicted
−200
Measured A full non-linear modeling of dynamics of the tho-
−300
−400
rax, the major mechanical component of the MFI is
0 20 40 60 80 100 120 140 160 180 200
Frequency (Hz) presented. A strategy for choosing the various fab-
(b) Vi,2 → V0,1 rication parameters is presented in order to optimize
the expected performance of the system. Two kinds of
Figure 8: Comparison of predicted (dashed) and ac- sensors which are presently used on the MFI are dis-
tual (solid) strain gage dynamics. (a) α = 0◦ (b)α = cussed. The non-linear aspects of the system are vali-
25◦ . dated by observing the frequency response of the sys-
tem and its variation with changing operating points.
Inspite of several constraints in the present fabrica-
proceed by measuring Y (ω), which can be calculated tion setup which were mentioned earlier, we observed
as: satisfactory wing trajectories from the thorax. In par-

V0
 ticular, we achieved inertial matching so that both ac-
Vin  tuators had the same loaded resonant frequencies, en-
Y (ω) = jωCs  (25)
1− VV0 abling them to be operated simultaneously with equal
in effectiveness.
V0 /Vin is readily measured using a DSA. Subtracting We are presently investigating other materials for
the effect of C0 , R0 using eqn. 24 yields the required making flexures such as HS-2 (silicone-rubber), which
mechanical admittance of the structure Ym (ω). has a low elastic modulus and has potential for mak-
Figs. 10(a) and 10(b) show a comparison of the ing very low parallel stiffness flexures. Problems such
measured mechanical admittance Ym (ω) and with the as clamping, adhesion will have to be overcome for
frequency response predicted by the model for the successful implementation. Future work will also in-
cases where α = 0◦ and α = 25◦ . A single scaling volve system identification on the structure to fine-
factor was used to match the maximum amplitudes tune the values of various model parameters using the
of the measurement and the simulation. This scaling techniques presented in section 6. Control strategies
takes into account the transmission ratios such as Tp , will be employed to assess the extent of uncoupling
the electromechanical transmission ratio of the PZT required to generate the required wing trajectories.
and Nf , the amplification ratio of the fourbar. We see
a validation of an important effect of the system non- Acknowledgments
linearity, where a change in the set-point changes the
frequency response of the system radically. Fig. 10(a) The authors would like to thank Joseph Yan for
shows a double resonance while Fig. 10(b) shows a sin- valuable discussions on structural kinematics, Gabe

7
 Normalized Magnitude
Measurement a micromechanical flying insect,” IEEE Int. Conf.
1
Simulation on Robotics and Automation, pages 1509–1516, San
0.8
0.6
Francisco, CA, April 2000.
0.4 [6] M. Goldfarb and N. Celanovic, “Modelling Piezolec-
0.2
tric Stack Actuator for Control of Micromanipula-
0
0 20 40 60 80 100 120 140 160 180 200 tion,” IEEE Control Sys. Magazine, vol. 17, no. 3,
pp. 69–79, 1997.
50
Measurement [7] M. Goldfarb, J.E. Speich, “A Well-behaved revolute
Phase (degrees)

0
Simulation
−50 flexure joint for compliant mechanism design,” Trans-
−100
actions of the ASME, Vol 121, pages 424-429, Septem-
−150
ber 1999.
−200 [8] T.N. Pornsin-Sirirak, S.W. Lee, H. Nassef, J. Gras-
0 20 40 60 80 100 120 140 160 180 200
Frequency (Hz) meyer, Y.C. Tai, C.M. Ho, and M. Keennon, “MEMS
(a) α = 0◦ wing technology for a battery-powered ornithopter,”
In IEEE 13th Annual Int. Conf. on MEMS, pages
Normalized magnitude

1 Measurement 799–804, Piscataway, NJ, Jan 2000.

0.8 Simulation [9] D. A. Ruth and J. M. McCarthy, ”The Design of
0.6
Spherical 4R Linkages for Four Specified Orienta-
0.4
tions,” Computational Methods in Mechanisms, ed. J.
0.2
Angeles, Spring-Verlag, 1998.
0
0 20 40 60 80 100 120 140 160 180 200
[10] E. Shimada, J.A. Thompson, J. Yan, R.J. Wood, and
100 R.S. Fearing, “Prototyping millirobots using dextrous
Measurement microassembly and folding,” Symp. on Microrobotics
Phase (deg)

0
Simulation ASME Intl. Mechanical Engineering Cong. and Expo.,
−100 Orlando, FL, Nov 2000.
−200 [11] I. Shimoyama, H. Miura, K. Suzuki, and Y. Ezura,
−300 “Insect-like microrobots with external skeletons,”
0 20 40 60 80 100 120 140 160 180 200
IEEE Control Sys. Magazine, 13:37–41, February
(b) α = 25◦ 1993.
[12] M. Sitti, D. Campolo, J. Yan, R.S. Fearing, T. Su,
Figure 10: Comparison of experimental and simulated
D. Taylor, and D. Sands “Development of PZT and
frequency responses using piezo-electrical sensing. PZN-PT based unimorph actuators for micromechan-
ical flapping mechanisms,” Int. Conf. on Robotics and
Automation, pages 1509–1516, Seoul, South Korea,
Moy for numerous suggestions on design and fabrica-
21-26 May 2001.
tion issues and Metin Sitti and Eric Park for helping
with the fabrication and modeling of the PZT actua- [13] R. Wood, R.S. Fearing, “Flight Force Measurements
tors. for a Micromechanical Flying Insect,” to appear Int.
Conf. Intelligent Robotics and Systems 2001, Maui,
HI, Oct. 29-Nov. 3, 2001
References
[14] J. Yan, R.J. Wood, S. Avadhanula, M. Sitti, and
[1] S. Avadhanula, “The design and fabrication of a dy- R.S. Fearing, “Towards flapping wing control for a
namically tuned MFI thorax,” MS Thesis, University micromechanical flying insect,” IEEE Int. Conf. on
of California, Dec 2001. Robotics and Automation, pages 3901—3908, Seoul,
South Korea, May 2001.
[2] A. Bicchi, A. Caiti, and D. Prattichizzo, “Optimal
design of a multi-axis force/torque sensor,” 38th Conf. [15] K. Abe, T. Miwa, and M. Uchiyama, “Development
on Decision and Control, pages 2981–2986, Phoenix, of a 3-axis planar force/torque sensor for very small
USA December 1999. force/torque measurement,” Trans Jpn Soc Mech
Eng, Vol 42, No. 2, pages 376-382, 1999.
[3] A. Cox, E. Garcia, and M. Goldfarb, “Actuator devel-
opment for a flapping microrobotic MAV,” In SPIE [16] J. Smits and W. Choi, “The constituent equations of
Microrobotics Symp., pages 102–108, Boston, MA, piezoelectric heterogeneous bimorphs,” IEEE Tran.
Nov 1998. On Ultrasonics, Ferroelectrics, and Freq. Control, vol.
38, pages 256-270, May 1991.
[4] M.H. Dickinson, F-O. Lehmann, and S.P. Sane,
“Wing rotation and the aerodynamic basis of insect [17] M. Weinberg, “Working equations for piezoelectric
flight,” Science, 284:1954–1960, June 1999. actuators and sensors,” J. of Microelectromechanical
Systems, vol. 8, pages 529-533, Dec. 1999.
[5] R.S. Fearing, K.H. Chiang, M.H. Dickinson, D.L.
Pick, M. Sitti, and J. Yan, “Wing transmission for