AIAA Guidance, Navigation, and Control Conference and Exhibit AIAA 2002-4559

5-8 August 2002, Monterey, California

ATMOSPHERIC ASCENT GUIDANCE FOR ROCKET-POWERED LAUNCH

VEHICLES

Greg A. Dukeman∗
NASA Marshall Space Flight Center
Huntsville, AL 35812

Nomenclature
Downloaded by NASA LANGLEY RESEARCH CENTRE on April 4, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2002-4559

Abstract Acronyms
POST Program to Optimize Simulated
Trajectories
An advanced ascent guidance algorithm for rocket-
TAEM Terminal Area Energy Management
powered launch vehicles is developed. The algorithm
cyclically solves the calculus-of-variations two-point
Symbols
boundary-value problem starting at vertical rise
A aerodynamic axial force magnitude
completion through main engine cutoff, taking into
CA coefficient of axial force
account atmospheric effects. This is different from
CN coefficient of normal force
traditional ascent guidance algorithms which operate in
g gravity acceleration vector
a simple open-loop mode until the high dynamic
H Hamiltonian function
pressure portion of the trajectory is over, at which time
h altitude
guidance operates under the assumption of negligible
M Mach number
aerodynamic acceleration (i.e., vacuum dynamics).
m vehicle mass
Judicious approximations are made to reduce the order
N aerodynamic normal force magnitude
and complexity of the state/costate system. Multiple
q dynamic pressure
shooting is shown to be a very effective numerical
technique for this application. In particular, just one qα product of dynamic pressure and
intermediate shooting point, in addition to the initial angle of attack
shooting point, is sufficient to significantly reduce qβ product of dynamic pressure and
sensitivity to the guessed initial costates. An abort to angle of sideslip
downrange landing site formulation of the algorithm is R great circle range
presented. Results comparing guided launch vehicle r vehicle position vector
trajectories with POST open-loop trajectories, for both rx, ry, rz x, y, and z components of position
sub-orbital cutoff conditions and orbit insertion S aerodynamic reference area
conditions, are given verifying the basic formulation of T thrust magnitude
the algorithm. tCutoff predicted cutoff time from guidance
v vehicle velocity vector
vx, vy, vz x, y, and z components of velocity
x column vector containing position
and velocity vectors
xb, yb, zb x-, y-, and z-body axes unit vectors
α vehicle angle of attack
β vehicle angle of sideslip
γ vehicle flight path angle
δ angle between velocity costate vector
and xb vector
* Guidance and Navigation Specialist, Vehicle φ angle between position vector and
Flight Mechanics Group velocity costate vector
φ0 angle between Earth-relative velocity
vector and velocity costate vector

1
American Institute of Aeronautics and Astronautics
Copyright © 2002 by the American Institute of Aeronautics and Astronautics, Inc. No copyright is asserted in the United States under Title 17, U.S. Code.
The U.S. Government has a royalty-free license to exercise all rights under the copyright claimed herein for Governmental purposes.
All other rights are reserved by the copyright owner.
 ∆Τ actual thrust minus vacuum thrust increase, it will be feasible to use more sophisticated
λ column vector containing position algorithms capable of increasing the reliability and
and velocity costates safety of the next generation of launch vehicles.
λr position costate Some of the earliest research into extending the
λry, λrz components of position vector costate capabilities of ascent guidance algorithms so that they
λv velocity costate would be effective just after liftoff was done by Brown,
σdir variable set to +1 or –1, specifying et. al.4. In Ref. 4 a linearized aerodynamics model was
heads-up or heads-down flight, resp. used to obtain the optimal control (thrust direction)
ω0 parameter associated with invocation from the optimality condition in closed form. Curve
of linear gravity field assumption fits for the lift and drag coefficients and atmospheric
density, pressure and speed of sound were used in the
Subscripts guidance formulation to reduce the computational
I, I vector or quantity relative to the burden imposed by aerodynamics modeling. The
Downloaded by NASA LANGLEY RESEARCH CENTRE on April 4, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2002-4559

inertial frame conventional shooting method combined with a

r, r vector or quantity relative to the homotopy procedure was used to solve the two-point
Earth-fixed frame boundary-value problem. The easily obtained vacuum
0 initial solution was first computed, from which a homotopy
procedure was used to re-introduce increments of the
atmospheric effects. Despite the use of homotopy to
Introduction
reduce sensitivity to the initial costate guesses, reliable
NASA, through a program known as the Space convergence was not always attained. Kelly develops a
Launch Initiative1 (also, Second Generation Reusable similar algorithm formulation in Ref. 5 with similar
Launch Vehicle program), has set for itself the goals of convergence difficulties reported, despite attempted
significantly increasing the safety and reliability of the homotopy procedures, due to inclusion of atmospheric
second generation of launch vehicles by two orders of terms. Bradt,et. al.6, use a formulation similar to that of
magnitude (to loss of crew in 1 in 10,000 flights), while Ref. 4 and add a penalty function to reduce bending
reducing the launch costs by an order of magnitude (to moment loads. Cramer, et. al.7, use a nonlinear
$1,000/pound payload). In order for the flight programming approach to guidance and take advantage
mechanics discipline to contribute to these goals, it is of measured day-of-launch winds in the guidance to
important that guidance and control algorithms be provide load relief. Hanson, et. al. develop and test
highly robust and adaptive to in-flight failures such as several atmospheric ascent guidance options in Ref. 8.
partial engine loss. Leung and Calise use a perturbation approach in Ref. 9.
Traditional ascent guidance algorithms have been Calise and Melamed10 use a hybrid collocation
relatively simple. The common methodology is to approach and demonstrate reliable convergence in
operate in ‘open-loop’ mode during the (early) high dispersed guided trajectory simulations. In Ref. 11,
dynamic pressure portion of flight and then, based on a Gath and Calise extend previous work to normal force
pre-determined time or event, switch to a closed-loop and angle of attack path constraints and optimization of
vacuum guidance scheme which operates on the burn-coast-burn sequences.
premise that aerodynamic forces can be neglected. The The rest of this paper is organized as follows. The
open-loop mode typically makes use of pre-loaded next section describes the ascent guidance problem,
tables of Euler attitude commands versus time or followed by the formulation of a robust atmospheric
speed2,3. The closed-loop logic is based on explicit ascent guidance formulation suitable for nominal flight.
formulas and simplified dynamics that result in a semi- Next, an abort guidance formulation is presented.
analytical solution for the optimal steering angles. This Numerical results are then given using the guidance
partitioning of the flight into distinct phases was formulations. The paper is ended with conclusions and
necessary primarily due to computer throughput and recommendations.
memory limitations as well as unavailability of
advanced algorithms that took into consideration Nominal Ascent Guidance Formulation
aerodynamic forces. Introduction of aerodynamic
forces into the problem formulation makes the problem Trajectory Optimization Problem
much more sensitive and computationally-intensive.
Simple, well-understood formulas don’t exist as they The equations of motion for a thrusting rocket in
do in the vacuum case, making the complete liftoff-to- atmospheric flight are:
burnout optimization problem difficult to solve reliably
in real-time. However, as computer power continues to

2
American Institute of Aeronautics and Astronautics
 r& = v ψ i = ψ i (r (tCutoff ), v (tCutoff )) = 0,
(3)
(T − A)x b − Nz b (1) i = 1,K, k , k ≤ 6
v& = g +
m
where the thrust magnitude, T, and the axial and normal Examples of terminal constraints include final position
forces, A and N are given by: magnitude, flight path angle, semi-major axis,
argument of perigee, inclination and longitude of the
T = Tvac + ∆T (h) ascending node.
(2)
A = qSC A N = qSC N In general, path constraints of the form:

Some notes on guidance modeling follow. We’ve S (x, u, t ) ≤ 0 (4)

assumed that all the thrust is aligned along the x-body
axis, xb. The velocity vector, v, can be taken as the are imposed where x is the state and u is the control.
Downloaded by NASA LANGLEY RESEARCH CENTRE on April 4, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2002-4559

Earth-relative velocity vector or inertial velocity vector Examples include maximum axial and normal
depending on the context. For high-speed flight acceleration, maximum normal force, minimum throttle
(typically occurring ‘outside’ the atmosphere), inertial level, angle of attack, angle of sideslip, and min/max
velocity is used whereas for low-speed flight and sub- values of qα, qβ and maximum dynamic pressure. The
orbital missions, it is sufficient to let the velocity in axial force and minimum throttle constraints are fairly
question be Earth-relative velocity. Note that we’ve straightforward to handle. Gath and Calise considered
assumed the force along the y-body axis is small and angle of attack and normal force constraints in Ref. 11
hence we ignore it in the equations of motion and using the mathematical rigor contained in Ref. 12,
consequently in the optimization (but not in the guided section 3.10. Constraints on qα and qβ can be handled
simulations). In the proceeding developments, the in an analogous way because they too are functions of
position and velocity vectors will be expressed in the the state and control variables. The maximum dynamic
guidance coordinate frame illustrated in Figure 1. The pressure constraint, however, is a state variable
guidance coordinate frame is an Earth-centered, right- constraint. It is well-known that state variable
handed, inertial coordinate system with the x-axis inequality constraints are difficult to treat using optimal
aligned with the local vertical and the z-axis aligned control theory. One can also easily conceive of
along the expected downrange direction. It is re- alternative, less rigorous methods of constraining
defined each guidance cycle with the vehicle’s current dynamic pressure that may be just as effective and
latitude, longitude and an azimuth angle which more appropriate for onboard guidance, e.g., Corvin13
approximates the downrange direction of travel. This is uses a feedback control law which provides maximum
a convenient frame to work in because, for example, dynamic pressure control via throttle modulation.
the initial position vector expressed in the guidance We consider now the question of where to point
frame has y- and z-components equal to zero and the y the z-body axis, zb, which is a function of the control xb
components of position and velocity are typically near but is not fully specified given xb. This decision is a
zero. function of how we want the vehicle to fly. We can
The aerodynamic coefficients are modeled using choose to construct zb so that the vehicle flies at zero
least squares polynomial coefficients interpolated with angle of sideslip:
cubic spline functions of Mach number, a common
technique for reduction of aerodynamics coefficient z b || x b × (x b × v ) (5)
data in trajectory optimization. The thrust difference
term, ∆T, due to the effects of the atmosphere on thrust,
or so the vehicle flies a zero degree (“heads-up) or 180
can in general be represented by a cubic spline
degree (“heads-down”) bank angle trajectory:
function. The density, ρ, is represented by a least
squares curve fit of the standard atmosphere and
matches the latter to within 5 percent up to 70 km. z b || x b × (x b × r0 ) (6)
Similarly, a least squares 3rd-order polynomial fit is
used to accurately model the speed of sound, a. The zero-sideslip option requires a non-zero roll
At the time of engine cutoff, tCutoff, k terminal state angle and possibly excessive roll maneuvering. It is
constraints, nonlinear functions of the states, are likely that future launch vehicles will have limited roll
imposed: control authority. The heads-up/heads-down option
will inherently result in larger angles of sideslip but this
can be attenuated fairly easily, if need be, by imposing

3
American Institute of Aeronautics and Astronautics
 a sideslip path constraint. In the proceeding, we adopt In the next subsection, we apply the maximum
the heads-up/heads-down option. principle to obtain the optimal control for atmospheric
The optimization problem can be stated as follows. flight.
Determine the x-body axis history, xb(t), that
maximizes the final vehicle mass (equivalent to Optimality Condition
minimizing fuel usage or minimizing flight time) Applying the maximum principle to the
subject to the equations of motion (1), the terminal Hamiltonian results in the optimization sub-problem:
constraints (3) and the path constraints (4)
  (T − A)x b − Nz b 
Costate Differential Equations max λ v ⋅   (10)
xb
  m 
We note that the atmospheric portion of flight
occurs over a very small ground track, enabling the use Note that the optimal control, xbo (and hence,
Downloaded by NASA LANGLEY RESEARCH CENTRE on April 4, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2002-4559

of the flat-Earth approximations14: optimal z-body axis, zbo) lies in the plane defined by the
(initial) position and velocity-costate vectors. Thus,
h ≅ rx − rE with reference to Figure 2 (after Gath and Calise11) , the
optimization sub-problem (10) can be written simply as
g ≅ [g 0 0 0]T (7)
max{(T − A) cos(δ ) + N sin (δ )} (11)
With these, the state equations become (with explicit δ
state dependencies called out):
Note that we need to be able to evaluate α in terms
r&x = vx r&y = v y r&z = v z of δ so that we can evaluate A and N in the previous
equation. From Figure 2, it is clear that α is some
(T (rx ) − A(rx , v ))xbx − N (rx , v )zbx x
constant, α0, (that is, constant with respect to δ) minus
v&x = − g 0 + x

m δ:
(T (rx ) − A(rx , v ))xby − N (rx , v )zby (8) α = α0 − δ (12)
v& y = x

m
v&z =
(T (rx ) − A(rx , v ))xbz − N (rx , v )zbz The formula for α0 can be derived by setting δ to
m zero and solving for α via:

The costate equations then are given by: v xb = v ⋅ xb = v cos(φ0 )

λ&ry = λ&rz = 0 xb × (xb × rˆ0 ) cos(φ )xb − rˆ0
z b = σ dir = σ dir
sin (φ ) sin (φ )
∂H − (Th − Ah )λ v ⋅ x b + N h λ v ⋅ z b
λ&rx = −
∂rx
=
m vzb = v ⋅ z b = σ dir
[v cos(φ )cos(φ0 ) − vx ] (13)
sin (φ )
∂H A λ ⋅ x + N vx λ v ⋅ z b
λ&vx = − = −λ rx + vx v b  σ dir [v cos(φ )cos(φ0 ) − v x ] 
∂v x m α 0 = tan −1 
Avy λ v ⋅ x b + N vy λ v ⋅ z b
(9)  v sin (φ )cos(φ0 ) 

∂H
λ&vy = − = −λ ry + In the preceding, note that φ and φ0 are simple functions
∂v y m
of the state and costate.
∂H A λ ⋅ x + N vz λ v ⋅ z b The maximization sub-problem can be solved in
λ&vz = − = −λ rz + vz v b
∂v z m many ways. One option is to take the derivative with
respect to δ, set to zero and use an iterative procedure
where the subscripts h, vx, vy, and vz denote partial (e.g., Newton’s method) to get the root which
differentiation with respect to those variables. No corresponds to the optimum δ. This approach was
known analytic solutions for the atmospheric found to be problematic because there are situations
state/costate system exist so we resort to a second-order when the Hamiltonian (as a function of δ) is very flat
Runge-Kutta numerical integration scheme to and Newton’s method is very slow to converge. A
propagate the state/costate system. Ten integration more direct method is to do a Golden Section search15.
steps are sufficient to obtain a good guidance solution. The Golden Section algorithm is relatively inefficient
but the function to be optimized in this case, the
Hamiltonian, is fairly inexpensive to evaluate.

4
American Institute of Aeronautics and Astronautics
 Once the Hamiltonian is maximized, we need to admissible, or ‘transversality vectors’, are obtained a
construct the x- and z-body axes so that we can priori and the inner products of the final costate with
evaluate the state/costate differential equations. Start these vectors are iteratively driven to zero (using a
by expressing the x-body axis as a linear combination modified Newton’s method) simultaneously with the
of costate and initial position vector terminal state constraints. As an example, whenever
argument of perigee is free, the admissible state
variation corresponds to a rotation about the angular
x b = a1 λˆ v + b1rˆ0 (14)
momentum vector, hence, one of the 6-k required
transversality vectors is16:
Dot the preceding with velocity costate and with
initial position to get  r × h  (r ⋅ v )r − r 2 v 
δx =  = 2  (19)
 v × h  v r − (r ⋅ v )v 
cos(δ ) = λˆ v ⋅ x b = a1 + b1 cos(φ )
Downloaded by NASA LANGLEY RESEARCH CENTRE on April 4, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2002-4559

(15)
cos(φ + δ ) = rˆ0 ⋅ x b = a1 cos(φ ) + b1 Note that, because the costates can be arbitrarily
scaled by a positive factor, the end-point condition on
Solve the preceding two equations for a1 and b1 in the Hamiltonian is equivalent to requiring that the
terms of the angles φ and δ: Hamiltonian be positive. This condition is usually
satisfied and is indeed satisfied whenever all the
cos(δ ) − cos(φ )cos(φ + δ ) terminal state constraints are Keplerian constants (e.g.,
a1 = semi-major axis, eccentricity) and the costate
sin 2 (φ ) conditions are satisfied. To see this note that for any
(16)
cos(φ + δ ) − cos(φ )cos(δ ) Keplerian constant, ψ
b1 =
sin 2 (φ )
T
 rT 
The z-body axis is constructed via: ψ& = ψ x x& = ψ x  v T ,− µ 3  = 0 (20)
 r 

σ dir
z b = xb × y b = xb × (xb × rˆ0 )
sin (φ + δ ) hence, [ vT, g(r)T ]T is a transversality vector whenever
 (a1 cos(φ ) + b1 )2 − 1  all the terminal constraints are Keplerian constants
(17)
σ dir   (also, whenever true anomaly is free) and thus
= a1λv y (a1 cos(φ ) + b1 ) satisfaction of the costate conditions imply that the
sin (φ + δ ) 
a1λv z (a1 cos(φ ) + b1 ) ‘Keplerian’ part of the Hamiltonian (at the final time) is
 
zero. Now, note that the sign of the non-Keplerian part
of the Hamiltonian is of definite sign:
Transversality Conditions
T T
The transversality conditions consist of conditions λ Tv uˆ = λ v >0 (21)
m m
on final costate, referred to in the sequel as ‘costate
conditions’, and on the final Hamiltonian:
Hence, H(tf) > 0 and the end-point condition on the
( )
λ T t f = ν Tψ x
(18)
Hamiltonian can generally be replaced by the simpler
(non-constraining) condition that the final costate
H (t f ) = 1 magnitude be unity.

where ν is a column-vector of constant Lagrange Vacuum Guidance Formulation

multipliers, ψ is the column-vector of terminal state
constraints and ψx is the ‘constraint gradients matrix’, Whenever dynamic pressure is reasonably small,
that is, the k by 6 matrix whose k rows are the gradients say, less than 50 psf, the vacuum assumption can be
(wrt state, x) of the terminal state constraints. The invoked. Under the assumption of linear steering, i.e.,:
costate conditions are equivalent to requiring that the
final costate vector be orthogonal to the space spanned At + B
x (t ) = (22)
by all admissible final state variations δx. For a given b At + B
set of k terminal state constraints, expressions for 6-k

5
American Institute of Aeronautics and Astronautics
 and constant thrust magnitude or constant thrust of just one additional shooting point (besides the initial
acceleration, a closed-form solution exists for the thrust shooting point), placed just after peak dynamic
integral requiring no numerical integration or pressure, dramatically reduces the sensitivity compared
quadrature17. The gravity integral is obtained using a to single shooting.
predictor-corrector step. Because the gravity vector
doesn’t change significantly over the ascent trajectory, Abort Guidance Formulation
one step is usually sufficient.
Next, for the costate vacuum solution, we assume a
linear central gravity field 18-19, that is: The abort guidance formulation is very similar to
the nominal guidance formulation. In the case of abort
µ to downrange or abort to launch site, the target values
g (r ) ≅ − r ≡ −ω 02r (23) are obtained from an entry profile21, i.e., a table of
r03
reference altitude, speed, and flight path versus range-
Downloaded by NASA LANGLEY RESEARCH CENTRE on April 4, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2002-4559

to-HAC ‘break points’, available from onboard data

so that the costate differential equations become: used (or generated) by the entry guidance function. An
additional constraint for aborts to a landing site is that
 λ& r   03 ω 02I 3   λ r  fuel be depleted during the ascent burn so that, e.g.,
&  =    (24)
landing gear loads are not exceeded.
λ v  − I 3 03  λ v 
A set of terminal constraints for a downrange abort
case is given by:
whose solution is:
r − rd = 0
λ r (t ) = cos(ω 0t )λ r (t0 ) + ω0 sin (ω0t )λ v (t0 ) v ⋅ r − rd v d sin (γ d ) = 0
(26)
sin (ω0t ) (25) v ⋅ ((r × rHAC )× r ) − v d cos(γ d )r 2 sin (θ ) = 0
λ v (t ) = − λ r (t0 ) + cos(ω 0t )λ v (t0 )
ω0 v ⋅ (r × rHAC ) = 0
where the HAC radius vector, rHAC, is a unit vector
Typically, for a given mission, a good initial guess directed from the center of the Earth to the heading
for the mission-elapsed time at which dynamic pressure alignment cone, the desired values (‘d’ subscripts) are
becomes small is available. Up until that ‘simulated’ obtained from the re-entry profile, and the range angle,
time, the atmospheric equations developed earlier are θ, is the angle subtended by the vehicle position vector
used in the (numerical) propagation of state/costate. At and the HAC radius vector. The first constraint fixes
that point, the current propagated state/costate is the altitude, the second and third fix the vertical and
handed off to the ‘vacuum propagator’ (after Ref. 8). It horizontal speeds, resp., and the fourth nulls the vehicle
is advantageous to do this, that is, only use the heading error with respect to the HAC. Note that the
atmospheric equations when absolutely necessary, third ensures that the vehicle is actually headed toward
because the atmospheric equations are much more the HAC and not away from it.
complicated than the analogous vacuum equations. For the case of downrange aborts, we want to burn
Note that it is not critical that the predicted time of all the propellant so, instead of optimizing fuel usage,
‘atmospheric exit’, tExo, be especially accurate. If need we choose to maximize final speed v. The
be, the predicted dynamic pressure at tExo can be transversality conditions are:
monitored and tExo increased if necessary.
(λ − φ x ) t =t ⋅ δxi = 0, i = 1,2
Numerical Solution Method f
(27)
µ
20
The multiple shooting method is used to solve the
( )
H t f = λ ⋅ f = λr ⋅ v − λv ⋅ 3
r=0
r
resulting two-point boundary-value problem. The where we’ve assumed that we intercept the re-entry
values of the six initial costates, and the engine cutoff profile on a coast arc, ϕx is the gradient of the
time are the free variables that must be iterated upon to performance index with respect to state, x, and where
null the k terminal state constraints, the 6-k costate
the ‘admissible’ state vectors, δxi, are given by:
conditions and the constraint on final costate
magnitude. Multiple shooting allows the user to guess
state/costate values at more than just the initial point,  r × rHAC 
δx1 =  
significantly reducing the well-known sensitivity to the  v × rHAC  (28)
initial costates. In fact, it has been found that insertion

6
American Institute of Aeronautics and Astronautics
 and parameters to be iteratively determined are the six
initial costates and the time of flight.
 ∂r r r h × r 
 + 
∂R r Re h × r (29) Numerical Results
δx 2 =  
 ∂v v h × v  ∂γ 1 
 −v  − 
 ∂R v h × v  ∂R Re  The algorithm formulation as described above has
been coded in C and integrated into the high-fidelity
MAVERIC/X-33 trajectory simulator which includes 6
The transversality vector δx1 reflects the fact that a degrees of freedom, a day-of-launch winds model, and
perturbation of the final position normal to the flight the GRAM atmosphere model. For sub-orbital
plane combined with a perturbation of the final velocity missions, the X-33 vehicle model was used. To enable
normal (but opposite in sign) to the flight plane is orbital insertion missions, the Isp of the X-33 vehicle
admissible. The transversality vector δx2 is a
Downloaded by NASA LANGLEY RESEARCH CENTRE on April 4, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2002-4559

was doubled. The end of the atmospheric phase was

calculation of the admissible change in final position taken as 140 seconds inside the guidance solution
and velocity given a perturbation of final range (treated process whereafter the vacuum solution is used. One
as a free variable). intermediate shooting point was used. A qα constraint
The transversality conditions (27) can be written: of 500 psf-deg was imposed inside the guidance
λ ⋅ δx1 − φx ⋅ δx1 = 0 solution process. For the two orbital missions, an axial
λ ⋅ δx 2 − φx ⋅ δx 2 = 0 (30) acceleration limit of 3.5 g’s was imposed. As a rough
( )
H tf = 0 indicator of algorithm efficiency, the 3 degree-of-
freedom simulations run significantly faster than real-
time on a three-year old DEC alpha
or, equivalently as: computer/processor. The initial guess was generated
from the vacuum solution although other simpler
λ ⋅ δx1 − φx ⋅ δx1 = 0 methods would probably work just as well. Results for
(φx ⋅ δx1 )(λ ⋅ δx2 ) − (λ ⋅ δx1 )(φx ⋅ δx2 ) = 0 (31a,b,c) three mission simulations are shown here: 1) A 6
( )
H tf = 0 degree-of-freedom sub-orbital mission simulation
launching from Edwards Air Force Base to Michael
Army Air Field in Utah, 2) a 3 degree-of-freedom
Suppose all the necessary conditions are satisfied orbital mission launching from Kennedy Space Center
except (31a). Then, (31a) can be solved by scaling the to a 100 by 100 nautical mile circular, 28.5 degree
costate vector by an appropriate factor. Note that the inclination orbit, and 3) a 3 degree-of-freedom orbital
arbitrary scaling has no effect on the necessary mission launching from Kennedy Space Center with
conditions or on the trajectory. If we enforce (31b) simulated 50 percent thrust loss at 90 s into the mission
instead of both (31a) and (31b), we can retain our logic requiring a downrange abort to a runway in Spain. For
from the nominal guidance formulation wherein we the first mission, the X-33 vehicle model3 was used and
keep the initial costate vector normalized, |λv(t0)|=1 and for the remaining two, the X-33 vehicle with increased
impose the non-constraining constraint |λv(tf)|=1. Now, Isp was used. The guidance in all cases was executed
substituting v for ϕ, we have: at 1 Hz with complete updates (re-optimized
 v v y vz  trajectories) each cycle starting at liftoff, although for
ϕ x = 0 0 0 x  (32) the first four seconds, the guidance solution was
 v v v  ignored and the guidance commands were set to
execute a vertical rise. In all cases, open-loop reference
Then, using (31b), (31c), and the second of (32), the trajectories (from POST) are available and comparison
transversality conditions reduce to the simple form: plots were made to verify acceptable performance of
the guidance. In the orbital case, the guided trajectory
λ ⋅ δx1 = 0 ⇒ λ r ⋅ (r × rHAC ) + λ v ⋅ (v × rHAC ) = 0 was very comparable to the open-loop trajectory with
( )
H tf = 0
(33) the guided trajectory out-performing the POST
trajectories by an insignificant 200 pounds fuel.
Figures 3-5 pertain to the sub-orbital mission.
To sum up the constraints, we have the four final Figure 3 compares the ascent altitude and flight path
state constraints (26), the transversality conditions (33), angle histories of the MAVERIC closed-loop and
and the (non-constraining) constraint |λv(tf)| = 1. The POST open-loop simulations. Although the histories
are noticeably different in the middle of the trajectory,

7
American Institute of Aeronautics and Astronautics
 the end-points are virtually identical. Figure 4 shows
the qα and qβ hisories and Figure 5 shows the ground
track. References
Figures 6-8 pertain to the 100 nmi circular orbit
insertion mission. Again, very comparable 1
performance and insertion accuracy are obtained. The Anon., “The Space Launch Initiative: Technology to
Pioneer the Space Frontier,” NASA Marshall Space Flight
ground tracks differ by the geodetic to geocentric
Center, Pub. 8-1250, FS-2001-06-122-MSFC, June, 2001.
correction. 2
McHenry, R. L., Brand, T. J., Long, A. D., Cockrell, B. F.,
Figures 9-15 pertain to the abort-to-downrange site and Thibodeau, J. R. III, “Space Shuttle Ascent Guidance,
mission. Figure 9 shows the time and magnitude of the Navigation, and Control”, Journal of the Astronautical
thrust loss. Figure 10 shows the altitude and speed Sciences, Vol. XXVII, No. 1, pp. 1-38, January-March, 1979.
3
profiles for the entire ascent through entry. The latter Hanson, J. M., Coughlin, D. J., Dukeman, G. A., Mulqueen,
phase was flown with closed-loop entry guidance to J. A., McCarter, J. W., “Ascent, Transition, Entry, and Abort
Downloaded by NASA LANGLEY RESEARCH CENTRE on April 4, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2002-4559

further verify that the ascent guidance delivered the Guidance Algorithm Design for the X-33 Vehicle”, AIAA
vehicle to an entry-compatible state. Figure 11 shows Paper 98-4409, 1998.
4
Brown, K. R., Harrold, E.F., Johnhson, G. W., “Some New
the ground track which is compatible with landing in
Results on Space Shuttle Atmospheric Ascent Optimization”,
Spain. Figure 12 shows the angle of attack and AIAA Paper No. 70-978, 1970.
sideslip. Apparently, the optimal thing to do to adapt to 5
Kelly, W. D., “Formulation of Aerodynamic Quantities for
thrust loss is to pitch up. Figure 13 shows qα and qβ Minimum Hamiltonian Guidance”, AIAA Paper 92-4380,
which are both fairly benign. Figure 14 shows the 1992.
6
heading error with respect to the HAC. This verifies Bradt, J. E., Jessich, M. V., and Hardtla, J. W., “Optimal
that the ascent guidance is successful at nulling heading Guidance for Future Space Applications,” AIAA Paper 87-
error at main engine cutoff. Figure 15 shows the 2401, August 17-29, 1987, Monterey, CA.
7
commanded cutoff time, an output from guidance. The Cramer, E. J., Bradt, J. E., Hardtla, J. W., “Launch
Flexibility Using NLP Guidance and Remorte Wind
prediction is quasi-constant indicating that the guidance
Sensing”, AIAA Paper 90-3330-CP, 1990.
is sufficiently accurately modeling aerodynamics, 8
Hanson, J., Shrader, M., Cruzen, C., “Ascent Guidance
gravity and thrust. Options”, AIAA Paper 94-3568-CP, 1994.
9
Leung, M. S. K., Calise, A. J., “A Hybrid Approach to Near-
Summary and Conclusions Optimal Launch Vehicle Guidance”, AIAA Paper 92-4304-
CP, 1992.
10
This paper describes an ascent guidance algorithm Calise, A. J., Melamed, N., and Lee, Seungjae, “Design and
that re-optimizes the entire ascent trajectory each Evaluation of a Three-Dimensional Optimal Ascent Guidance
guidance cycle from liftoff to main engine cutoff. Algorithm”, Journal of Guidance, Control, and Dynamics,
vol. 21, no. 6, pp. 867-875, Nov-Dec 1998.
High-fidelity guided trajectories compared with POST 11
Gath, P., and Calise, A., "Optimization of Launch Vehicle
open-loop trajectories demonstrate that it provides Ascent Trajectories With Path Constraints and Coast Arcs",
near-optimal performance despite flat-Earth AIAA Paper 99-4308, August 1999.
simplifications of the costate equations. The high- 12
Bryson, Arthur, E., Jr., and Ho, Yu-Chi, "Applied Optimal
fidelity trajectory simulator with the guidance cycling Control", Hemisphere Publishing Corporation, 1975.
13
at 1 Hz runs significantly faster than real-time, Corvin, M.A., ‘Ascent Guidance for a Winged Boost
indicating the algorithm’s efficiency. An abort-to- Vehicle’, NASA-CR-172083, August 1988.
14
downrange site formulation is given along with guided Vinh, N.X., “Integrals of the Motion for Optimal
abort trajectory results. Trajectories in Atmospheric Flight”, AIAA J., 11, (1973),
700-703.
Future work will involve using day-of-launch 15
Press, W.H., Teukolsky, S.A., Vetterling, W.T., Flannery,
winds in the guidance solution and incorporating B.P., “Numerical Recipes in C: The Art of Scientific
Mach-scheduled angle of attack constraints. Several Computing”, 2nd ed., Cambridge University Press, 1992.
candidate RLVs are 2-stage vehicles with gimbaled 16
Teren, Fred, and Spurlock, Omer F., “Optimal Three
engines, possibly requiring modification of the Dimensional Launch Vehicle Trajectories with Attitude and
assumption that all thrust is directed along the x-body Attitude Rate Constraints”, Lewis Research Center, NASA
axis. Technical Note, NASA TN-D-5117, NASA, Washington,
D.C., March 1969.
17
Acknowledgement Burrows, R. R., McDaniel, G. A., “A Method of Trajectory
Analysis With Multi-Mission Capability and Guidance
The POST trajectory data generated by Terri
Application”, AIAA Paper No. 68-844, August 1968.
Schmitt of the Vehicle Flight Mechanics Group at
NASA-Marshall is gratefully acknowledged.

8
American Institute of Aeronautics and Astronautics
 18
Jezewski, D.J., “Optimal Analytic Multiburn Trajectories,”
AIAA Journal, Vol. 10, No.5, 1972, pp. 680-685.
19
Azimov, D.M., Bishop, R.H., “Extremal Rocket Motion 5
2 10 100
with Maximum Thrust in a Linear Central Field,” Journal of
Spacecraft and Rockets, Vol. 38, No.5, September-October
80
2001, pp. 765-776. 5
1.5 10
20
Stoer, J., and Bulirsch, R., “Introduction to Numerical
altitudeFt_Ref 60

altitudeFt
Analysis”, Springer-Verlag, New York, 1980.

fpaDeg
1 10
5 altitudeFt
21
Lu, P., Shen, Z., Dukeman, G., Hanson, J., “Entry
40
Guidance by Trajectory Regulation”, AIAA paper 2000-3958, fpaDeg_Ref
Proceedings of Guidance, Navigation and Control 5 10
4 fpaDeg
20
Conference, August 14-17, 2000, Denver, CO.
Downloaded by NASA LANGLEY RESEARCH CENTRE on April 4, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2002-4559

0 0
0 2000 4000 6000 8000 10000
vRelMagFt

Figure 3: Altitude/flight path, POST/MAVERIC

North Pole
Launch site
Comparison, Sub-orbital Mission

zg
xg 1500 500
ψ zg
1000
0

qAlphaPsf 500

qBetaPsf
yg -500
0

qBeta_Ref -1000
-500 qAlpha_Ref qBeta
qAlpha

Figure 1: Guidance Reference Frame -1000 -1500

0 2000 4000 6000 8000 10000
vRelMagFt

Figure 4: q-α, q-β, POST/MAVERIC Comparison,

ro Sub-orbital Mission
-zb
N λv

δ xb
φ

φ0

Figure 2: Angles Pertaining To Optimality Condition

9
American Institute of Aeronautics and Astronautics
 28.6
36.2
28.5
36
latDeg_Ref 28.4
35.8
latDeg

GeodLat
28.3
35.6
latDeg

28.2
35.4
28.1 GeodLat_POST
35.2 latDeg
28
Downloaded by NASA LANGLEY RESEARCH CENTRE on April 4, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2002-4559

35
27.9
34.8
-82 -80 -78 -76 -74 -72 -70
-117.8 -117.6 -117.4 -117.2 -117 -116.8 -116.6 -116.4 Long
lonDeg

Figure 8: Ground Track, POST/MAVERIC

Figure 5: Ground Track, POST/MAVERIC
Comparison, Orbital Mission
Comparison, Sub-orbital Mission

5
7 10 100 5
5 10
5
6 10
altitudeFt 80 4 10
5
5
5 10 altitude_POST
5
60 3 10
altitudeFt

5
4 10
fpaDeg

thrust

5
5
3 10 2 10
40
fpaDeg
5 5
2 10 RelFPA_POST 1 10
20
5
1 10
0

0 0 5
-1 10
0 5000 10000 15000 20000 25000
vRelMagFt 0 100 200 300 400 500 600
time
Figure 6: Altitude/flight path, POST/MAVERIC
Figure 9. Thrust (lbf), Abort Mission
Comparison, Orbital Mission

1000 1500 5 4
3 10 2.5 10
1000
qBeta 5
qBeta_POST 2.5 10 4
500 2 10
500
5
2 10
qAlphaPsf

qBetaPsf

0 4
vRelMagFt

1.5 10
altitudeFt

0 5
-500 1.5 10
4
1 10
5
-1000 1 10 altitudeFt
-500 qAlpha
qAlpha_POST -1500 4 5000
5 10
vRelMagFt
-1000 -2000
0 0
0 1000 2000 3000 4000 5000
vRelMagFt 0 500 1000 1500 2000
time

Figure 7: q-α, q-β, POST/MAVERIC

Figure 10: Altitude/Speed, Abort Mission
Comparison, Orbital Mission

10
American Institute of Aeronautics and Astronautics
 45 10

40 5

headErrHACdeg
latDeg

35 0

30 -5

25 -10
-100 -80 -60 -40 -20 0 -500 0 500 1000 1500 2000
Downloaded by NASA LANGLEY RESEARCH CENTRE on April 4, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2002-4559

lonDeg time

Figure 11: Ground Track, Abort Mission Figure 14: HAC heading error, Abort mission

30 10 600

25 8
550
20 6
500
15 4
alphaDeg

alphaDeg
betaDeg

tCutoff
10 2 450

5 0
400
0 -2
betaDeg
350
-5 -4

-10 -6 300
0 100 200 300 400 500 600 0 100 200 300 400 500 600
time time

Figure 12: Angle of attack/sideslip, Abort mission Figure 15: Guidance predicted cutoff time, Abort
mission

2000

1500

1000 qAlpha
qBeta
qAlpha

500

-500

-1000
0 100 200 300 400 500
time

Figure 13: qα, qβ, Abort mission

11
American Institute of Aeronautics and Astronautics