ICCA2010-SDRE Final

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

SDRE Missile Guidance Law

Feng Tyan and Jeng Fu Shen

Abstract— In this work, a state dependent Riccati equation the computational effort of solving SDRE can be greatly
(SDRE) missile guidance is formulated and studied. With the reduced since it is only a third order system. In general
aid of a line of sight (LOS) fixed coordinate, the order of SDRE technique is a regulator in nature, while for guidance
the system equations describing the three dimensional missile
guidance problem is reduced to three. In addtion, the compu- problem it is not necessary to regulate all the states to zero.
tational effort of solving SDRE is less than the usual guidance Hence we change the state variables by a simple shifting to
formulations which often have the order greater than three. accommodate this nature of SDRE.
Three different state dependent coefficient forms are proposed This paper is organized as followings. In Section II, the
and compared under the consideration of optimality and equation of motion describing the relative dynamics between
implementation. For more constructive design, a simple shift
of variable is also introduced to achieve better performance. target and missile is derived in a LOS fixed coordinate. Then
From the results of numerical experiments, we also find that a brief introduction of SDRE technique is given in section
the proposed SRRE guidance law outperforms the conventional III for completeness. In Section IV, we utilize the SDRE
PPN guidance (for navigation constant equals 5) in the sense technique to synthesize the nonlinear guidance law. Different
of time but at the cost of energy. SDC forms will be taken into account therein. For more
Keywords – State dependent Riccati equation, guidance law
constructive design, we simply change the state variables to
I. I NTRODUCTION achieve better performance. In Section V, we compare the
Most of the guidance laws adopted either the Cartesian performance of the proposed SDC forms for the case of both
coordinate or polar coordinate system. The difficulty, when nonmaneuvering and maneuvering target through numerical
adopting the aforementioned systems, is that the equation of experiments. Finally some conclusion are drawn in Section
motion becomes complicated. This inevitably arise a certain VI.
difficulty for analysis. In addition, many of the approaches
II. R ELATIVE DYNAMICS IN 3D SPACE
focused on two dimensional case to avoid the complexity
of the equations of motion as in three dimensional space. It Assume that both missile and target are particles in a 3-
has been shown, the line of sight (LOS) fixed coordinate will dimension space and define the LOS vector as
lead to a simple equation of motion, with only three states to
fully describe three dimensional missile guidance [1]. In this r = rT − rM = ρer , (1)
paper, we will adopt the LOS fixed coordinate for deriving where
the three dimensional relative dynamics between missile and
rT , rM are the position vectors of the target and missile
target then design the guidance law for missile to pursuit the
in an inertial frame, respectively,
target.
Recently, a nonlinear suboptimal method, namely SDRE ρ is the range between target and missile and
technique [2–4], has been recognized and applied to numer- er the unit vector in the direction of LOS.
ous nonlinear problems. The advantages of the method is that The relative velocity and acceleration are
the control law can be obtained systematically and relatively
ṙ = ρ̇er + ρėr = vT − vM , (2)
easy to derive with respect to the optimal theorem. The
performance of the controller can be directly affected by the r̈ = ρ̈er + 2ρ̇ėr + ρër = aT − aM , (3)
weighting matrices of the cost function and the choice of sys- where vT , aT , vM and aM are the velocity and acceleration
tem’s state dependent coefficient (SDC) form. Unfortunately, vectors of target and missile, respectively. Assume that the
the global stability of SDRE technique remains no conclusion angular velocity vector of LOS, Ω, is orthogonal to LOS and
up to now, only local stability or region of attraction can be and define a LOS fixed coordinate system (er , et , eΩ ) where
obtained [5–7].
SDRE missile guidance had also been studied in the past ėr Ω
et , , eΩ , . (4)
few years, see e.g. [8, 9]. In most formulations, the order kΩk kΩk
of the system exceeds three and hence, computational effort
The geometry of the coordinate is shown in Fig. 1.
of solving state dependent Riccati equation (SDRE) is quit
If the accelerations of both target and missile are expressed
large. In this work, with the aid of a LOS fixed coordinate,
in this (er , et , eΩ ) coordinate,
Feng Tyan is with Computational Dynamics and Control Lab, Department
of Aerospace Engineering, TamKang University, Tamshui, Taipei County aT , aTr er + aTt et + aTΩ eΩ ,
25147, Taiwan, R.O.C. tyanfeng@mail.tku.edu.tw
Jeng Fu Shen was with the Computational Dynamics and Control Lab aM , aMr er + aMt et + aMΩ eΩ ,
V For completeness of SDRE technique, the following the-
T

VM Target
orems are provided. The corresponding proofs can be found
et r in [2, 5, 11].
Theorem 3.1: Assume that both f (x), A(x), B(x) Q(x)
er and R(x) are continuous. Further assume that the pair
Missile rT
(A(x), B(x)) is controllable parameterization and the pair
eΩ rM Z (A(x), Q(x)) is observable parameterization. Then the non-
linear system with SDRE feedback law has local asymptot-
ically stability.
X
O Y Theorem 3.2: Let B be an arbitrary large open ball cen-
tered at the origin with radius r > 0. Assume that the func-
Fig. 1. LOS Fixed Coordinate tions A(x), B(x), P(x), Q(x) and R(x), along with their
gradients, Axi (x), Bxi (x), Pxi (x), Qxi (x) and Rxi (x),
i = 1, 2, · · · , n, are bounded in B. Then, in SDRE feedback
law, under asymptotic stability, as the state x is driven to
it can be shown that the equations of motion are the follow-
zero, the necessary condition for optimality is asymptotically
ings [10],
satisfied at a quadratic rate.
d The process of SDRE technique can be summarized as
ρ̇ = kΩk2 + (aTr − aMr ), (5a)
dt follows:
d
ρkΩk = −ρ̇kΩk + (aTt − aMt ), (5b) 1. Bring the nonlinear system (7) into it’s SDC form (9).
dt 2. Choose the weighting matrices Q(x) and R(x).
along with one kinematic equation 3. Solve the SDRE (11) to obtain P, and the related
d controller is given by (10).
ρ = ρ̇. (6) Note that A(x) should be both controllable and observable
dt
The three scalar differential equations fully describe the parameterization to insure the existence of P.
relative dynamics between target and missile. IV. SDRE BASED G UIDANCE L AW
III. I NTRODUCTION OF SDRE T ECHNIQUE Assume that target does not maneuver and let the state
and control variables be
Consider a nonlinear, input affine system  
ρ̇  
ẋ = f (x) + B(x)u, (7) −aMr
x , ρkΩk , u ,
  , (12)
−aMt
subject to the performance index ρ

1 ∞ T
Z then the state equation of the system can be expressed as
J= [x Q(x)x + uT R(x)u]dt (8)
2 0 ẋ = f (x) + Bu, (13)
where the weighting matrices are chosen such that Q ≥ 0 where
and R > 0. Equation (7) can be written into an SDC form
 
x22
 
1 0
as  xx13x2 
f (x) = − x3  , B = 0 1 . (14)
ẋ = A(x)x + B(x)u . (9) 0 0
x1
Note that the expression of A(x) is not unique. Associate At first, we define two state dependent coefficient matrices
with the SDC form (9), the following definitions are consid- as
ered [2]. x2
 
0 2 xx23 − x22
Definition 3.1: A(x) is a controllable (stabilizable) para- 3 
A1 (x) = − xx2 − xx1 x1 x2 2  (15)

meterization of the nonlinear system in a region D if the pair 3 3 x3
(A(x), B(x)) is point-wise controllable (stabilizable) in the 1 0 0
linear sense for all x in D. and  x2 
Definition 3.2: A(x) is a observable (detectable) parame- 0 x3 0
terization of the nonlinear system in a region D if the pair A2 (x) = − xx23 0 0 , (16)
(A(x), Q(x)) is point-wise observable (detectable) in the 1 0 0
linear sense for all x in D.
The SDRE feedback control law is given by where A1 is chosen such that A1 (x)x = ∂f∂x (x)
x. For the
SDC form A1 , the optimality criterion [4] is relatively
u = −R−1 BT Px, (10) simple compared to that of A2 . Hence, we expect that the
SDRE feedback law of A1 will have a better performance
where P solves state dependent Riccati equation
than that of A2 . The choice of A2 is worthy to study since
AT P + PA − PBR−1 BT P + Q = 0. (11) it involves only the term xx23 , which is the magnitude of
1

0.9

0.8
the angular rate of LOS. It has the advantage in realization Z
VM
since the most commonly measured state is the magnitude of 0.7

angular rate of LOS. (Note that it doesn’t mean that SDRE Y


0.6
feedback law associated with A2 is not a full state feedback.)
The SDRE technique is a regulator in nature, but in fact 0.5 φe
we don’t want to regulate the range rate, x1 . It is obvious that 
we can achieve this goal simply by nullifying the weighting 0.4
) θa X
O
on x1 in Q matrix. But for a more constructive design, we
shift the state x1 and define the states as 0.3
Fig. 2. Conventions of Azimuth and Elevation angles
   
x̄1 x1 + η 0.2
TABLE I
x̄ , x̄2  =  x2  (17)
NONMANEUVERING TARGET
x̄3 x3 0.1

A1 A2 A3
where η is a preselected positive constant. Then the dynamic 0
0 45o
θe = 0.1 0.2 tf − t00.3 6.1929
0.4 18.8521
0.5 4.1882 0.7
0.6 0.8 0.9
equation for this shifted variables is φa = 30o Jc (×103 ) 1.9793 2.2905 1.6263
θe = 135o 6.2390 18.9339 4.2771
x̄22
   
1 0 φa = 30o 2.0901 2.4025 1.7430
x̄ 3
θe = 225o 6.4004 19.1880 4.5215
x̄˙ = − (x̄1 −η)x̄2  + 0 1 u. (18)
 
x̄3 φa = 30o 2.4920 2.8138 2.1548
x̄1 − η 0 0 θe = 315o 6.4410 19.1293 8.7552
φa = 30o 2.4254 2.7639 2.9155
In this case, we choose SDC form as θe = 45o 6.3363 19.1010 4.6871
φa = 225o 2.6772 2.9091 2.3317
x̄2
2 x̄x̄23
 
0 − x̄22 θe = 135o 6.3743 19.2724 8.9349
3 φa = 225o 2.6153 2.9066 3.1196
A3 (x̄) = 0 −2 x̄1 −η (x̄1 −η)x̄ . (19)
 2

x̄3 2 x̄3 θe = 225o 6.1945 18.9481 4.3036
φa = 225o
1 0 − x̄η3 2.0896 2.3896 1.7460
θe = 315o 6.2159 19.0387 4.4110
φa = 225o 2.2154 2.5009 1.8762
Note that A3 (x̄) 6= ∂f∂(x̄)
x̄ . The performance and properties
of the aforementioned SDC forms will be examined through
numerical experiments in next section.
go and energy expenditure. All the three SDC forms capture
V. N UMERICAL E XAMPLES nonmaneuvering target under the considered scenario.
Example 5.1: In this example, the SDC forms proposed Example 5.2: Although maneuvering target was not for-
in Section IV are studied. The initial positions of target and mulated in section IV, we still consider the case that target
missile are at has a spiral motion as shown in Fig. 3. The radius of spiral
is 300m at a constant speed of 400m/sec in the Y direction,
rT (t0 ) = (1000, 3000, 1000), rM (t0 ) = (0, 0, 0), the magnitude of velocity is 500m/sec. The scenario is the
in inertial Cartesian coordinate. Target is assumed to be not same as that given in example 5.1. The results are shown in
maneuvering and moving in the direction of Y axis at a Table II. Unfortunately, both SDC forms A1 and A2 couldn’t
constant speed 300 (unit length/second). The direction of capture target under these conditions. while SDC form A3
the initial velocity of missile is determined by azimuth and performs well even though target maneuvers. After some
elevation angles, θa and φe , respectively (see Figure 2). The observations, we conclude that the regulation of ρ̇ via SDC
initial speed of missile is twice as that of target. Various
launch angles are considered. The weighting matrices asso-
ciated with SDC forms A1 , A2 and A3 (with η = 300) are
Initial Position of Target
all chosen as Trajectory of Target
  1000
1 0 0  
5 0
Q = 0 1 0 , R = 800
0 5
0 0 1 600
Z

400
for illustration purpose. To compare the energy expenditure,
200
we define the cost as
Z tf 0
15000
Jc = ||aM ||dt . 10000 1000
t0 5000
500
0
−500
The simulation stops when the end condition ρ ≤ 1 is Y 0 −1000 X

satisfied. From Table I, it is obvious that although SDC form


A2 requires less information of states, but the performance is Fig. 3. Illustration of Spiral Motion
the worst. Furthermore, A3 outperforms A1 in both time to
TABLE II TABLE III
MANEUVERING TARGET NONMANEUVERING TARGET: PPN AND SDC F ORM A3

A1 A2 A3 A3 PPN
θe = 45o tf − t0 - - 3.2773 θe = 45o tf − t0 4.1882 11.1171
φa = 30o Jc (×103 ) - - 3.2860 φa = 30o Jc (×103 ) 1.8607 0.9058
θe = 135o - - 3.2657 θe = 135o 4.2771 11.3290
φa = 30o - - 3.4783 φa = 30o 1.7430 0.7613
θe = 225o - - 3.4839 θe = 225o 4.5215 12.9390
φa = 30o - - 4.1450 φa = 30o 2.1548 1.2396
θe = 315o - - 3.5084 θe = 315o 8.7552 12.4375
φa = 30o - - 4.0303 φa = 30o 2.9155 1.1535
θe = 45o - - 3.7338 θe = 45o 4.6871 19.1197
φa = 225o - - 4.2285 φa = 225o 2.3317 1.5843
θe = 135o - - 3.7912 θe = 135o 8.9349 14.4775
φa = 225o - - 4.1427 φa = 225o 3.1196 1.4456
θe = 225o - - 3.4979 θe = 225o 4.3036 11.3275
φa = 225o - - 3.1424 φa = 225o 1.7460 0.7419
θe = 315o - - 3.4519 θe = 315o 4.4110 11.7545
φa = 225o - - 3.3793 φa = 225o 1.8762 0.8595

TABLE IV
SDC F ORM A3 : EFFECT OF η
Initial position of target
Initial position of missile tf − t0 Jc (×103 )
1500 Trajectory of target
SDC A
1
η = 50 5.2269 1.8530
Target
SDC A2 η = 100 4.8911 1.7871
1000 A1 SDC A3 η = 150 4.6543 1.7359
A2
η = 200 8.9760 2.3383
500
η = 250
Z

4.3176 1.6574
A3
η = 300 4.1882 1.8607
0
η = 350 8.2849 2.5144
η = 400 8.1069 2.5721
−500
15000
η = 450 7.9455 2.6294
10000 2000 η = 500 7.7972 2.6866
5000
1000 η = 550 3.7294 1.5215
0
Y −1000
0 X
η = 600 7.5349 2.8012

Fig. 4. Trajectory of SDRE Guidance


VI. CONCLUSIONS
With the unified approach based on the LOS fixed co-
ordinate, three dimensional SDRE missile guidance was
forms A1 and A2 would result in miss of capturing target.
formulated and studied. The order of the system is reduced
Fig. 4 depicts the trajectories of one of the cases in the 3-
to three only and has its computational advantages in solving
dimensional space. It can be observed that missile guided
solutions of Riccati equation. Different SDC forms have been
by SDC forms A1 and A2 has a motion similar to that of
proposed and studied. From the numerical experiments, SDC
target.
form with shifted variables outperforms others in the sense of
Example 5.3: A most practical missile guidance law, time to go and defined cost. Furthermore, for maneuvering
namely PPN, is herein compared to SDC form A3 . By using target, missile guided by SDC forms A1 and A2 missed
PPN law (aM = N Ω × vM ) with navigation gain N = 5 target under the considered scenario, only SDC form A3
and same scenario given in example 5.1, the simulation performs well for each case.
results are shown in table III. These results reveal that in
the performance index time to go, missile guided by SDC R EFERENCES
form A3 outperforms the one guided by PPN.
Example 5.4: In the previous experiments the results [1] F. Tyan, “An Unified Approach to Missile Guid-
show that SDC form A3 performs better than PPN in the ance Laws: A 3D Extension,” IEEE Transactions on
sense of time to go. In this example we adopt different value Aerospace and Electronic Systems, vol. 4, no. 41, pp.
of η to evaluate the performance of SDC form A3 . General 1178–1199, Oct 2005.
speaking, the choice of η depends on experience, but it can be [2] J. R. Cloutier, “State-Dependent Riccati Equation Tech-
thought of as the closing velocity at the instant of intercepting niques: An Overview,” in Proceedings of American
target. As we can see in table IV, some of the values of η Control Conference, vol. 2, June 1997, pp. 932–936.
perform poor as compared with the others. The reason is not [3] ——, “The Capabilities and Art of State-Dependent Ric-
clear now. Yet, the performance under these values are still cati Equation-Based Design,” in Proceedings of Ameri-
superior to that of PPN. can Control Conference, vol. 1, May 2002, pp. 86–91.
[4] J. R. Cloutier, C. N. D’Souza, and C. P. Mracek,
“Nonlinear Regulation and Nonlinear H∞ Control Via
the State-Dependent Riccati Equation Technique: Part 1,
Theory, Part 2, Examples,” in Proceedings of the Inter-
national Conference on Nonlinear Problems in Aviation
and Aerospace, May 1996, pp. 117–130.
[5] H. T. Banks, B. M. Lewis, and H. T. Tran, “Nonlinear
Feedback Controller and Compensators: A States De-
pendent Riccati Equation Approach,” in Proceedings of
American Control Conference, vol. 1, May 2002, pp.
86–91.
[6] E. B. Erdem and A. G. Alleyne, “Estimation of Stability
Regions of SDRE Controlled Systems Using Vector
Norms,” in Proceedings of American Control Confer-
ence, vol. 1, May 2002, pp. 80–85.
[7] J. W. Curtis and R. W. Beard, “Ensuring Stability of
State-depend Riccati Equation Controllers Via Satisfic-
ing,” in Proceedings of the 41st IEEE Conference on
Decision and Control, Dec 2002, pp. 2645–2650.
[8] M. Xin and S. N. Balakrishnan, “Integrated Guidance
and Control of Missile with θ − d Method,” American
Control Conference.
[9] P. Menon and E. Ohlmeyer, “Integrated Design of Ag-
ile Missile Guidance and Autopilot Systems,” Mediter-
ranean Control Conference, pp. 28–30, June 1999.
[10] F. Tyan and J. F. Shen, “A Simple Adaptive GIPN
Missile Guidance Law,” in Proceedings of American
Control Conference, Minneapolis, Minnesota, U.S.A.,
June 14-16 2006.
[11] J. R. Cloutier and P. H. Zipfel, “Hypersonic Guid-
ance Via the State-Dependent Riccati Equation Control
Method,” Proceedings of IEEE Conference on Control
Applications, vol. 1, pp. 219–224, Aug 1999.

You might also like