AIAA Guidance, Navigation, and Control Conference and Exhibit AIAA 2005-6287

15 - 18 August 2005, San Francisco, California

Powered Descent Guidance Methods For The Moon

and Mars
Ronald R. Sostaric ∗ and Jeremy R. Rea†
NASA Johnson Space Center, Houston, TX, 77058, USA

This study consists of a performance assessment of the various types of powered descent
guidance methods previously used for missions to the Moon and Mars. The strengths and
weaknesses of the various methods are discussed and simulation results are shown. Past
methods have been used as a starting point for newer guidance schemes and one particular
guidance scheme based on the Apollo Lunar Module Landing Guidance is discussed in
greater detail. The guidance results are compared to optimal trajectory results, and a
general outline for future powered descent trajectory guidance is suggested.

Nomenclature
a Acceleration, m/s2
acmd Commanded acceleration, m/s2
aT Total acceleration, m/s2
c0 , c1 , c2 Unknown coefficients
d1 , d 2 , d 3 , d 4 Unknown coefficients
~
C Constraint vector(s)
DN Legendre psuedospectral differentiation matrix
g Acceleration due to gravity, m/s2
Isp Specific impulse, sec
LN −1 (t) Legendre polynomial of order N − 1
L0N −1 (t) First derivative of the Legendre polynomial
L/D Lift-to-drag ratio
M0 Initial mass, kg
Mf Final mass, kg
Mp Propellant mass, kg
Rx , R y , R z , r Position, m
r0 Initial position, m
rt Target position, m
t Time, sec
t0 Initial time, sec
tgo Time to go, sec
T /W Thrust-to-Weight ratio
Vx , V y , V z , v Velocity, m/s
v0 Initial velocity, m/s
vt Target velocity, m/s
wk Weighting for Gaussian quadrature
β Thrust heading angle in the horizontal plane measured CCW from the X-axis, deg
∆V Delta-V, m/s
θ Thrust angle from horizontal, deg

∗ Aerospace Engineer, Flight Mechanics and Trajectory Design Branch, NASA Johnson Space Center, AIAA Senior Member.
† Aerospace Engineer, Flight Mechanics and Trajectory Design Branch, NASA Johnson Space Center, AIAA Member.

1 of 20

This material is declared a work of the U.S. Government and American Institute
is not subject of protection
to copyright Aeronauticsin theand Astronautics
United States.
 Introduction
Successfully landing a spacecraft on a planetary surface requires some method to slow the descent rate
of the vehicle and lessen the impact, such as parachutes, airbags, engine firings, or some combination. In
particular, the use of rocket engine thrust to steer the vehicle to a soft landing allows for greater control
during the descent and the ability to avoid hazards, such as rocks or craters. A trajectory guidance algorithm
is also necessary to provide the thrust and attitude commands to deliver the vehicle to the surface safely.
This algorithm must provide a trajectory that decreases the descent rate to an acceptable value while limiting
the amount of propellant, and may also be required to achieve a certain landing accuracy and/or provide
for hazard avoidance capability. The trajectory must also meet constraints on acceleration, attitude, and
attitude rate.

I. Powered Descent at the Moon

A. Background
1. Powered Descent Envelope
The powered descent is the final phase of flight leading to touchdown, and for that reason it is also sometimes
referred to as “terminal” descent. Figure 1 shows an example Moon mission profile from the First Lunar
Outpost study.1 The sequence begins with a de-orbit burn, followed by a coast phase. Following the coast
phase is the powered descent, which begins near periapse (or perilune) and terminates at touchdown.

Figure 1. Example Moon Descent and Landing Profile

The powered descent itself may be broken into phases. Figure 2 shows an example of powered descent
at the Moon. The longest phase is the initial braking phase. The throttle was designed to a constant
92% during this phase, to provide margin for potential throttle increases due to navigation updates. At 84
sec remaining, the braking phase ends and the pitch-up phase begins. During the pitch-up phase, a linear
throttle down and linear pitch profile is commanded. At 24 sec remaining, the vehicle is directly above the
landing site at an altitude of 100m. It then descends vertically at 33% throttle (1.2 Moon G’s) to touchdown.

2 of 20

American Institute of Aeronautics and Astronautics

 Figure 2. Example Moon Landing Powered Descent Trajectory

2. Apollo Lunar Descent Guidance

The Apollo Lunar Descent Guidance was designed to steer the vehicle from about 15 km altitude in an
elliptical lunar orbit to the lunar surface. The guidance issued thrust and attitude commands given the
vehicle state from the navigation routines.
The descent was broken into three separate phases (shown in Figure 32 ):
1. Braking Phase (P63)
2. Approach Phase(P64)
3. Terminal Descent Phase (P66)

Figure 3. Apollo Lunar Descent Guidance Trajectory

The braking and approach phase each used the same guidance law, but the targets were different for each
phase. The braking phase began at a slant range to target of 492 km. The transfer to the start of approach
took about 514 sec and was near optimal. The target for the braking phase was a point below the surface
and further downrange than the start of the approach phase.
The approach phase began with initial conditions of approximately 2.2 km altitude, 7.5 km ground range,

3 of 20

American Institute of Aeronautics and Astronautics

-44 m/s vertical velocity, and 129 m/s forward velocity. A typical approach phase took 146 sec. During this
time the landing site could be manually re-designated. A target point was chosen that was not expected to
be achieved, but would steer the vehicle to the desired point near to directly above the landing site with the
desired conditions for P66 Terminal Phase start.
The algorithm designers were concerned with delay times due to computer processing, and derived the
guidance law with a parameter to account for the delay. With current computing power, processing delay
times are not a concern for this type of guidance algorithm. For a computational delay time of zero, the
guidance equation (used for P63 and P64) reduced to:2

12(rt − r) 6(vt + v)
acmd = + + at (1)
t2 t
The terminal descent phase began at 30-m altitude and 11-m ground range from the target. During
this phase, the vertical and horizontal velocity were controlled separately. The vertical guidance algorithm
controlled the altitude rate by adjusting the throttle. The horizontal guidance algorithm nulled the horizontal
velocity by making small changes to the thrust angle. It assumed the vertical component of thrust was
constant. There was a 20◦ limit to the thrust angle during this phase.
The guidance operated in a surface-fixed coordinate frame with X-axis vertical, Z-axis in-plane and
forward, and Y-axis completing the right hand coordinate system. Re-designation of the landing target
(performed during the approach phase, if necessary) changed the origin of the coordinate system.

B. Lunar Powered Descent Guidance Derivation and Simulation

The following guidance method was developed independently of the Apollo LM Guidance. However, the
final result is similar to the explicit form of the Apollo LM Guidance law shown in equation (1).

1. Braking Phase
A powered descent trajectory at Moon typically begins with a long braking phase, designed to slow the
vehicle velocity from orbital speeds. It can be shown, using optimal control theory, that the optimal steering
law for this phase is the bilinear tangent law (if range is constrained):3
d1 t + d2
tan θ = (2)
d3 t + d4
where θ is the thrust angle, and d1 , d2 , d3 , and d4 are unknown coefficients.
If range is not constrained, the steering law reduces to a single linear function, commonly known as the
linear tangent law.

tan θ = d1 t + d2 (3)
It is not possible, in general, to find a closed-form solution for the coefficients when solving this optimal
control problem. Some methods have been developed to approximate the optimal solution.4
A simple method for calculating the trajectory is to develop an analytic profile that closely approximates
the optimal solution. This can be accomplished by restricting the acceleration profile to a linear function in
each axis.

a = c1 + c2 t (4)
The acceleration profile is integrated to develop the analytic equations for velocity and position.
Z t
1
v= a = c1 (t − t0 ) + c2 (t − t0 )2 + v0 (5)
t0 2
Z t Z tZ t
1 1
r= v= a = c1 (t − t0 )2 + c2 (t − t0 )3 + v0 (t − t0 ) + r0 (6)
t0 t0 t0 2 6
The equations can be rewritten as follows,

tgo = t − t0 (7)

4 of 20

American Institute of Aeronautics and Astronautics

 1
vt − v0 = c1 tgo + c2 tgo2 (8)
2
1 1
rt − r0 − v0 tgo = c1 tgo2 + c2 tgo3 (9)
2 6
In matrix form,
" # " #" #
1 2
vt − v0 tgo 2 tgo c1
= 1 2 1 3
rt − r0 − v0 tgo 2 tgo 6 tgo c2

Solve for the coefficients, c1 and c2, by inverting the tgo matrix.
" # " #" #
2 6
c1 − tgo tgo2 vt − v0
= 6 12
c2 tgo2 − tgo 3 rt − r0 − v0 tgo

In simplified form,

−2(vt + 2v0 ) 6(rt − r0 )

c1 = + (10)
tgo tgo2
6(vt + v0 ) 12(rt − r0 )
c2 = − (11)
tgo2 tgo3
The acceleration command is then given by

acmd = c1 + c2 t (12)
The length of burn, tgo, and position and velocity targets, r0 and v0 , respectively, must be provided.
These targets are found using a combination of the “optimal” routines and trial and error from repeated
simulation (Monte Carlo runs).
The following plots in figure 4 are an example nominal descent trajectory at the Moon using the scheme
described above for the braking phase. Following the braking phase (which ends at t=3725.1 sec), a quadratic
acceleration scheme is used for the approach phase. The third and final phase is a vertical descent to the
landing site at a constant acceleration. These phases are described further in the following sections.
The braking phase performance is dependent on the maximum engine thrust level. A series of simulations
were run varying the maximum thrust for each run. The performance variation is shown in Figure 5 below.
The minimum delta-V of about 1608 m/s occurs around an initial T/W = 0.55. The braking phase makes
up the majority of the required delta-V for powered descent. More discussion of performance follows later
in the “performance” section.

2. Quadratic Approach Phase

Following the braking phase, it is desirable to have a 2nd guidance phase which commands a pitch-over from
the mostly horizontal attitude commanded during the braking phase to a vertical attitude in preparation
for landing. During this time the engine will also be throttled down as the vehicle nears the surface. The
throttle-down minimizes the disturbance of surface dust and allows for a less dynamic landing, as well
protects for dispersions. This phase is called the “approach” phase, since the target will be directly above
the landing site at some specified altitude with null horizontal velocity.
A simple method of calculating the thrust command for this phase is to restrict the acceleration profile
to a quadratic equation of time.

a = c0 + c1 t + c2 t2 (13)
The acceleration profile is integrated to develop the analytic equations for velocity and position.
Z t
1 1
v= a = c0 (t − t0 ) + c1 (t − t0 )2 + c2 (t − t0 )3 + v0 (14)
t0 2 3

5 of 20

American Institute of Aeronautics and Astronautics

(a) (b)

(c) (d)

Figure 4. Simulated Guided Lunar Descent Trajectory

Figure 5. Braking Phase Performance Variation with T/W

6 of 20

American Institute of Aeronautics and Astronautics

 Z t Z tZ t
1 1 1
r= v= a= c0 (t − t0 )2 + c1 (t − t0 )3 + c2 (t − t0 )4 + v0 (t − t0 ) + r0 (15)
t0 t0 t0 2 6 12
The equations can be rewritten as follows,

tgo = t − t0 (16)

at = c0 + c1 tgo + c2 tgo2 (17)

1 1
vt − v0 = c0 tgo + c1 tgo2 + c2 tgo3 (18)
2 6
1 1 1
rt − r0 − v0 tgo = c0 tgo2 + c1 tgo3 + c2 tgo4 (19)
2 6 12
In matrix form,
    
at 1 tgo tgo2 c0
1 2 1 3 
vt − v0 = tgo 2 tgo 3 tgo   c1 
   

1 2 1 3 1 4
rt − r0 − v0 tgo 2 tgo 6 tgo 12 tgo c2

Solve for the coefficients, c0 , c1 , and c2 , by inverting the tgo matrix.

6 12
    
c0 1 − tgo tgo2 at
6 30 48
 c1  =  − tgo − tgo vt − v0
    
tgo2 3  
c2 6 24 36 rt − r0 − v0 tgo
tgo2 − tgo3 tgo4

Reducing into equation form,

(vt + v0 ) (rt − r0 )
c0 = at − 6 + 12 (20)
tgo tgo2
at (5vt + 3v0 ) (rt − r0 )
c1 = −6 +6 − 48 (21)
tgo tgo2 tgo3
at (2vt + v0 ) (rt − r0 )
c2 = 6 2
− 12 3
+ 36 (22)
tgo tgo tgo4
The length of burn, tgo, and position, velocity, and acceleration targets, rt , vt , and at , respectively, must
be provided. Typically, an altitude directly above the landing site is targeted, with zero horizontal velocity
and a vertical orientation (zero horizontal acceleration). The vertical acceleration target is chosen to be
equal to 1.2g, and is held constant through the 3rd and final “vertical” phase. The vertical velocity target
is then determined from the vertical position and acceleration targets.
The above coefficients are computed at each guidance call, in each axis. The initial state for the two-point
value boundary value problem is the current vehicle state (as estimated by the navigation algorithm). The
targets for the final state typically do not change during the phase, unless a target re-designation is desired.
The burn length to go, tgo, will decrease as the vehicle approaches the target state.
As tgo becomes very small the coefficient terms will tend toward infinity. There are typically two different
methods employed to prevent this from happening. One method is to use a target state that forces the vehicle
to follow the desired trajectory path without the intent of actually reaching that target. The guidance phase
is then terminated prior to tgo approaching zero. The second method is to utilize the fine count, where the
coefficients are not recalculated after tgo goes below a pre-selected value (typically 1-2 sec).
The acceleration command is then given by

acmd = c0 + c1 t + c2 t2 (23)
At each guidance call, the coefficients are recalculated and t is set to zero. For t=0, the acceleration
command is equal to c0 .

7 of 20

American Institute of Aeronautics and Astronautics

 (vt + v0 ) (rt − r0 )
acmd = c0 = at − 6 + 12 (24)
tgo tgo2
This is the same result as the simplified version of the explicit equation used by the Apollo Lunar Descent
Guidance in equation (1).
During the fine count, the coefficients are not recalculated, and t is not reset at each guidance call.

3. Vertical Phase
The vertical phase is the final guidance phase, ending at touchdown. During this phase, a vertical orientation
is nominally commanded at a constant vertical acceleration. Current baseline lunar trajectories use a 100m
altitude for the start of the phase and a 1.2g vertical acceleration throughout. This altitude allows time to
divert to a different landing site in case of hazards.
If a secondary landing site is required, the vehicle still has the time before touchdown to achieve a new
landing site with the necessary touchdown conditions (vertical orientation, low touchdown velocity). The
nominal vertical phase begins at an altitude of 100 m and vertical velocity of -8.2 m/s (towards the surface).
A constant thrust command of 0.3 brings the vehicle to the surface with a velocity of slightly greater than
-1.0 m/s. The nominal vertical phase takes 20.9 sec and costs 40.9 m/s of delta-V.
An attempt was made to quantify the amount of time and delta-V that is necessary to perform a desired
amount of divert. A series of cases with increasing divert was simulated. The pitch profile for these cases
involved three segments of constant pitch rate. The first is a pitch over to a maximum of 25 deg, to start
the vehicle in the desired direction of the divert. The second is a constant rate pitch over in the opposite
direction (to a max of -25 deg), in order to null the vehicle horizontal velocity prior to touchdown. The final
segment is a pitch over to 0 deg. The pitch rate for these cases was limited to 5 deg/sec. Also, the throttle
rate was limited to 0.1 sec-1 . The minimum throttle was set to 25%, with a max of 100%.
Table 1 shows the amount of extra time and delta-V that is required for the divert maneuver. Up to
50 m of divert can be performed at essentially no extra cost. The engine throttle is increased slightly to
start the vehicle heading in the direction of the divert. It is then decreased slightly as the vehicle travels the
desired divert distance. Then it is increased slightly to take the velocity back out. However, this does not
impact the delta-V much since the average throttle is close to 0.3. Also note that the average descent rate
is close to the nominal. Since most of the delta-V for this phase comes from gravity losses, the delta-V is
very closely related to the length of the phase.

Table 1. Divert Deltas From Nominal

Divert(m) ∆Time (sec) ∆Delta-V(m/s)

25.0 -0.9 -0.9
50.0 -0.6 0.6
75.0 4.0 8.2
100.0 8.3 15.5
125.0 11.1 20.3
150.0 15.4 27.4

As divert distance is increased, the amount of extra time and delta-V is increased. It might be possible
to further optimize the profile for these cases with larger number of segments in the pitch profile. However,
this is the final phase right before touchdown, and every effort should be made to minimize the complexity
and number of maneuvers at this stage of the flight. Thus, the simplest possible method (three segments
only) was chosen.
Another method of performing the divert is selecting a new landing site earlier in the trajectory, during
the approach phase. If this method is preferred, it may be advantageous to lower the start altitude of the
vertical phase. During the Apollo missions, the landing site change was performed during the approach
phase. The vertical phase started at an altitude of 30m above the surface. The horizontal velocity at the
start of the vertical phase was not expected to be zero. No position constraint was enforced during this
phase, only velocity constraints. It was called the Terminal Descent Phase, since a strictly vertical descent
was not commanded.

8 of 20

American Institute of Aeronautics and Astronautics

4. Performance
In order to determine the total performance for the Lunar Descent, the simulation was run from 100km
circular orbit to the surface. A de-orbit burn was targeted for a 18.5 km perilune. The start of the braking
phase burn was determined using NPSOL, an optimization code created at Stanford. The braking phase was
performed using PEG, a variation of the Shuttle ascent guidance code, which provides a near optimal braking
phase. This phase uses a max throttle burn. The PEG targets were also determined using the optimizer.
Following the braking phase is a 60 sec pitch-over/throttle down phase. Both pitch over and throttle down
are commanded at a constant rate for this phase. The phase targets a location 100-m directly above the
landing site. The final vertical phase is a constant throttle descent, at a throttle setting determined using
the optimizer. The end of phase target is less than 2 m/s velocity at touchdown.
Figure 6 shows the delta-V required for the entire descent as it varies with initial thrust-to-weight ratio.
The minimum occurs at a initial T/W of 0.6. However, past experience has shown that overall system mass
is optimized at a lower T/W. Figure 6 shows that the required delta-V does not increase rapidly until the
T/W goes below 0.4. The amount of extra engine mass required to achieve T/W=0.6 tends to be greater
than the extra propellant mass needed to meet the delta-V requirement for T/W=0.4, and so the overall
optimal initial T/W is closer to 0.4.

Figure 6. Total Delta-V Required for Lunar Descent from Parking Orbit

As a comparison, the Apollo powered descent phase needed about 2055 m/s of delta-V.1 The primary
reasons for the difference are that Apollo:
• Used a lower T/W (initial T/W =
˜ 0.33)
• The total time of the burn was longer
• The engine could not throttle in the region between 65% and 92% throttle
• Had a surface viewing constraint

5. Divert Out-of-Plane
Depending on the sensor suite and type of navigation algorithms employed for a Lunar Descent mission, the
accuracy of the knowledge of the vehicle state may change quickly during flight. One navigation solution

9 of 20

American Institute of Aeronautics and Astronautics

being considered by engineers at the Johnson Space Center involves the use of a very accurate sensor with
limited range. This sensor would be activated relatively close to the surface. It was determined that a near-
worst case scenario would be to turn on the sensor only to discover that the vehicle was 2000 ft from the
desired landing location in the out-of-plane direction. Navigation engineers were interested in understanding
when this information must be available to guidance and control in order to still reach the target.
The trajectory shown in figure 4 was used as the nominal trajectory. The divert is commanded during
the approach phase. If the divert is commanded at the start of approach, figure 7 shows that cost in delta-V
is only a few m/s. As the time goes on and the vehicle nears the (soon to be realized as incorrect) target,
the cost increases.

Figure 7. Out-of-Plane Divert Delta-V Requirements

Figure 7 shows both the altitude and range on the same plot using two different y-axes. The two curves
are actually the same runs, one shows altitude and the other shows range.
Note that the nominal approach phase delta-V is about 230 m/s. The results show that if you wait until
1000 ft altitude, you can still make the new target, but it costs about 70 m/s extra.
Cases were run every 10 sec from the start of the approach phase (from 0-50 sec after approach phase
start). After about 20 sec, the delta-V starts to take off. For the 30 sec case and up, the total length of the
approach phase was increased in order for the vehicle to make the target. The recommendation was not to
wait much longer than 20 sec after the start of approach to make the decision for this kind of divert. The
range to target at that point is 2000 ft, the altitude is about 3500 ft, and the extra cost in delta-V is about
22 m/s.

C. Gravity Turn
Another type of powered descent method that is commonly used is gravity turn steering. Gravity turn
steering requires that the thrust vector be oriented opposite to the instantaneous velocity vector during the
descent.

10 of 20

American Institute of Aeronautics and Astronautics

 The trajectory described in figure 2 can be modified to illustrate gravity turn steering. Starting at the
same initial conditions depicted in figure 2, the gravity turn trajectory with a constant thrust was simulated,
and is shown as the blue trajectory in figure 8. Note that this trajectory holds a -180 deg angle of attack
throughout. For this case, it takes approximately 280 sec to travel 300 km downrange and descend 17.5 km
in altitude. Note in the flight path angle plot that the vehicle velocity vector is along the horizontal at the
start of the trajectory, but is still traveling mostly horizontal (at -15 deg) at touchdown. Also, note in the
altitude-velocity plot (row 3, column 1) that the relative velocity at touchdown is still greater than 500 m/s.
Clearly, this trajectory must be redesigned to something more reasonable. One simple solution is to start
the gravity turn earlier, at a higher altitude.

(a)

(b)

Figure 8. Redesigned Gravity Turn Trajectory

Figure 8 shows the redesigned gravity turn (green line) in comparison to the trajectory above. The
powered descent is initiated at a higher altitude (30.6 km). This would require starting the burn earlier
along the “coast arc” depicted in figure 1, at a slightly slower velocity. It also would cause some impact to

11 of 20

American Institute of Aeronautics and Astronautics

the Earth-Moon transfer trajectory, but that will be neglected for the purposes of this example. Notice that
the redesigned gravity turn achieves the small and nearly horizontal velocity required for safe touchdown.
The ideal delta-V for the gravity turn solution is 1716.5 m/s, compared to a delta-V of 1852.7 m/s for
the solution with the pitch up maneuver depicted in figure 2. Since the gravity turn steering is thrusting
opposite the velocity vector at all times, the velocity losses are minimized.
It is important to note that this “true” gravity turn method is only theoretical and could not be used as
a guidance algorithm, because it can only meet final state constraints by adjusting the initial state, i.e. when
you turn it on. However, modified versions of gravity turn guidance can and have been used successfully,
in particular for the Viking program. Still, these types of algorithms have limitations. They are good at
minimizing propellant usage but lack the landing precision and re-targeting capability of schemes similar to
the Apollo LM Guidance.

II. Powered Descent at Mars

Powered descent to the Martian surface presents a separate set of challenges than the Lunar descent. Due
to the presence of an atmosphere, the powered descent at Mars is preceded by an atmospheric entry phase
and a parachute phase, which slow the vehicle and reduce the velocity at the start of powered descent. At the
Moon, the velocity at the start of powered descent is primarily horizontal. However, at Mars, the parachute
removes most of the horizontal velocity, so the initial velocity is much more vertical. In order to minimize
propellant, the engine should be ignited as late as possible and commanded at the highest setting. However,
if the intent of the mission is to achieve an accurate landing, the powered descent must solve the remaining
amount of position error, which can be substantial. Since earlier engine ignition is generally cheaper for
flying large horizontal (“divert”) distances, this becomes a trade off. Past work has shown that conditions
at the start of powered descent can vary widely, so this remains an interesting and challenging problem.

A. Background
Viking Powered Descent Guidance
The Viking landers used a powered descent guidance scheme based on a gravity turn solution.5 Due to wind
and atmospheric variation at Mars, the velocity at the end of the parachute phase was expected to be in
the range of 40-75 m/s vertical, and 0-80 m/s horizontal. The guidance scheme used two altitude-velocity
reference profiles to attempt to bound the solution space–one high speed contour and one low speed contour.
These were gravity turn solutions. Following this was a constant velocity phase that targeted 8 ft/s (2.44
m/s). This final phase began at an altitude of 16.8 m.
The engine was ignited at 1.5 km altitude, and after a 2 sec warm-up sequence the parachute was released.
A pitch maneuver to the correct attitude was performed, then the H-V reference contour was commanded
the rest of the way to the surface. Originally, the same H-V contour was followed for all cases, but it
was discovered during simulation work that for the lower velocity cases the guidance would sometimes get
behind the contour and end up impacting at a high velocity. This was the reason for the addition of the
low velocity contour. For initial velocities between the two contours, interpolation was used to determine
the H-V reference contour. For initial velocities outside of the range of the low and high speed contours, the
nearest contour was used.
Figure 9 shows a reconstructed version of the Viking H-V contours. The simulation was set up using a
gravity turn with constant thrust, starting at the initial conditions of the two contours. It was discovered
that the H-V contour could be approximated extremely well using a third order function.
Low speed contour curvefit:

H = 3.06 ∗ 10−7 + 1.53 ∗ 10−1 ∗ V + 8.795 ∗ 10−5 ∗ V 2 − 1.029 ∗ 103 ∗ V 3 (25)

High speed contour curvefit:

H = 3.23 ∗ 10−7 + 1.20 ∗ 10−4 ∗ V + 1.856 ∗ 10−4 ∗ V 2 − 2.479 ∗ 104 ∗ V 3 (26)

Figure 10 shows a few other plots from the reconstructed trajectory. The plot in the top left corner shows
the thrust command. The nominal thrust command for the high speed contour was about 0.64, and for the

12 of 20

American Institute of Aeronautics and Astronautics

 Figure 9. Viking H-V Contours

Figure 10. Viking Powered Descent Reconstructed Trajectory

13 of 20

American Institute of Aeronautics and Astronautics

low speed contour it was 0.53. Though designing to a higher thrust level would have saved some propellant
mass, the designers chose to save some throttle capability to protect for dispersions.
Both the high speed and low speed contours start at the same altitude, but the low speed contour requires
a lower thrust level, resulting in a longer burn than the higher thrust high speed contour.
The nominal required delta-V for this descent profile was 235 m/s.

B. Pinpoint Landing
Future missions to Mars may require an accurate landing on the surface. The atmospheric entry phase
can be guided accurately, given that the spacecraft has enough L/D during entry. However, limitations in
knowledge of the vehicle state for the current state-of-the-art navigation sensor suites and algorithms show
that substantial position error may still exist at the end of the entry phase7 .8 Also, wind drift on a parachute
may further increase position error from the desired landing site.
In order to achieve the desired landing position accuracy, the terminal descent system must be sized to
be able to accommodate significant position errors at engine ignition. Past studies have shown that a single
quadratic approach phase would use too much fuel to fly out large horizontal errors.
Results obtained using the optimization method described as follows suggest that a high-low-high thrust
profile is needed. The initial high thrust would be used to increase the vehicle horizontal velocity in the
desired direction. During the low thrust, or “coast” phase, the vehicle would travel a large horizontal distance
without much change in velocity. The results suggest that this phase would be commanded at the lowest
possible throttle setting. The final high thrust phase is used to the null the horizontal velocity that was
imparted to the vehicle during the first high thrust phase. A 4th and final phase, which has not been included
here, would be needed for the final descent to the landing site.

1. Optimization Code
A Legendre Pseudospectral Method for trajectory optimization, created by Mike Ross and Fariba Fahroo of
the Naval Postgraduate School, was applied to the terminal descent problem. It combines pseudospectral
methods with numerical nonlinear optimization. The time domain of the trajectory is discretized at a
special set of points called the Legendre-Gauss-Lobatto (LGL) points. The pseudospectral differentiation
matrix is then used to transform the equations of motion from nonlinear differential equations to nonlinear
algebraic equations. These equations are then posed in the form of a nonlinear programming problem,
and a numerical optimizer is used to solve the problem. The method has been used on many trajectory
optimization problems.9–12 A short description of the method follows:
The Legendre pseudospectral differentiation matrix operates on a discrete set of points called the Legendre-
Gauss-Lobatto (LGL) points. The matrix is based on Lagrange interpolation using the LGL points. A set
of N LGL points are defined as follows:9

t1 = −1
tj = roots of L0N −1 (t) (j=2..N-1)
tN = 1

where LN −1 (t) is the Legendre polynomial of order N −1 and L0N −1 (t) is the first derivative of the polynomial.
These points lie in the range of [-1,1].
It is necessary to map the real time domain to the LGL time domain by the following equation:13

2(τi − τo ) − (τf − τo )
tLGLi = (27)
τ f − τo

where tLGLi = the ith LGL time point; τi = the ith real time point; τo = the initial real time point; τf =
the final real time point.

14 of 20

American Institute of Aeronautics and Astronautics

 The elements of the N xN Legendre pseudospectral differentiation matrix (DN ) are defined by:13

N −1 (tLGLi )
 L
1
 i 6= j
L −1 (tLGLj ) (tLGLi −tLGLj )


 N





− N (N4−1)

i=j=1


Dij = (28)

N (N −1)

i=j=N


4








0 otherwise

It is also useful to have the weights for Gaussian quadrature:13

2 1
wk = k = 1, ..., N (29)
(N − 1) N L2N −1

The terminal descent problem is modeled using a flat planet assumption with a constant gravitational
acceleration in the vertical axis. Atmospheric effects are neglected. The equations of motion are:

Ṙx = Vx
Ṙy = Vy
Ṙz = Vz
V̇x = aT cos(θ) cos(β)
V̇y = aT cos(θ) sin(β)
V̇z = aT sin(θ) − g (30)

where the R0 s are the position components, the V 0 s are the velocity components, aT is the thrust acceleration,
θ is the thrust pitch angle from the horizontal plane, and β is the thrust heading angle in the horizontal
plane, measured counter-clockwise from the X-axis.
It is desired to find the optimal control history of the thrust acceleration and direction angles that
will minimize the required amount of propellant. The initial state and final state are specified, and the
time-of-flight is free. This is equivalent to minimizing the integral of the thrust acceleration.
Z tf
min J = aT dt (31)
to

The D-matrix is used to write a discretization of the differential equations as a system of algebraic
equations. After transforming the real time domain to the LGL time domain, the D-matrix is applied to
(30). The equations are then rearranged into sets of equality constraints. The result is:

~ ~x − (τf − τo ) h ~ i
C Ṙx = DN R Vx = 0
2
~ ~y − (τf − τo ) h ~ i
C Ṙy = DN R Vy = 0
2
~ ~z − (τf − τo ) ~ i
h
C Ṙz = DN R Vz = 0
2
(τf − τo ) h    i
~
C V̇x
~x −
= DN V ~aT cos θ~ cos β~ = 0
2
(τf − τo ) h    i
~
C V̇y
~y −
= DN V ~aT cos θ~ sin β ~ = 0
2
(τf − τo ) h   i
~
C V̇z
~z −
= DN V ~aT sin θ~ − g = 0 (32)
2
where each vector represents the value of the discretized states and controls at each LGL point. The C’s are
vectors of constraints that must be met at each LGL point.

15 of 20

American Institute of Aeronautics and Astronautics

 The objective function to minimize the integral of the thrust acceleration becomes:

(τf − τo ) T
min J = w
~ ~aT (33)
2
The optimization vector for the parameter optimization problem is:

~x
 
R
 ~ 
 Ry 
 ~ 
 Rz 
 
 V~ 
 x 
 ~ 
 V 
~xopt =  y 
 V~ 
 z 
 ~aT 
 
 ~ 
 θ 
 ~ 
 β 
τf

Lower and upper bounds are placed on the optimization vector so that the initial and final states are
fixed. Upper and lower limits are also placed on the thrust acceleration.

2. Results
Figure 11 shows a sample commanded acceleration profile for a stress case that flies 4 km in range over only
1.5 km for altitude change. Note the high-low-high thrust command as discussed earlier. The minimum and
maximum acceleration was limited in this case to 2.0-10.78 m/s2 . The maximum limit roughly corresponds
to T/W of 1.1 (in Earth G). The minimum limit is about 18.5% of the max thrust and is based on the Mars
Science Laboratory (MSL) configuration as of mid-2004. Figure 11 also shows the thrust angle variation
with time. Note a roughly linear relationship with time.

(a) Thrust Acceleration (b) Thrust Angle

Figure 11. Sample Commanded Profile Using Legendre Pseudospectral Method

Figure 12 shows the amount of delta-V required to fly various divert distances using this scheme. The
cost increases with divert distance. Figure 12(a) shows the delta-V needed for an initial descent rate of 50
m/s, and the 100 m/s data is shown in figure 12(b). The different curves correspond to an initial horizontal
velocity, shown in the legend. The horizontal velocities are in the opposite direction as that of the divert.
The start altitude is optimized for each case. The highest engine start altitude for the 50 m/s cases is 1.6
km, and the longest burn is 60 sec. For the 100 m/s cases, the highest start altitude is 2.7 km and the
longest burn is just under 60 sec.

16 of 20

American Institute of Aeronautics and Astronautics

 (a) 50 m/s Descent Rate

(b) 100 m/s Descent Rate

Figure 12. Divert Delta-V Requirment

17 of 20

American Institute of Aeronautics and Astronautics

 The kinds of initial conditions used for these runs are likely to be encountered for robotic missions in
which a parachute has been used to slow the majority of the horizontal velocity, and the remaining velocity
is mostly vertical. The results show the cost of providing the capability to solve the remaining position error
from the entry phase and wind drift. For example, a 1000-kg landed mass with a terminal parachute sized to
a 50 m/s descent rate would require 34 kg of propellant in the best case scenario (assuming Isp = 210 sec).
However, if the mission required capability for up to 4 km of divert and 30 m/s horizontal velocity in the
wrong direction, this would cost 189 kg of propellant. This is an increase of 155 kg of propellant, without
considering margin. This extra propellant must be carried from launch and would have a large impact on the
overall mission design. The cost of diverting during powered descent is significant and should be minimized
as much as possible by improvements in accuracy during the entry phase. Steerable parachutes may also
help increase accuracy by minimizing wind drift and should be considered for pinpoint missions.
To show the improvement this type of scheme has over previously considered schemes, the data is com-
pared to the old MSL guidance14 in figure 13. The initial conditions are a 50 m/s descent rate and zero
horizontal velocity. The blue curve represents the optimal solutions. The green curve is the ∆V required
using the old MSL guidance, which employs a restricted quadratic scheme. The ∆V increases as divert
distance increases. For the small divert cases, the MSL guidance and optimal solutions are close in magni-
tude. As divert increases, the optimal solution performs much better. The old MSL guidance solution was
designed with capability for small divert distances in mind (< 200 m), not for large, pinpoint class divert
requirements.

Figure 13. Comparison of old MSL Guidance to Optimal Solution

Conclusion
A combination of different methods for powered descent guidance should be pursued, depending on the
application. At the Moon, the braking phase must be at or near optimal due to the large amount of delta-V
necessary to slow the vehicle from orbit. Two-pt boundary value methods with proper targeting may prove
to be adequate, but more advanced methods should be pursued. For the succeeding phases, two-pt boundary
value methods are sufficient. At Mars, the potential range of initial conditions for the terminal phase can
vary widely depending on the mission. Parachutes are typically used for robotic missions but a chute system
may not be feasible for the more massive human missions, which would lead to higher initial velocities and
a greater delta-V requirement for the propulsive phase. In addition, with current state-of-the-art Entry,
Descent, and Landing, a large downrange change during the propulsive phase may be necessary and must be
accounted for in the vehicle design. Optimal methods must be pursued for these pinpoint cases but two-pt
boundary value methods should remain available for less stressful mission requirements or phases.

18 of 20

American Institute of Aeronautics and Astronautics

 Appendix
Propellant Mass
Throughout this paper, the main performance measure cited is delta-V, since it is a more general performance
index and can be applied to any size vehicle. However, a more tangible measure of the performance is the
propellant mass required to perform the trajectory maneuvers. The amount of mass change during the
maneuver is related to the delta-V via the rocket equation, the derivation of which is commonly found in
many aerospace texts.
M0
= e∆V /g0 Isp (34)
Mf
The propellant mass required, Mp , is the difference between the initial mass, M0 , and the final mass,
Mf .

Mp = M0 − Mf (35)
Figure 14 shows the amount of propellant mass required for a 1000-kg vehicle given a particular delta-V
and Isp . Each curve represents a different value of Isp , as shown in the legend.
As seen in equation (34), the result of the rocket equation is a mass ratio. Thus, the data in figure 14
is easily scalable to any size vehicle, by multiplying the amount of propellant read from the figure by the
vehicle mass and then dividing by the reference 1000-kg mass.

Figure 14. Propellant Mass

Acknowledgments
Thank you to Lee Bryant for serving as technical reviewer, and also for advice and guidance during this
study.
Thank you to the Aeroscience and Flight Mechanics Division for support in allowing us to perform and
present this work.

19 of 20

American Institute of Aeronautics and Astronautics

 References
1 Condon, Jerry; presentation entitled “First Lunar Outpost Powered Descent Design and Performance”, NASA JSC Systems

Engineering Division, Performance Analysis Branch; September 14, 1992.

2 Klumpp, Allan R., “Apollo Lunar-Descent Guidance”, Charles Stark Draper Laboratory, R-695, June 1971.
3 Bryson, A.E., Ho, Y., Applied Optimal Control, Ginn, Waltham, 1969.
4 Jaggers, Roland F., “Shuttle Powered Explicit Guidance (PEG) Algorithm”, Rockwell Space Operations Company, Novem-

ber 1992, JSC-261 22.

5 Ingoldby, R. N. (Martin Marietta Aerospace, Denver, Colo.), “Guidance and control system design of the Viking planetary

lander”, AIAA-1977-1060; Guidance and Control Conference, Hollywood, Fla., August 8-10, 1977, Technical Papers. (A77-42751
20-35) New York, American Institute of Aeronautics and Astronautics, Inc., 1977, p. 164-171.
6 Viking Lander System Primary Mission Performance Report, Martin Marietta Corporation (Denver, CO), April 1977,

NASA CR-145148.
7 Mendeck, G., Carman, G., ”Apollo-Derived Entry Guidance for Mars Smart Landers (AIAA-2002-4502),” AIAA Atmo-

spheric Flight Mechanics Conference, Monterey, CA, August 2002.

8 Crain, T., Bishop, R., ”Mars Entry Navigation: Atmospheric Interface Through Parachute Deploy (AIAA-2002-4501),”

AIAA Atmospheric Flight Mechanics Conference, Monterey, CA, August 2002.

9 Fahroo, F. and Ross, I. M., ”A Spectral Patching Method for Direct Trajectory Optimization,” The Journal of the

Astronautical Sciences, Vol. 48, No. 2/3, 2000.

10 Ross, I. M. and Fahroo, F., ”A Direct Method for Solving Nonsmooth Optimal Control Problems,” Proceedings of the

International Federation on Automatic Control, 15th World Congress, Barcelona, Spain, July 2002.
11 Ross, I. M. and Fahroo, F., ”User’s Manual for DIDO 2001: A MATLAB Application Package for Dynamic Optimiza-

tion”, NPS Technical Report AA-01-003, Department of Aeronautics and Astronautics, Naval Postgraduate School, Monterey,
California, June 2001.
12 Rea, J., ”Launch Vehicle Trajectory Optimization Using a Legendre Pseudospectral Method (AIAA-2003-5640),” AIAA

Guidance, Navigation, and Control Conference and Exhibit, Austin, Texas, August 2003.
13 Fahroo, F. and Ross, I. M., ”Costate Estimation by a Legendre Pseudospectral Method,” Journal of Guidance, Control

and Dynamics, Vol. 24, No. 2, Mar.-Apr. 2001, pp. 270-277.

14 Wong, E., Masciarelli, J., and Singh, G., “Autonomous Guidance and Control Design for Hazard Avoidance and Safe

Landing on Mars”, AIAA Paper 2002-4502, Atmospheric Flight Mechanics Conference, Monterey, CA, August 2002.

20 of 20

American Institute of Aeronautics and Astronautics