Dileep M V et al. Int. Journal of Engineering Research and Applications www.ijera.

com
ISSN : 2248-9622, Vol. 4, Issue 1( Version 1), January 2014, pp.116-121

RESEARCH ARTICLE OPEN ACCESS

Trajectory Optimization of Launch Vehicles Using Steepest

Descent Method – A Novel Approach
Dileep M V1, Vishnu G Nair2, Prahalad K R3,Surekha Kamath4
1
(Department of Aeronautical Engineering, MIT, Manipal,)
2
(Department of Instrumentation and Control Engineering, MIT, Manipal.)
3
(Department of Instrumentation and Control Engineering, MIT, Manipal.)
4
(Department of Instrumentation and Control Engineering, MIT, Manipal

ABSTRACT
Trajectory optimization of a generic launch vehicle is considered in this paper. The direct application of a
nonlinear programming method is used in recent literature, which transforms the original problem into a
nonlinear optimization problem.To study the rocket motion under the influence of gravitational field, 2-D
simulator is developed. The rocket motion is analyzed for a gravity turn trajectory. The objective is to ensure
desired terminal conditions as well as minimum control effort in the low dynamic pressure region. Design of
optimal trajectory for a single stage rocket is a two point boundary problem. Trajectory is designed for a single
stage liquid rocket.Trajectory is computed for a given initial and final condition using equations of motion of
rocket in 2-D plane. Hamiltonian is formulated for the given constraints. The non-linear equations are solved
using steepest descent method.It is assumed that the launch vehicle is not experiencing any perturbations. Results
are compared for Runge-kutta and Euler‟s integration methods,which clearly brings out the potential advantages
of the proposed approach.
Keywords – Trajectory optimization, Steepest descent method, Euler‟s method, Runge-Kutta method
.

I. INTRODUCTION
The subject of optimization of a continuous relation between these two approaches, and the
dynamical system has a long and interesting history. development of these two approaches. In it, Betts
The first example is the Brachistochrone problem points out some of the disadvantages with indirect
posed by Galileo, later by Bernoulli and solved by methods which are mentioned below First, it is
Newton in 1696. The problem can be simply stated necessary to derive analytic expressions for the
as the determination of a trajectory that satisfies necessary conditions, and for complicated nonlinear
specified initial and terminal conditions, i.e., satisfies dynamics this can become quite daunting. Second,
the system governing equations, while minimizing the region of convergence for a root finding
some quantity of importance. We use the term algorithm may be surprisingly small, especially
trajectory here as representing a path or time history when it is necessary to guess values for the adjoint
of the system state variables. Our experience is variables that may not have an obvious physical
primarily in the field of spacecraft and aircraft interpretation. Third, for problems with path
trajectory optimization so that the trajectories are inequalities it is necessary to guess the sequence of
literal [1].There are many techniques for numerically constrained and unconstrained subarcs before
solving trajectory optimization problems. Generally iteration can begin[8]. One of the standard
these techniques are classified as either indirect or procedures for optimizing non-linear system is the
direct. Indirect methods are characterized by Gradient or Steepest-decent technique. Reference [3]
explicitly solving the optimality conditions stated in discusses implementation of this method to a launch
terms of the adjoint differential equations, the vehicle carrying a hypersonic vehicle as payload. In
Pontryagin‟s maximum principle, and associated paper[7] the BDH method that is one of the direct
boundary conditions. Using the calculus of collocation methods is used. In the direct collocation
variations, the optimal control necessary conditions method, not only the control variables but also the
can be derived by setting the first variation of the state variables are discretised. The BDH is using
Hamiltonian function zero. The indirect approach linear interpolation for this discretization.An
usually requires the solution of a nonlinear optimization algorithm Combination of Gauss
multipoint boundary value problem. There is a Pseudospectral Method and Genetic Algorithm is
comprehensive survey paper by Betts [8] that presented to solve the optimal finite-thrust trajectory
describes direct and indirect optimization, the with an input constraint in the paper[9]. The

www.ijera.com 116 | P a g e
Dileep M V et al. Int. Journal of Engineering Research and Applications www.ijera.com
ISSN : 2248-9622, Vol. 4, Issue 1( Version 1), January 2014, pp.116-121

simulation results indicate that the GPM-GA for many applications, and the
optimization algorithm has high accuracy, and the phases are sequential, that is
error with results solved by indirect method is very
small. In paper[10], two different approaches are
proposed for simultaneous optimization of staging However, neither of those assumptions is required.
and trajectory of multistage launch vehicles. In the Within phase the dynamics of the system are
first approach, the problems of staging optimization described by a set of dynamic variables.
and trajectory optimization are solved separately. In
the second approach, the optimal staging and
trajectory are achieved during trajectory optimization Made up of the state variables and the
in an integrated problem. Both approaches can lead control variables, respectively. In addition, the
to very similar solutions in spite of their differences
dynamics may incorporate the parameters
in staging formulation. Integrated approach can lead
to better results because of simultaneous which are not dependent on . For clarity the phase-
consideration of objective functions and effective dependent is dropped notation from the remaining
constraints of two optimization problems of staging discussion in thissection. However, it is important to
and trajectory. In paper[11] analysis of Euler remember that many complex problem descriptions
approximation to a state constrained control problem require different dynamics and/or constraints, and a
is carried out. it shows that if the active constraints phase-dependent formulation accommodates this
satisfy an independence condition and the requirement.
Lagrangian satisfies a coercively condition, then Typically the dynamics of the system are
locally there exists a solution to the Euler defined by a set of ordinary differential equations
discretization, and the error is bounded by a constant written in explicit form, which are referred to as the
times the mesh size. The paper[12] analyze second- state or system equations,
order Runge-Kutta approximations to a nonlinear
optimal control problem with control constraints. If Where is the dimension state vector.
the optimal control has a derivative of bounded
Initial conditions at time are defined by,
variation and a coercively condition holds, and it
shows that for a special class of Runge-Kutta
schemes, the error in the discrete approximating Where,
control is O(h2), where h is the mesh spacing.In this
paper the trajectory optimization problem solved by And terminal conditions at the final time are
steepest descent method, which is the type of indirect defined by,
gradient method. In reference[6], the control variable
is randomly chosen which will affect the accuracy to
a great extent. This problem is solved in our paper, Where,
by carefully choosing the initial control variable by
an new approach which is described in the following In addition, the solution must satisfy algebraic path
sections. To the author‟s best knowledge, the tuning constraints of the form,
of weighting is not done in any of the available
literature. In this the papers the weighting factor is Where is a vector of size , as well as simple
selected by proper tuning, and the proposed method
is explained in the following sections. Also for the bounds on the state variables.
quick convergence of the objective function a
multiplication factor is introduced. This will And control variables,
drastically increase the performance of the proposed
system. The prescribed optimization technique is Note that an equality constraint can be imposed if the
implemented using Euler‟s and Runge-kutta[2] upper and lower bounds are equal,
integration methods and the performance is verified.
for some k.
1.1. Trajectory optimization problem Finally, it may be convenient to evaluate expressions
A general problem statement for finite- of the form,
thrust trajectory optimization can be stated as
follows [7]. A trajectory optimization or optimal
control problem[14]can be formulated as a collection
of phases. In general, the independent variable Which involve the quadrature functions
(time) for phase is defined in the region collectively we refer to those functions evaluated
during the phase, namely,

www.ijera.com 117 | P a g e
Dileep M V et al. Int. Journal of Engineering Research and Applications www.ijera.com
ISSN : 2248-9622, Vol. 4, Issue 1( Version 1), January 2014, pp.116-121

In opimal control the is integrating to get

final values.

As the vector of continuous functions, similarly

functions evaluated at a specific point, such as the
boundary conditions and
are referred to as point
Where is the integration step
functions. size. are the initial values of .
The basic optimal control problem is to
Similarly are the initial values of
determine the k dimensional control vectors
and parameters to minimize the .For optimal solutions, The optimality
performance index. conditionis given by,

Notice that the objective function may depend on

quantities equations of motion with non-rotating 1.3 Runge-Kutta method
spherical earth[5] is given by, Runge-Kutta methods[7] are very popular
because of their efficiency; and are used for in most
solving problem numerically. They are single-step
methods, as the Euler methods[7].
There are many ways to evaluate the right-
hand side that all agree to first order, but that
have different coefficients of higher-order error
terms. Adding up the right combination of these, it
Here neglecting and since can eliminate the error terms order by order. That is
the rocket propagating in above vacuum, therefore the basic idea of the Runge-Kutta method.
the equations becomes

Trajectory optimization problem is The fourth-order Runge-Kutta method

formulated as optimal control problem. The requires four evaluations of the right-hand side.
nonlinear problem has to solve using numerical
method. Steepest decent method is used to solve the Methodology
problem. In this section mathematical model of
rocket trajectory is discussed first. To study the
1.2 Euler’s method rocket motion under the influence of gravitational
Euler method is a first-order numerical field, 2- D simulator is developed. The rocket motion
procedure for solving ordinary differential equations is analyzed for a gravity turn trajectory. It is solved
(ODEs) with a given initial value. The Euler method by Newton-Raphson method. Using gravity turn
is a first-order method, which means that the local trajectory target is achieved by varying initial
error (error per step) is proportional to the square of conditions to reach the target. From this we will get
the step size, and the global error (error at a given the characteristics of the required trajectory. Indirect
time) is proportional to the step size. It also suffers methods require initial guesses for the control as
from stability problems.Euler‟s method can be well as adjoint variables.
implemented both in simulator program and the For finding the feasible steering profile,
opimal control problem. In simulator program the acceleration is assumed to be a linearly increasing
integration of happens. That is given as quantity. So rate of change of velocity is
follows. approximated as linear function.This linear function
is selected by polynomial approximation[16] that is
by changing the functions randomly to fit the curve
properly.

www.ijera.com 118 | P a g e
Dileep M V et al. Int. Journal of Engineering Research and Applications www.ijera.com
ISSN : 2248-9622, Vol. 4, Issue 1( Version 1), January 2014, pp.116-121

variables are calculated by manual calculations using

the terminal conditions of rocket.
For optimal solutions, the optimality
condition is given by,

The weighting factor should be selected as

follows:-
The vehicle acceleration is a linear function
of time as given in the equation the constants
are obtained by varying initial and final steering
angle value. Thus tuning of constants is
done. The initial and final is thus obtained to be Thus weighting factor, thus obtained is
⁡and respectively from boundary applied to the control variable ,to update the initial
conditions. alpha profile.
Therefore by solving the equation
using the initial and final values the appropriate
values for can be found out; Where previous stage of
Equation is the gradiant function
A simulator program for rocket trajectory is which optimizes the control variable.
then developed by putting the initial guess for In the above equation,multiplication
control variable...The target can be successfully factor, has to be addedto converge objective
achieved only with t continuous monitoring of the function value. The value has to multiplied with the
position, velocity and flight path angle of the rocket. weighting factor, .
The state and control vectors are defined as,

1.4 Summary
The steps given below are adopted from [6]
1. Start with an initial guess of control variable
where
In this a trajectory optimization problem of
2. Propagate the states from to using with
a single stage launch vehicle is considered. The
objective here is to generate the guidance command initial conditions (Forward Integration of the
history such that the following SystemDynamics).
concerns are takencare of, 3. Obtain by using the boundary conditions
(terminal boundary conditions).
(1) At the final time , the specified terminal 4. Propagate the costate vector from to using
constraintsmust meet accurately. The terminal step (iii) values as the initial values (Backward
constraints include constraints on altitude, velocity integration ofCostate Equations).
and flight path angle (which is the angle made by the 5. Calculate the gradient from to
velocity vector with respect to the local horizontal).
6. Calculate the control update
(2) The system should demand minimum guidance where is the
command, which can be ensured by formulating a learningrate.
„minimum time‟ problem. 7. Repeat from step (2) to step (6) until optimality
To achieve the above objectives, the conditions are meet within a specified tolerance.
following cost function is selected, which consists of
terminal penalty terms and a dynamic control
minimization term.

An optimal control problem is developed for rocket

trajectory and control parameter is optimized by
steepest descent method. The method of steepest
descent will generally converge linearly to the
solution [14, 15]. The initial values for ad-joint

www.ijera.com 119 | P a g e
Dileep M V et al. Int. Journal of Engineering Research and Applications www.ijera.com
ISSN : 2248-9622, Vol. 4, Issue 1( Version 1), January 2014, pp.116-121

Table.1 EULER‟S METHOD Table.2RUNGE-KUTTA METHOD

Parametr Intial Final Target Error
Parame Intialva Final Target Error es valu Value Value Obtaine
tres lue Value Value Obtai e d
ned Height
Velocity
Height
Velocit
Flightpat
y m/s h Angle
Flightp
ath Final objective function value=
Angle
Total number of iterations =116
Final objective function value=
Total number of iterations=81

I. Result analysis
The results obtained by Euler‟s method and
Runge-Kutta methods are given in table 1 and
2.Runge-kutta and Euler‟s methods of integration are
performed. In Runge-kutta method of integration, the
step size is different for each integration step. But in
Euler‟s method the steps sizes are equal. Due to this
Runge-Kutta method of integration is comparatively
accurate.
Figure(1) shows the variations in control
variable with respect to time. Since the initial value
of the control variable is a guess, it is not optimal,
hence rate of change of initial control variable is Fig.2Objective function variations in Runge-Kutta
high and therefore it requires more fuel to complete and Euler‟s method
the trajectory.

Fig.1: Initial and final control variable array

Figure(2) shows the variation of objective

function in two different methods. From this graph
we can find that the convergence characteristics of
two methods are same but Runge-Kutta method
needs more iteration to find its minima then also it is
efficient by calculating the efficiency. Figure(3)
shows the variation in flight path angle with respect
to time we can see that the performance in achieving Fig.3 Final value of velocity difference in Runge-
target in both methods are almost same. kutta method and Euler‟s method

www.ijera.com 120 | P a g e
Dileep M V et al. Int. Journal of Engineering Research and Applications www.ijera.com
ISSN : 2248-9622, Vol. 4, Issue 1( Version 1), January 2014, pp.116-121

ComputerScience, Univ. of California,

Berkeley, Berkeley, CA, 1996.
[3] C. V. Sirisha and M. Manickavasagam, "A
Specific Energy Based Guidance Algorithm
for the Hypersonic Launch Vehicle",
International Conference on Avionics
Systems–2008Conference, RCI, Hyderabad.
[6] Lukkana Varaprasad and Radhakant Padhi
“Ascent Phase Trajectory Optimization Of
A Generic Launch Vehicle” XXXII
NATIONAL SYSTEMS CONFERENCE,
IIT-Roorkee. NSC 2008
[7] Nobuhiro YOKOYAMA, Shinji SUZUKI
Fig.4Final position achieved by rocket using Runge- “Flight Trajectory Optimization Using
kutta method and Euler‟s method Genetic Algorithm Combined with Gradient
Method” e-journal: Information
Figure(4) and (5) indicates the velocity Technology for Economics &Management
profile and position with respect to time and by (ITEM) Volume 1, No 1, 2001 Paper 10
analyzing both graphs we can estimate the efficiency [9] QiboPeng (2012). Finite-Thrust Trajectory
of Runge-Kutta method. Optimization Using a Combination of
Gauss Pseudospectral Method and Genetic
II. Conclusion Algorithm, Genetic Algorithms in
Rocket trajectories are optimized to achieve Applications, Dr.RustemPopa (Ed.), ISBN:
the target, by either minimum time, control forces or 978-953-51-0400-1, InTech, DOI:
fuel. To study the rocket motion under the influence 10.5772/36857.
of gravitational field, 2-D simulator is developed. [10] Reza Jamilnia∗ , AbolghasemNaghash
The rocket motion is analyzed for a gravity turn “Simultaneous optimization of staging and
trajectory and target is achieved by varying initial trajectory of launch vehicles using two
conditions to reach the target. Design of optimal different approaches” Aerospace Science
trajectory for a single stage rocket is a two point and Technology 23 (2012) 85–92
boundary problem. Trajectory is designed for a [11] A. L. Dontchev and William W. Hager “
single stage liquid rocket, for given initial and final The Euler Approximation in State
conditions using equations of motion of rocket in 2- Constrained Optimal Control “)
D plane. The trajectory optimized for minimum time Mathematics of Computation, Vol. 70, No.
of flight using the pitch angle as control 233 (Jan., 2001), pp. 173-203
variable.Hamiltonian is formulated for the given [12] A. L. Dontchev, William W. Hager and
constraints. The non-linear equation is solved using Vladimir M. Veliov “ Second-Order Runge-
steepest descent method.Results are compared for Kutta Approximations in Control
Runge-kutta integration and Euler integration Constrained Optimal Control” SIAM
method.Runge-kutta integration gives high accuracy. Journal on Numerical Analysis, Vol. 38,
By analyzing the results the converging No. 1 (2001), pp. 202-226
characteristics of two methods are almost same but [13] Ulrich Walter“ Astronautics” © 2008
Runge-Kutta method needsmore time to settle down WILEY-VCH Verlag GmbH &Co. KgaA,
to the minimum value. Weinheim ISBN:978-3-527-40685-2
[14] Bryson, A.E., Jr., Ho, Y.-C., “Applied
Reference Optimal Control”, Revised Printing,
[1] Bruce A. Conway“A Survey of Methods Hemisphere Publishing Corporation, 1975
Available for the Numerical Optimization [15] Desineni Rama Naidu “Optimal control
of Continuous Dynamic Systems” J Optim Systems” Boca Raton London New York
Theory Appl (2012) 152:271–306 © Washington, D.C.© 2003 by CRC Press
Springer Science+Business Media, LLC LLC ISBN 0-8493-0892-5
2011 [16] Gordon K. Smyth “Polynomial
[2] Schwartz, A. L., “Theory and Approximation” John Wiley & Sons, Ltd,
Implementation of Numerical Chichester, 1998 ,ISBN 0471 97576
MethodsBased on Runge–Kutta Integration
for Solving Optimal ControlProblems,”
Ph.D. Thesis, Electrical Engineering and

www.ijera.com 121 | P a g e