Professional Documents
Culture Documents
1.Introduction
The development of flappingwing MAVs has been
spearheaded by the demand of developing autonomous,
lightweight, smallscale flying machines that are
appropriate for a variety of missions, including
reconnaissance over land, in buildings and tunnels, and
other confined spaces. Of particular interest is the ability
ofthesevehiclestooperateintheurbanenvironmentand
perch on buildings to provide situational awareness to
militarycombatants.
Figure1.HummingbirdsandFlappingwingMAVs
Neural networkbased (NNbased) control has been
widelyappliedinvarioussystemssee,forinstance,[10
18];however,mostNNbasedcontrolalgorithmsconfront
oneorsomeofthefollowingissues:1)Alargeamountof
trainingdataisneededtopreconstructthenetwork,and
although many improved NNs are able to update
themselves online, their accuracy heavily relies upon the
selected training data. 2) There is no existing theory to
regulate how to build a NN, such as how many layers
there should be or what kind of base function should be
used. As a matter of fact, constructing a NN is always
time consuming and empirical. 3) A practical NN
containslargequantitiesofneurons,whichbringswithit
a heavy computational burden and requires a great deal
ofmemoryspace.4)Usually,networkreconstructionisa
necessity when the system dynamics change, even very
slightly. 5) There is no theoretical proof to guarantee the
stabilityofthecontrolsystem.
Morespecifically,NIMNisveryeasytoconstructitdoes
notrelyuponaprecisesystemmodelanditdemandsless
computationEnergycomparedwithmostothermethods.
It is able to learn from both past experience and current
observed information to improve its performance.
Moreover,thereisnoneedfornetworkreconstructionor
consistent weights to update even if the systems
dynamics change significantly. These features have been
verified via both theoretical analysis and simulation
study.
2.NeuralMemoryNetwork
A.Fundamentals
Human memory is one of the most intriguing biological
phenomena in nature [25] [26]; it is what makes possible
so many of our complex cognitive functions, including
communicatingandlearning.
Figure2.Humanmemory/learningsystem
B.AnalogiesbetweenaHumanMemorySystemandaNeural
MemoryNetwork
The proposed NeuralMemory Network (NMN)
simulatesthemechanismsandproceduresoftheHuman
Memory System (HMS) described above. The analogies
betweenHMSandNMNarepresentedinTable1.
HMS NMN
Pastexperience Pastcontrolsignal
Current(final)response Currentcontroloutput
Observations Availablesysteminformation
Feedbacks(comparison) Controlerrors
Objective Desiredsystemresponse
Table1.AnalogiesbetweenHMSandNMN
C.NeuralMemoryNetworkStructure
Considering an arbitrary system with the following
systemdynamics:
x f gu f (1)
Wingbeat pattern
control
Hummingbird
Flapping wing MAV
Flight
Control
Modeling
Inspiration Inspiration
System
dynamics
Wingbeat pattern
control
Hummingbird
Flapping wing MAV
Flight
Control
Modeling
Inspiration Inspiration
System
dynamics
Other
Observation
Action Action
Final Control Action
Memorizing
Objective
Feedback & Reasoning
2 Int. j. adv. robot. syst., 2013, Vol. 10, 305:2013 www.intechopen.com
where f(.) and g(.) are two nonlinear functions, f
representsuncertaintiesinthissystemanduisthecontrol
signal. Defining a symbol e = x x* to denote the control
error, where x* represents the desired state, a 1
st
order
NeuralMemoryNetworkisdepictedasinFig.3.
Figure3.1
st
orderNeuralMemoryNetwork
This 1
st
order Network includes four types of neurons.
These are past experience neurons, objective neurons,
feedback neurons and observation neurons, whereby w1,
w2, w3 and w4 represent their respective weights and k is
the step counter. Note that observation neurons are not
necessarilyrequired.
An m
th
order NeuralMemory Network is illustrated in
Fig.4,inwhichmdenotesmstepsbackwards.
Figure4.m
th
orderNeuralMemoryNetwork
D.StabilityAnalysis
In this section, the theoretical analysis for the systems
stability is given. However, and for simplicity, only the
proofforthe1
st
ordernetworksstabilityisdetailedstep
bystep.
AccordingtothenetworkstructuredepictedinFig.3,we
have:
1 * 1 1 1
k 1 2 3 4
u w U w g X w g t E w g F
(2)
where
T
k 1
U [u ]
,
k k 1
* * T
X [x x ]
,
T
k k 1
E [e e ]
,
T
k k 1
F [f f ]
andtisthesamplingperiod.
Theweightsforthe1
st
orderarechosenas:
w1=1(3)
w2=[11](4)
w3=[21](5)
w4=[11](6)
Please refer to our previous work for the details of the
weight settings [23]. Therefore, using the Euler
approximation,weget:
*
k 1 k k k k k
e e t(f gu f x )
(7)
*
k k 1 k 1 k 1 k 1 k 1
e e t(f gu f x )
(8)
With(7)minusing(8),ukbeingdefinedin(2),andweights
givenin(3)to(6).Itcanbereadilyshownthat:
k 1 k k 1
e t( f f )
(9)
Therefore:
2
k 1 0
e t c
(10)
where:
0
d f
c max
dt
,(11)
donates the maximum variation rate of the disturbances
and uncertainties f, which is assumed to be bounded
due to the fact that in general such variation cannot be
infinitely fast, otherwise no control strategy is able to
work.
Asaresult,andbecausethesamplingintervaltisavery
small number, any tracking error is confined within a
narrowenvelopdefinedbyt
2
c0.Theobservationneurons,
F, are not necessarily required for the same reason,
namely that it is not possible that f be able to change
infinitelyquickly.
Asimilaranalysiscanbemadeforthehigherordercase.
Presumably, a higherorder NeuralMemory Network
leads to better control precision because more of the
earlier (longerterm) memory is incorporated within the
controlscheme,thoughmorecomputationsareinvolved.
3 Liguo Weng, Min Xia, Kai Hu and An Wang: Micro Aerial Vehicle (MAV) Flapping
Motion Control Using an Immune Network with Different Immune Factors
www.intechopen.com
Refer to Table 2 for the weights of higherorder Neural
MemoryNetworks.
w1 w2 w3 err
[1]
T
[1,1]
T
[2,1]
T
t
2
c0
[2,1]
T
[1,2,1]
T
[3,3,1]
T
t
3
c0
[3,3,1]
T
[1,3,3,1]
T
[4,6,4,1]
T
t
4
c0
[4,6,4,1]
T
[1,4,6,4,1]
T
[5,10,10,5,1]
T
t
5
c0
Table2.Higherorderweights
4.ApplicationtoMicroAerialVehicles
A.SystemDynamics
Based upon a study of the skeletal structure of
hummingbirds and the existing literature [25], [26], an
artificial vibratory flapping system as shown in Fig. 5
isdevelopedtodescribethewingflappingmotion.
Tworeferenceframesconsistingofabodyfixedaxis{e1,e2,
e3}andawingfixedaxis{x,y,z},aredefined,bothhaving
their origin at the shoulder joint. An ideal column with a
length l and a radius r is used to represent the humerus.
Located at distance l1 from the shoulder joint, a pair of
exogenous forces is introduced to represent the depressor
muscle, whereby each of f1 and f2 composes an opposite
angle to z axis. At the same location, a vertical spring
with a stiffness k donating the elevator muscle is placed,
the justification for representing this muscle with a single
spring is that this muscle travels around and over the top
of the coracoid, reversing direction and attaching to the
sternum; therefore, this constrains the line of action of the
elevatormuscleinpassingthroughapointatthetopofthe
coracoid,justliketheforcegeneratedbythespring.
Figure5.Schematicrepresentationofthevibratoryflapping
system
Theflappingmotiondynamicsisthusgivenas:
T T 2 T
1 2 1 b
J[p q] G[F F ] Nkl vC[p q] = + + A (12)
T T
[ ] Y[p q] u =
(13)
where
1 1
1 1
l cos cos l cos cos
G
l sin cos l sin cos
o o
o u o u
(
=
(
,
T
N [ sin cos sin ] u =
and C = diag(c1 c2). v is the free stream speed; therefore,
G[F1 F2]
T
, Nkl1
2
and vC[p q]
T
correspondingly denote
moments due to actuating forces, restoring forces and
damping forces. b represents the model constructing
error and J is the moment of inertia. [p q]
T
represents the
angular velocity of the wingfixed frame with respect to
the bodyfixed frame, and and are Euler angles
respectively referred to as flapping and folding motions.
Yisthetransformationmatrix.
B.NeuralMemoryNetworkControlDesign
To derive the control input for the flapping motion, the
systemdynamicsisrewrittenasfollows:
b
f gu O = + + A
(14)
1 1 2 1
1
f YY YJ (Nkl vCY )
= O+ O
(15)
1
g YJ G
= (16)
T
1 2
u [F F ] = (17)
where=[]
T
,u=[F1F2]
T
.Moreover,notethatbecause
neither of the Euler angles nor could physically reach
/2,matrixGandYarealwaysinvertible.
Definingsymbol s e e | = + ,where>0,itiseasytoprove
that s converges to zero as e goes to zero. Using the 1
st
orderNeuralMemoryNetwork,itiseasilyshownthat:
2
k 1 0
s t c
+
s (18)
b
0
d
c max
dt
A
= (19)
For the same token, s is bounded and therefore e is
boundedwithinanarrowenvelop.
C.SimulationVerifications
To verify the effectiveness of the NeuralMemory
Network,numericalsimulationsonflappingwingmotion
control are conducted using the firstorder network. The
parametersusedforsimulationarechosenas:t=0.01s,k=
10N/m, v = 10m/s, = 40
o
, C = diag(0.1, 0.06) and the
uncertaintyb=[10sin(t)+515cos(4t)8]
T
,whichistime
varyingconstantly.
Twotypesofmotionaresimulated.
[1] Highspeedcruising
Figure6.Aforwardinghummingbirdswingtipmotion
Figure7.Simulatedwingtipmovementtrajectories
[2] Reverse.
Figure8.Controloutputs
Figure9.AHummingbirdflyingbackwards
Figure10.Wingtiptrajectoriesforareversingmotion
where rand is a function generating a random number
from 0 to 1. In addition, a more complicated motion
hovering is simulated. The corresponding simulation
results are shown in Figs. 12 and 13. Note that the
proposed NeuralMemory Network still works well,
whereas most existing control strategies would fail in
suchacomplicatedenvironment.
Figure11.Controloutputsforreversingmotion
High speed forwarding High speed forwarding
0
2
4
6
-5
0
5
-6
-4
-2
0
2
4
6
Folding Movement
F
l
a
p
p
i
n
g
M
o
v
e
m
e
n
t
Tracking
Desired
Joint
humerus
0 1 2 3 4 5 6 7 8 9 10
-2
-1
0
1
2
3
x 10
4
F
1
0 1 2 3 4 5 6 7 8 9 10
-2
-1
0
1
2
x 10
4
t
F
2
-5
0
5
-2
-1
0
1
2
0
1
2
3
4
5
F
l
a
p
p
i
n
g
M
o
v
e
m
e
n
t
Folding Movement
Tracking
Desired
Joint
humerus
0 1 2 3 4 5 6 7 8 9 10
-1
-0.5
0
0.5
1
1.5
x 10
4
F
1
0 1 2 3 4 5 6 7 8 9 10
-2
-1
0
1
2
x 10
4
t
F
2
5 Liguo Weng, Min Xia, Kai Hu and An Wang: Micro Aerial Vehicle (MAV) Flapping
Motion Control Using an Immune Network with Different Immune Factors
www.intechopen.com
Figure12.Ahoveringhummingbird
Figure13.Wingtiptrajectoriesforahoveringmotion
From the simulation results above, it is clearly seen that
the tracking performances in either scenario (constant
free stream and random free stream) are fairly good.
However, the control output magnitudes are relatively
large at the beginning of every simulation and so this
phenomenon is understandable because little experience
has been gained at that point, and hence the Neural
Memory Network is relatively immature. In order to
overcome this problem, the Neural
Immunology/Memory Network has been developed,
whichwillbeaddressedinthenextsection.
5.Neuralimmunology/memorynetwork
A.ImmunologyInspiration
The human immune system is a dynamic and intelligent
system [2729]. Once one kind of antigen invades our
body, a general immune response is initiated by the
immune system. After the invading antigen gets
acquainted, a more specific immune response begins to
workagainsttheantigen.
Thehumanimmunesystemworksontwolevelswiththe
general goal of pathogen control: a general response
mechanism, called innate immunity that does not
directly respond to any specific pathogen, and a specific,
antibodymediated response mechanism called acquired
immunity.
At the beginning, and because of the lack of knowledge
about the antigen, the immune system selects a sort of
universalantibodytotakeeffect(innateimmunity).After
the antigen becomes better understood, a much more
matching and effective kind of antibody is activated to
fight(acquiredimmunity).
B.NeuralImmunology/MemoryNetworkStructure
The overall immune system is similar to a complicated
and intelligent network, which dynamically adjusts its
neuron(antibody)weightstoreacttotheantigens.
Figure14.NeuralImmunology/MemoryNetworkStructure
C.ImmuneFactors
Inthesimulation,innateresponsenetworkischosenasui=
ke,whichisatraditionalproportionalcontrolapproach.The
acquired response ua is the NeuralMemory Network.
However, the question of how to choose an appropriate
immune selection factor is an issue. In this paper, two
candidatesarepresentedanddiscussed.
1. HardSwitch
0
0
0 t t
1 t t
(22)
Thismeansthatatthebeginning,onlyuiworksand,after
a certain period, u2 replaces u1 completely and suddenly.
Fig.15illustratesahardswitchfunction.
0
2
4
6
-1
-0.5
0
0.5
1
-4
-2
0
2
4
Folding Movement
F
l
a
p
p
i
n
g
m
o
v
e
m
e
n
t
Desired
Actual
Joint
humerus
6 Int. j. adv. robot. syst., 2013, Vol. 10, 305:2013 www.intechopen.com
Figure15.Ahardswitch
For the purpose of clarity and concision, only a forward
cruising motion is tested in this section. The tracking
result is shown in Fig. 16 and the control outputs are
depictedinFig.17.
Figure16.Trackingtrajectoriesofthehardswitch
2. SelfadaptiveSwitch
t
1 exp , 0, 0
e
o c
c
| |
= > > |
|
+
\ .
(23)
Obviously, this immune selection factor incorporates
control error and varies as the magnitude of e changes.
When||e||islarge,whichmeansthattheantigenisnot
well understood, the function value approaches 0 and
therefore the innate response tends to function more;
whereas when ||e|| is small, which means that learned
experience is precise, the function value goes to 1 and
thus the acquired response begins to work. Fig. 18
illustrates the tracking trajectories, Fig. 19 shows the
networkoutputsandFig.20givesthecurveof.
Figure17.Networkoutputsunderthehardswitch
Figure18.Trackingtrajectoriesoftheselfadaptiveswitch
Figure19.Controloutputsoftheselfadaptiveswitch
6.ComparisonandAnalysis
Inthissection,theNIMNiscomparedwiththeNMNand
a traditional BPNN. All of these 3 approaches are
applied to flapping control and the results are shown in
table3.
0 2 4 6 8 10 12 14 16 18 20
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
t
o
0
2
4
6
-5
0
5
-6
-4
-2
0
2
4
6
Folding Movement
F
l
a
p
p
i
n
g
M
o
v
e
m
e
n
t
Tracking
Desired
Joint
humerus
0 2 4 6 8 10 12 14 16 18 20
-50
0
50
100
150
200
F
1
0 2 4 6 8 10 12 14 16 18 20
-5
0
5
10
15
20
t
F
2
0
2
4
6
-5
0
5
-6
-4
-2
0
2
4
6
Folding Movement
F
l
a
p
p
i
n
g
M
o
v
e
m
e
n
t
Tracking
Desired
Joint
humerus
0 2 4 6 8 10 12 14 16 18 20
50
100
150
200
F
1
0 2 4 6 8 10 12 14 16 18 20
12
14
16
18
t
F
2
7 Liguo Weng, Min Xia, Kai Hu and An Wang: Micro Aerial Vehicle (MAV) Flapping
Motion Control Using an Immune Network with Different Immune Factors
www.intechopen.com
Figure20.Theworkingcurveoftheselfadaptiveswitch
training
data
control
energy
tracking
error
comp.
burden
reconstruction
NIMN Noneed
Low
(26.3)
Low
(0.27)
Low
(0.021)
Noneed
NMN Noneed
High
(312.2)
Low
(0.53)
Low
(0.015)
Noneed
BP
NN
Needed
(1000)
Med
(43.5.)
High
(0.82)
High
(1.23)
Needed
Table3.ComparisonResults
Data training is time consuming and the selection of
trainingdataisverytricky.NIMNandNMNdonotneed
anytrainingdata,butdatatrainingisessentialtoBPNN.
500 sets of training data are applied to train the NN
initially,butthecontrolresultsareveryunstableinusing
the trained BPNN. Therefore, the number of training
data is increasedto1,000sets andthen theNN begins to
workproperly.
. The control
input magnitude of NMN is very high at the beginning,
and hence the control energy that NMN needs is
relatively high compared with that of both NIMN and
BPNN. Because no memory (experience) is gained for
NMN at the beginning, so the control input varies
drastically (a large magnitude) in searching for the
balance point. However, the control energy that both
NINM and BPNN needs is low or medium because of
thehelpoftheinnateresponseandthetrainingdata.
) of BPNN is
almost as good as the other two approaches, but only
providedthatthetrainingdatais sufficientand properly
chosen. However, the time that BPNN needs for one
iterationismanytimeslongerthanthatneededbyeither
NIMN or NMN, as there are many neurons involved in
BPNNscomputation.
7.Conclusions
In this paper, a novel NeuralImmunology/Memory
Network is proposed. It is inspired by the memory
systemandtheimmunesystem.
The proposed NeuralImmunology/Memory Network
overcame the shortages of the NeuralMemory Network.
However, different immune selection factors resulted in
differentnetworkoutputs.Itcanbeobservedthatwitha
more intelligent immune selection factor, the control
outputstendedtobesmoother.
9 Liguo Weng, Min Xia, Kai Hu and An Wang: Micro Aerial Vehicle (MAV) Flapping
Motion Control Using an Immune Network with Different Immune Factors
www.intechopen.com