(Preprint) AAS 13-006

ORION EXPLORATION FLIGHT TEST-1 CONTIGENCY DROGUE

DEPLOY VELOCITY TRIGGER

Robert S. Gay*, Susan Stochowiak†, and Kelly Smith‡

As a backup to the GPS-aided Kalman filter and the Barometric altimeter, an

“adjusted” velocity trigger is used during entry to trigger the chain of events that
leads to drogue chute deploy for the Orion Multi-Purpose Crew Vehicle
(MPCV) Exploration Flight Test-1 (EFT-1). Even though this scenario is multi-
ple failures deep, the Orion Guidance, Navigation, and Control (GN&C) soft-
ware makes use of a clever technique that was taken from the Mars Science La-
boratory (MSL) program, which recently successfully landing the Curiosity rov-
er on Mars. MSL used this technique to jettison the heat shield at the proper time
during descent. Originally, Orion use the un-adjusted navigated velocity, but the
removal of the Star Tracker to save costs for EFT-1, increased attitude errors
which increased inertial propagation errors to the point where the un-adjusted
velocity caused altitude dispersions at drogue deploy to be too large. Thus, to
reduce dispersions, the velocity vector is projected onto a “reference” vector that
represents the nominal “truth” vector at the desired point in the trajectory. Be-
cause the navigation errors are largely perpendicular to the truth vector, this pro-
jection significantly reduces dispersions in the velocity magnitude. This paper
will detail the evolution of this trigger method for the Orion project and cover
the various methods tested to determine the reference “truth” vector; and at what
point in the trajectory it should be computed.

INTRODUCTION
The Orion Multi-Purpose Crew Vehicle (MPCV) Exploration Flight Test-1 (EFT-1) uses alti-
tude to trigger many of the key events during Entry Descent and Landing (EDL). One of the pri-
mary events is drogue parachute deploy. The drogue chutes are used to slow the descent to an
acceptable rate for the larger main chutes, which allow the capsule to land soft enough for the
crew to survive. There are two sources for navigated altitude on Orion: 1) The Primary Global
Positioning System (GPS)-aided Kalman filter, and 2) The backup barometric altimeter (set of
three). If the GPS and barometric altimeters fail for whatever reason, the inertial-only solution is
used to compute the altitude (Kalman filter with no GPS or backup inertial-only solution if the
Kalman filter is corrupted). The inertial-only solution is derived by propagating the Inertial
Measurement Unit (IMU) data along with a gravity model. This method is robust, but much less
accurate than filtered GPS or a barometric altimeter (especially in altitude). Consequently, an al-

*
Orion NASA Absolution Navigation Lead, NASA Johnson Space Center, Houston, TX 77058.
†
Orion Entry MODE Team member, NASA Johnson Space Center, Houston, TX 77058.
‡‡
Orion Nav and Entry MODE Team member, NASA Johnson Space Center, Houston, TX 77058.

1
ternatte trigger method of velociity magnitudee is used for ddrogue deployy if the GPS aand barometric
altimeeters fail. Maiin chute deplo
oy and other events
e are alll subsequentlyy based on tim
me.
Orriginally, this velocity triggger was deveeloped when O Orion still haad a Star Traccker for precisse
attitud
de-determinattion on orbit. This allowed d the inertial--only solutionn to be good enough to usse
the naavigated velo ocity magnitu ude directly with
w no augm mentation. Hoowever, for E EFT-1 the Staar
Track ker was remov ved to save costs,
c thus inccreasing the ppossible attituude error at E
Entry Interfacce
(EI). The increasee in attitude error degrades the inertiaal-only solutiion such thatt the velocityy-
triggeer dispersionss are too great for droguee deploymentt. Fortunatelyy, a techniquue used by thhe
Mars Science Labo oratory (MSL L) program1 that
t just recenntly landed thhe Curiosity rrover, was im
m-
plemeented to reducce the errors in velocity, and
a thus reduuced the droguue deploy disspersions. Thhis
techniique involvess projecting th he navigated velocity vecttor onto a refeerence “truth”” vector. Sincce
much of the navigation errors are a perpendicular to the acctual velocityy vector (largeely due to attti-
tude errors),
e the errror in the maagnitude is siignificantly rreduced. Obviiously, the keey to adjustinng
the veelocity vectorr is to properrly determine the referencee “truth” vecttor. Several ddifferent methh-
ods foor computing g the referencce “truth” vecctor were exaamined relativve to perform mance and efffi-
ciency w the velocity trigger waas used, how it was “adjussted”, and how
y. This paperr will detail why
well it
i performed.

VELO
OCITY TRIG
GGER NEEDED FOR IN
NERTIAL-O
ONLY NAV
Ass mentioned above,
a if GPS and the barrometric altim meters fail, thhe Orion navvigated altitudde
relies solely on thee IMU and an n on-board grravity model ffor state proppagation. Thiss leads to largge
errorss that generate drogue dep ploy dispersioons that are b eyond vehiclle capability. Thus, in ordeer
to pro
otect for this unlikely
u even
nt, an alternatee trigger is ussed that is bassed on the maagnitude of thhe
planett-relative velo ocity. Figure 1 shows how w the inertiall altitude erroors grow duriing entry. Attti-
tude errors
e at EI arre normally diispersed with h a 3-sigma, zeero-mean, vaalue of 0.4 degg.

Figure 1 Inertial-only
y Altitude Errror (EI to G
Ground)

2
 In order to determine at what velocity value to trigger drogue deploy (or actually the jettison
of the Forward Bay Cover (FBC) that starts the sequence leading to drogue deploy), a set of Mon-
te Carlo runs was examined2. Specifically, statistics were taken on the velocity magnitude at
which the desired drogue deploy altitude was reached. These results have varied slightly as the
simulations and the flight software have matured. Figure 2 shows some early results that put the
range between 435 ft/s and 503 ft/s. Choosing the higher side allows for more time to complete
the chute sequence, while picking the low side provides more time for the barometric altimeters
to come on line. Initially the high side was preferable, but later the value was reduced to the low
side to keep from triggering on velocity before the barometric altimeters had useable data (not
valid at high altitudes and Mach number). The current trigger value being used is 425 ft/s.

Figure 2 Relative Velocity at Drogue Deploy Altitude (24 kft)

As stated earlier, one of the primary inertial-only propagation error sources is the initial atti-
tude error at Entry Interface. Monte Carlo results show that initial attitude errors of 0.3 degrees or
greater will cause velocity-triggered drogue deployment to have altitude dispersions too great for
a successful landing. Conversely, the results also showed that initial attitudes less than 0.1 de-
grees would produce satisfactory results. Figure 3 includes performance results for initial attitude
errors of 0.1, 0.3, and 0.5 degrees2. The trigger value used was 503 ft/s.

3
 Figure
F 3 Stan
ndard Velociity-Triggered
d Drogue and
d Main Deplloyment (5033 ft/s trigger)

JUSTING” NAVIGATED
“ADJ N D VELOCIT
TY
Affter the Consttellation Program was canccelled the inittial mission sccope for Orioon was reduceed
to a flight
f test thaat consists of two earth orrbits. The seccond orbit is very ellipticaal generating a
high entry
e velocityy needed to teest aero-thermmal properties and create a similar envirronment to that
of retuurning from thet Moon, Mars, or some other celestiaal body. As part of this redduction in mis-
sion scope,
s many unnecessary
u components
c were
w removedd including thhe Star Trackeer used for pree-
cise inertial attitud de knowledgee while in orrbit. Since thhe mission is less than fivve hours, IMU U
propaagation is suffficient to main ntain attitude accuracy. In fact, the Kallman filter cann estimate attti-
tude due
d to dynamical correlatio ons accumulaated during atm mospheric fliight (both asccent and entryy).
Nonettheless, the vehicle
v must achieve misssion objectivees assuming no updates tto the attitudde.
Thus, the potentiall error at EI is up to 0.4 deegrees. This iis a substantiaal increase frrom the perfoor-
mancee with a Starr Tracker. As can be seen from the datta in Figure 33, attitude errors larger thaan
0.3 deegrees cause the standard velocity trigg ger to have ddispersions beeyond the vehhicle’s capabiil-
ity. Figure 4 show ws just how laarge those errrors can be. C Consequentlyy, some adjusstment must bbe
made in order to usse the planet-relative veloccity magnitudde as a droguee deploy triggger.

4
 Figure 4 Relative Velocity Magnitude Error at Drogue Deploy

Fortunately, there does exist such an “adjustment” to the velocity so that the errors can be re-
duced enough to make it useful. A few years ago the Orion and MSL Entry Descent and Landing
teams came together for a three-day meeting to exchange ideas and try to learn from each other.
One of those ideas was a clever method used to trigger the MSL heat shield jettison1. Originally,
the relative velocity was to be used, but the dispersions were too large to meet overall vehicle
constraints. After, examining the problem, MSL engineers determined much of the navigation
error is due to attitude errors, and that this manifested in such a way that the error is primarily
perpendicular to the truth vector. Figure 5 generalizes the velocity error accumulated during iner-
tial-only propagation in the atmosphere. Clearly, projecting the navigated velocity vector onto the
truth vector can reduce much of the magnitude error.

Nav vector
Nav error

Truth vector
Figure 5 Generalization of Inertial-only Velocity Propagation Error

Initially, this technique was not used on Orion due to the wide variety of possible entry trajec-
tories (driven by anytime-return requirements). As will be detailed in the following section, the
“truth” vector must be deterministic with this method. However, the current Orion EFT-1 trajec-
tory is very predictable making this technique useful. Figure 6 shows the results of a 3000 Orion
Entry Monte Carlo run where the velocity error vectors at three altitudes (25 kft, 50 kft, & 75 kft)
are all plotted in 3-D forming a “disc”. This “disc” confirms that the navigation vector “cones”

5
arounnd the truth veector such thaat the error veectors form a “disc” (or boottom of the ccone). Now aall
that reemains is deteermining the reference “tru uth” vector too project ontoo. This and thhe performancce
of thee “adjusted” velocity
v triggeer are coveredd in the follow
wing section.

Figure 6 Oriion Entry Monte Carlo – Velocity Errror Vectors ((25 kft, 50 kfft, & 75 kft)

ERMINING REFERENC
DETE CE “TRUTH
H” VECTOR
R
Th
here are two key
k issues witth generating the necessaryy reference “ttruth” vector to “adjust” thhe
planett-relative velocity vector magnitude: 1)
1 How to coompute it, annd 2) Where to compute iit.
Four different metthods were ex xamined to co
ompute the vvector along w with three diffferent altitudde
pointss in the entry
y trajectory. The
T sections below
b will deetail the methhods, locationns, and the peer-
formaance of each.
Meth
hods for Geneerating the Reference
R “Truth” Vectorr
Obbviously theree is no way to
o know what the actual plaanet-relative vvelocity really is, so it muust
be computed from m current navigation parammeters or compmputed prior too flight throuugh simulationn.
For siimplicity, the Orion EFT-1 1 design uses a constant veector that is ddetermined prre-flight. Fouur
differrent methods for computing the referencce “truth” vecctor were evaaluated for peerformance annd
efficieency. The following sectio
ons describe each
e of the meethods.
Noominal Run. This is the simplest metthod. All thaat is requiredd is to make a single nonn-
disperrsed (GPS-qu uality navigation) entry sim mulation run and snapshott the true plannet-relative vee-
locity
y vector at thee desired altittude. The vecctor is then noormalized to facilitate the use of the Doot
Produuct to perform
m the projectio on.
Mean
Me of Montee Carlo. This method is stiill quite simpple and only rrequires runniing one Montte
Carlo set and averraging the veelocity vectorrs at the desirre altitudes. T
This average vector is theen
normaalized for usee as the reference “truth” vector.
Veector Cross Product.
P This method is a little more commplicated. It requires that the navigatioon
velociity error vecto
ors from a Monte Carlo seet (taken at thhe desired altittudes) be used to form muul-
tiple cross products. These vecctor cross pro oducts are coomputed by rrandomly sam mpling a largge
numb ber of vector pairs
p taken from the Montte Carlo data. Then all of tthe resulting vvectors are avv-

6
eraged
d and the vecctor is normallized. This method is attem
mpting to findd the unit vecctor that is peer-
pendicular to the “disc” shown in
i Figure 6.
Grradient Desceent. This is th
he most comp plicated methhod. Using thhe data from a Monte Carllo
entry run, an iterattive Gradient Descent metthod is used tto minimize tthe angles bettween the choo-
sen veector and all the
t planet-relative velocity
y vectors takeen at the desirred altitudes. O
Once the iteraa-
tive process
p has coonverged, the vector is norm
malized.
Trajeectory Locatiions for Geneerating the Reference
R “T
Truth” Vectoor
Ass stated earlieer, the referennce “truth” vectors were geenerated at thhree different altitudes alonng
the enntry trajectory k and 75 kftt. The reason for choosing altitudes other than just thhe
y: 25 kft, 50 kft,
droguue deploy altittude, is to find a place wheere the distribbution of the pplanet-relative velocity vecc-
tors are
a relatively Gaussian.
G If the
t distributioon is not nearrly Gaussian, it will be diffficult to find a
singlee reference “ttruth” vector that will pro ovide reasonaable results. A As it turns ouut, the altitudde
nearest to the drog gue deploy po oint of 24 kft,, is not very G
Gaussian. In ffact, the 50 kkft point is alsso
not veery Gaussian.. The 75 kft point,
p howeveer, is fairly Gaaussian and thhe spread is ttighter than thhe
other two altitudess. The distribu utions at 25 kft
k and 50 kft ft are likely leess Gaussian due to the fact
that Guidance
G inters the “Termiinal Steering”” phase arounnd 60 kft. Thiis guidance pphase has morre
directtional variabiility than the previous phaase. Figure 7 through Figgure 9 show a general azzi-
muth//elevation histogram of thee velocity vecctors at each oof the altitudees respectivelly. The characc-
teristiics of the distrribution can be
b seen clearly in each of tthe 3-D plots..

Figure 7 Azimuth and Elevation Histogram

H o f Truth Veloocity Vector aat 25 kft

7
Figure 8 Azimuth and Elevation Histogram
H o f Truth Veloocity Vector aat 50 kft

Figure 9 Azimuth and Elevation Histogram

H o f Truth Veloocity Vector aat 75 kft

8
Referrence “Truth
h” Vector Perrformance
Eaach of the refference “truth
h” vectors, computed as ddescribed in the previouss sections, waas
d with a 3000 Monte Carlo set. The resu
tested ults are shownn in

Taable 1. Note that

t the veloccity trigger vaalue used wass 425 ft/s. Ass expected, thhe performancce
for the points at 25
5 kft and 50 kft
k was not accceptable. Thee spread was llarge and therre were several
hundrred runs wherre the Mains never deploy yed. The refeerence “truth”” vector compputed at 75 kkft
did ho owever perfoorm well. Forr this altitudee point, the ddeploy spreadd for all the vvector methodds
was close
c and mosst had similarr means (except for the veector cross prroduct). Sincce all the refeer-
ence “truth”
“ vectorrs performed similarly, it was
w decided to keep the pprocess simple and computte
the veector by takin
ng the mean of o the Monte Carlo set. Thhis method waas chosen oveer the Nominaal
Run method
m becauuse it is likely
y more robust than a singlle run. In adddition, this w was the original
metho od used to commpute the Baseline vector noted in the ffirst row of T Table 1. The BBaseline vectoor
was computed
c prio
or to this trad
de study and was
w also calcculated at the 75 kft altitudde point. Sincce
the orriginal Baselinne reference “truth”
“ vectorr performed sslightly betterr than the more current vecc-
tor, it was decided to retain the Baseline
B vecttor until simuulations provee that a change is needed.

Table 1 Refference “Tru

uth” Vector Performance
P e – Various M
Methods and
d Locations

Thhe Baseline reeference “truth h” vector wass then tested w with various iinitial attitudee errors at Enn-
2
try Interface. The Monte
M Carlo results
r for a velocity triggger of 503 ft/ss are illustrateed in Figure
10. It can clearly beb seen that thhe “adjusted” velocity triggger substantiaally reduced ddispersions for
initiall attitude errors 0.3 degreess and above asa compared tto the standarrd velocity triggger shown inn
Figure 3. The sensitivity to initiial attitude errror was also ggreatly reduceed. Finally, Fiigure 11
2
showss the sensitiviity of the veloocity trigger value
v . The hiigher value haas less spreadd and gives
more time for the chute
c deploy sequence, butt increases th e maximum ddeploy altitudde. The lowest
value deploys the chutes
c to too low. The mid ddle value proovides a goodd compromisee that deploys

9
all thee chutes high enough, but also
a gives thee most time foor the baromeetric altimeterrs to provide
valid data. As men ntioned earlierr, the current velocity triggger value is 4225 ft/s.

Figure
F 10 “A
Adjusted” Velocity-Triggeered Drogue and Main D
Deployment (503 ft/s trig)

Figure 11 “A
Adjusted” Velocity-Trigg
V gered Perforrmance – Varrious Triggerr Velocities

LOOKING BEYO
OUND ORIO
ON EFT-1
On ne of the draw
wbacks to deffining the refeerence “truth”” vector priorr to flight is thhat it limits it’s
usefullness to a giv
ven entry trajjectory relativ ve to the plannet of interest (Earth, Marrs etc.). Futurre
Orionn missions wiill likely requ uire that this technique w work with a vvariety of enttry trajectoriees.
Fortun variety” in thee trajectory iss mainly only planet-relativve. The “shappe” of the final
nately, the “v
guidaance phase is very
v similar even
e for long skip entries. The only reaal difference iis where on thhe

10
planet the vehicle is headed. Therefore, it should be possible to develop a backup drogue deploy
trigger scheme that is valid for all entries using the same guidance scheme. Fortunately, early
studies using Multivariate Logistical Regression techniques have already shown great promise.
Several navigation signals are utilized to evaluate if the proper conditions are met for deploy-
ment. The Logistical Regression technique properly weights each signal allowing for large errors
and dispersions. Future studies will also look at applying this technique as the primary method for
triggering many of the key Entry, Descent, and Landing events.

CONCLUSION
Although it is very unlikely that the GPS and the barometric altimeters will all fail, this study
shows that the Orion EFT-1 GN&C design has a robust alternate method to trigger drogue and
main parachute deployment. In addition, this design is a great example of successful collaboration
between two NASA programs: Orion MPCV and MSL. Finally, this work has also led the Orion
GN&C team to investigate other trigger methods that will likely increase robustness and accuracy
for future missions to come.

REFERENCES
1
Kipp, D., San Martin, M., Essmiller, J., Way, D., “Mars Science Laboratory Entry, Descent, and Landing Triggers,”
IEEEAC #1445, IEEE Aerospace Conference, March 3-10, 2007, Big Sky, MT.
2
Stachowiak, S., “Assessment of Velocity Trigger as a Backup Drogue Chute Deploy Option for Orion Exploration
Flight Test #1,” FltDyn-CEV-12-4, NASA Johnson Space Center, February 14, 2012, Houston, TX.

11