See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/225024879

In-Flight Demonstration of Formation Control based on Relative Orbital

Elements

Conference Paper
Source: DLR

CITATIONS READS
11 229

3 authors:

Simone D'Amico Jean-Sébastien Ardaens

Stanford University German Aerospace Center (DLR)
191 PUBLICATIONS 4,079 CITATIONS 66 PUBLICATIONS 1,310 CITATIONS

SEE PROFILE SEE PROFILE

Robin Larsson Nordström

Swedish Defence Research Agency
31 PUBLICATIONS 535 CITATIONS

SEE PROFILE

All content following this page was uploaded by Jean-Sébastien Ardaens on 28 May 2014.

The user has requested enhancement of the downloaded file.

 In-Flight Demonstration of Formation Control
based on Relative Orbital Elements

S. D’Amico1, J.-S. Ardaens1, R. Larsson2

1
German Space Operations Center, 82234 Wessling, Germany; 2Swedish Space
Corporation, Strandväg 86, P.O.Box 4207, Stockholm, Sweden

1 ABSTRACT

The fundamental objective of the PRISMA mission is to respond to the increasing demand of autonomous
formation flying and on-orbit servicing technology through the in-flight demonstration of novel guidance,
navigation and control (GNC) techniques. This paper addresses one of the primary experiments
conducted in the frame of the PRISMA mission to demonstrate broad autonomous formation keeping and
reconfiguration capabilities on a routine basis using GPS navigation, relative orbital elements, and
impulsive control. After a brief introduction of the adopted formation flying concept and its key algorithms,
the paper focuses on the experiment planning, operations and its performance in orbit. The obtained
results show the high readiness of the developed spaceborne GNC technology and pave the way for its
adoption in future advanced multi-satellite missions for remote sensing.

2 INTRODUCTION

This paper presents flight results from the first demonstration of autonomous formation keeping and
reconfiguration based on relative orbital elements in low Earth orbit. The results have been obtained in
the frame of the Spaceborne Autonomous Formation Flying Experiment (SAFE) [1] contributed by the
German Space Operations Center (GSOC) to the Swedish PRISMA mission [2]. SAFE is intended to
demonstrate fuel-efficient, collision free, accurate and robust relative motion control on a routine basis to
fulfill the typical requirements of future distributed sensors for Earth observation like synthetic aperture
radar interferometers and gravimeters. In this research a description in terms of relative orbital elements
has been preferred to the canonical Cartesian parameterization. In contrast to the fast varying Hill
variables, the use of orbital element differences simplifies the formation flying description and the satellite
relative position computation. Here the method of relative eccentricity/inclination vector separation, first
developed for the safe collocation of geostationary satellites, is generalized and applied to proximity
operations of formation flying spacecraft in orbit [3]. The spontaneous geometrical representation offers a
direct correlation between the relevant characteristics of the bounded relative motion in near circular orbit
and the magnitude/phase of the relative eccentricity/inclination vectors. This aspect extremely simplifies
the design of safe, passively stable formation-flying configurations. In particular minimum collision risk
conditions can be guaranteed by imposing the (anti-)parallelism of the eccentricity and inclination vectors
of the respective satellites, while J2-stable relative orbits are obtained by setting a specific nominal phase
for the configuration. The adopted approach is shown to be suitable either for the realization of synthetic
aperture radar interferometers with baselines below 1 km [4] or the application in longitude swap
operations with along-track separations above 200 km [5]. This paper addresses the first case where an
active relative orbit control strategy is necessary, in order to compensate for the main disturbance forces
represented by Earth’s oblateness perturbations and differential aerodynamic drag. The required velocity
budget for formation keeping and reconfiguration can be expressed in terms of relative orbital elements
and is directly proportional to the relative eccentricity and inclination offsets. The proposed analytical
feedback control law is adopted to maintain and reconfigure the PRISMA formation in a safe and fuel-
efficient way on a routine basis during the technology demonstration. After a brief description of the
relative orbital elements parameterization and the control strategy, the paper focuses on the second and
th th
final part of the SAFE experiment, executed in the time frame between 17 March and 4 April 2011. The
flight results shown in the paper demonstrate the fulfillment with margins of the prescribed objectives and
requirements which are at the basis of the guidance, navigation and control (GNC) system design [6,7].

4th International Conference on Spacecraft Formation Flying Missions & Technologies;

St-Hubert, Québec, 18-20 May 2011 1
3 ORBIT CONTROL CONCEPT

Under the assumptions of a Keplerian two-body motion, a circular chief orbit, and deputy spacecraft
separations which are small as compared to the chief orbit radius, the fundamental equations of dynamics
can be linearized to obtain the familiar Hill-Clohessy-Wiltshire (HCW) equations of relative motion [8].
These equations have a homogeneous analytical solution where the integration constants can be
recognized as a specific set of relative orbital elements [9]. The aforementioned solution can be
expressed to provide the relative Cartesian state vector in the orbital frame (aligned with radial R, along-
track T, and cross-track N, directions) at any mean argument of latitude u as a function of the relative
orbital elements {δa,δe,δi,δλ}T at epoch t0. δa represents the normalized difference of the spacecraft’s
semi-major axes, it introduces a constant offset in radial direction and a drift in along-track direction
proportional to the elapsed time u-u(t0). The relative eccentricity δe and inclination vectors δi drive the in-
plane and out-of-plane relative motion through their magnitudes (δe, δi) and phases (ϕ, ϑ) respectively. δλ
= δu + δiy⋅coti represents the relative mean along-track offset as a combination of relative mean argument
of latitude and the y-component of the relative inclination vector (i represents the chief orbit inclination).
The relative motion model can be extended by considering J2 and differential drag perturbations of the
orbital elements differences. The analytical treatment is based on the theory developed by Brouwer [10].
Here, only first-order terms in J2 are retained, thus small errors of the order of J22 and e⋅J2 are to be
expected on top of the quadratic terms related to spacecraft separation and eccentricity e, neglected by
the HCW equations. The secular variations of the relative orbital elements are easily obtained from the
differentiation of the secular variations of the Keplerian orbital elements for each formation-flying satellite
[9]. The inclusion of these variations provides a relative motion model which constitutes a first order
approximation of the solution of the equations of relative motion in the presence of J2 perturbations. The
secular motion of the relative eccentricity/inclination vectors caused by the Earth oblateness is illustrated
in Fig. 1. An initial configuration will ultimately be destroyed unless correction maneuvers are performed
to compensate for the natural drift of both vectors.

The proposed linearized relative motion model offers an ideal mathematical tool to process orbit
maneuvers. In particular we can express the relative orbital elements as a function of the relative position
and velocity by inverting the solution of the HCW equations. In other words the inversion of the linear
relative motion model provides the direct relation between an instantaneous velocity increment δv in the
Hill’s orbital frame and the consequent change of the orbital elements [9]

⎧ anδa = +2δv T
⎪ anδλ = −2δv − 3(u − u )δv
⎪ R M T
⎪⎪anδe x = +δv R sin u M + 2δv T cos u M
⎨ . (1)
⎪anδe y = −δv R cos u M + 2δv T sin u M
⎪ anδi x = +δv N cos u M
⎪
⎪⎩ anδi y = +δv N sin u M

For a given thrust, located at a specific mean argument of latitude uM, instantaneous variations of the
actual relative orbital elements are generated, as well as a net change of the mean longitude δλ within the
time interval between the maneuver execution and the epoch of the relative orbital elements. n and a
represents mean motion and semi-major axis of the chief orbit and amounts respectively to n = 0.00105s-1
and a = 7130688.2m at the beginning of the SAFE experiment. The control problem is fully decoupled
with respect to in-plane and out-of-plane. A thrust in cross-track direction affects only the relative
inclination vector, whereas a thrust in the orbital plane influences the relative eccentricity vector, the
relative semi-major axis and the relative longitude. The aim is to maintain the actual orbital differences
confined within symmetric control windows centered on the nominal values (superscript nom). As
explained in the previous section the relative e-/i-vectors are mainly affected by the Earth’s oblateness
perturbations. The former is characterized by a circular motion in the e-vector plane, the latter by a linear
drift proportional to the inclination difference (i.e., δix). We can take advantage of this natural secular
motion and define convenient control windows for the relative e-/i-vectors as depicted in Fig. 1. When the
magnitudes of the relative e-/i-vector tracking errors exceed the respective maximum allowed deviations,

2
δemax and δimax, the control scheme applies velocity increments at specific locations according to the linear
relative motion model. The computed velocity increments will transfer the relative orbital elements to the
opposite limit of the circular control window. Given the desired amplitude of the control windows and the
absolute reference orbit in terms of eccentricity and inclination, the desired relative e/i-vectors after each
orbit control maneuver can be determined uniquely. These are constants until a new nominal
configuration or new control windows are prescribed. The guidance strategy ensures that the time
between consecutive corrections of the relative e-/i-vectors (i.e., the maneuver cycle) is maximized and
represents as a consequence a fuel-efficient approach for formation maintenance.

Fig. 1. Representation of in-plane (left) and out-of-plane (right) maneuver locations and relationship to the
control windows for the relative eccentricity vector (left) and the relative inclination vector (right) [9].

Depending on the desired control window, or equivalently on the required orbit control accuracy, along-
track, radial and cross-track maneuvers (in the form of single or double-pulses) are executed at regular
time intervals in a deterministic fashion according to the following solutions of Eq. (1). An arbitrary
correction of the relative inclination vector δi can be realized through a single cross-track maneuver of
size δvN, at location uM, given by

⎧ δv N = naδi
⎨ , (2)
⎩u M = arctan(δi y / δi x )

which represents also the minimum delta-v solution for out-of-plane control according to Eq. (1). The
minimum delta-v solution for in-plane control provides an arbitrary correction of the remaining relative
orbital elements according to the following double-impulse scheme

⎧δv T1 = na (δa + δe) / 4 at u M1 = arctan(δe y / δe x )

⎨ , (3)
⎩δv T2 = naδa / 2 at u M2 = u M1 + π

where along-track maneuvers in flight or anti-flight direction are separated by half an orbit (the subscripts
1 and 2 indicate the first and second maneuver of the same pair). Here δa and δe represents the desired
corrections computed before the execution of the individual maneuvers, whereas δλ is indirectly
controlled through δa and does not appear explicitly in Eq. (3). An alternative approach for in-plane
control adopted within SAFE is based on the execution of radial maneuvers separated by half an orbit
given by

⎧δv R1 = na(δe − δλ / 2) / 2 and δv T1 = naδa / 4 at u M1 = arctan(δe x / δe y )

⎨ . (4)
⎩δv R2 = − naδλ / 2 and δv T2 = naδa / 2 at u M2 = u M1 + π

3
The choice of the most appropriate in-plane control mode (i.e., usage of Eq. (3) or (4)) is mission and
application dependant. Pairs of tangential-only maneuvers ensure minimum propellant consumption but,
on the other hand, when used for routine formation keeping, cause unintentional drifts in along-track
direction due to the continuous corrections of the semi-major axis. Pairs of radial maneuvers do not affect
the semi-major axis and can realize smaller corrections of the relative orbital elements due to the double
delta-v consumption. As shown in the SAFE experiment for PRISMA, the so called CL-T control mode
based on Eq. (3) can be efficiently used for large reconfigurations in along-track directions in rendezvous
scenarios, while the so called CL-R control mode based on Eq. (4) can be better used to accurately
control the formation for tight reconfigurations at short separations in fly-around and inspection phases.

4 EXPERIMENT PLAN

The PRISMA space segment comprises the fully maneuverable Mango (150 kg) small satellite as well as
the passive Tango (40 kg) sub-satellite which have been launched on June 15th 2010 into a Sun-
synchronous orbit at 750 km altitude. Mango features a three-axis reaction-wheel-based attitude control
and accommodates a hydrazine-based thrusters system that provides delta-v for orbit control. The system
provides thrust in all directions, using six 1-N thrusters that are capable of delivering impulse bits of 0.1
Ns. These translate to single velocity increments for formation control of ca. 0.6 mm/s at begin of life
which reduced to ca. 0.4 mm/s at the time of the experiment discussed in this paper. A total of 11 kg of
propellant provides a delta-v capacity of 115 m/s. Tango applies a coarse sun- magnetic-based three-axis
attitude control and does not have orbit control capability.

Fig. 2. Nominal formation geometries and associated relative orbital elements (from D to I, extract from Table
1) illustrated in the RTN orbital frame.

The SAFE experiment has been divided in two operational slots AFC1 (16 days) and AFC2 (19 days)
named from the Autonomous Formation Control (AFC) software implementing its functionalities and
integrated into the Mango FPGA-based LEON3 on-board computer. AFC receives 1-Hz estimates of the
absolute states of Mango and Tango provided by the GPS relative navigation system and implements a
state machine to realize the deterministic in-plane and out-of-plane maneuver planning discussed in the
previous section. The continuous navigation data are used to compute the deviation of the current relative
state from the desired relative state in order to generate velocity increments for autonomous maneuver
execution. The relative states are based upon a mean orbital elements representation [10].

At the time of paper writing the navigation system commissioning phase and both GNC operational slots
have been successfully completed [11-13]. AFC1 took place between September 20th and October 6th
2010 as an early harvest, whereas AFC2 was conducted between 17th March and 4th April 2011 as the
final part of the SAFE experiment. The plan of AFC2 is detailed in Table 1 which lists the nominal relative
orbital elements sent via telecommand to the AFC software for each day of operations. Each set of
relative orbital elements identifies a specific formation geometry (cf. ID in Table 1) which needs to be
acquired and maintained autonomously by the GNC system. Fig. 2 illustrates the nominal relative motion
of Mango w.r.t Tango which corresponds to a subset of the prescribed nominal relative orbital elements
(i.e., IDs from D to I). The adopted orbit and attitude control modes are also indicated.

4
Table 1. Nominal formations, orbit and attitude control modes planned for the 19-days SAFE (AFC2)
experiment between 17th March and 4th April 2011.
Day ID Control Attitude aδa aδex aδey aδix aδiy aδu
mode mode
1 A G Zenith 0 0 200 0 -200 4971.2
2 CL-T
3 B 3471.2
4 C 1971.2
5 D 471.2
6 Target
7 E Zenith -28.8
8 CL-R Target
9 F Zenith 150 -150 -21.6
10 G Zenith/Target 100 -100 -14.4
11 H Zenith 50 -50 -7.2
12 I Zenith/Target 30 -30 -4.3
13 J Zenith 30 -90 -12.9
14 K Zenith/Target 30 -150 -21.6
15 L Zenith 20
16 M Zenith/Target 100
17 N CL-T Zenith(-X) 15 -0.2 -100.1
18 O CL-R 0.2 -15 -104.3
19 P 0.2 -0.2 -100.1

First the relative orbit controller is operated in guidance (G) only mode for 1 day to verify the basic health
of the GNC functionalities, next AFC is operated in closed-loop (CL-T or CL-R) mode for the remaining 18
days. The relative eccentricity and inclination vectors are chosen anti-parallel and aligned with the y-axis
(i.e., aδex=aδix=0) to increase the safety of the formation and remove the secular J2 effects on aδiy. The
mean along-track separation is decreased gradually from 5 km to zero through formations A to E in a 6-
days long rendezvous phase (cf. aδu). Next the fly-around and inspection phase takes place for 9 days
until formation M. Here the relative eccentricity and inclination vectors are stepwise reduced in magnitude
to achieve a minimum separation of 20 m during formation L. In parallel the attitude control mode is
switched from the nominal pointing with quasi-zenith GPS antenna orientation on both satellites (cf.
“Zenith”), to the so-called “Target” pointing where the Digital Video System (DVS) camera mounted on
Mango is permanently pointing to the Tango satellite to take photos. Especially during the close-ups with
Mango flying along bounded relative orbits centered on Tango (i.e., IDs G, I, K and M, cf. Fig. 2), the
Target attitude mode poses challenges to the GPS navigation system which has to cope with rotating
GPS antennas and consequently unfavourable GPS visibility conditions.

The SAFE experiment is concluded by a 3-days phase characterized by a mean along-track separation of
-100 m and alternatively vanishing relative eccentricity/inclination vectors. Here the final goal is to
demonstrate the capability to keep an arbitrary hold point in along-track direction with nearly zero radial
and cross-track separations. Due to the usage of a Zenith pointing attitude mode (instead of Target
pointing), formations N and O gave the possibility to take photos with the Tango spacecraft moving in the
field of view of the DVS camera. The amplitude of 15 m for the relative eccentricity (N) or inclination (O)
vectors has been chosen to comply with the camera field of view of ca. 17 deg at 100 m along-track
separation.

5 FLIGHT RESULTS

This section presents key results obtained in orbit throughout the conduction of the SAFE (AFC2)
experiment described in the previous section. First of all a representative autonomous reconfiguration
which involves simultaneous relative e-/i-vector changes is described to show the overall functionality of
the GNC system, second detailed information is provided on the control performance for each formation

5
keeping and reconfiguration phase. Finally the accuracy of the absolute and relative GPS navigation
system is analyzed during the various attitude control modes used throughout the experiment.

Fig. 3. Autonomous in-plane and out-of-plane reconfiguration from formation F to G (cf. Tables 1-3) and
subsequent keeping on 25th March 2011.

Figure 3 illustrates the relative orbital elements estimated on-board and available in the telemetry stream
at 1 Hz on the 25th of March during the reconfiguration from F to G. Red dashed horizontal lines indicate
the nominal relative orbital elements commanded to the spacecraft for this phase (cf. Table 1, ID G). The
AFC state machine is implemented in such way to start the computation of the location uM, of the desired
in-plane or out-of-plane maneuver only after the violation of a so called “soft” control box. The soft control
box is set to 0.5 m during most of the SAFE experiment. At the acquisition of the desired location uM
along the orbit, AFC computes the desired size of the corrective maneuver δv, and sends the command
to the propulsion system for execution. After the single out-of-plane maneuver or after a pair of in-plane
maneuvers, the state machine restarts waiting for the next soft control box violation. In order to avoid
unnecessary control action due for example to maneuver execution errors, a deadband is introduced
through the implementation of a so called “hard” control box. If the control error lies in between soft and
hard limits, then an idle phase is entered which lasts a fraction of the previous maneuver cycle (typically
75%). The hard control box is set to 2.0 m during most of the SAFE experiment and is depicted with a
green dashed horizontal line in Figure 3.

The simultaneous in-plane and out-of-plane maneuver planning process is clearly visible from Figure 3
which shows how the relative orbital elements are controlled after the reconfiguration telecommand is
sent to the on-board software at 16:00 UTC prescribing a new set of nominal relative orbital elements (cf.
Table 1, ID G). The y-component of the relative inclination vector is transferred from ca. 150 m to 100 m
within the hard control box after one out-of-plane maneuver (cf. Fig. 3, bottom-right) at 17:18 UTC. The
size of the autonomously commanded maneuver is δvN = 5.2 cm/s and the maneuver location is uM = 89.5
deg, this is consistent with Eq. (2) for a desired correction aδiy = 50 m. At the occurrence of this
maneuver, cross-coupling effects due to small attitude misalignments cause undesired variations of all
other relative orbital elements and especially the relative semi-major axis and the relative mean argument
of latitude which vary respectively by 1.6 m (cf. Fig. 3, top-left) and 6 m (cf. Fig. 3, top-right). The
consequent linear drift of aδu caused by aδa is also visible during the subsequent 25 minutes (25% of one
orbit) of free motion. In parallel the in-plane maneuver planning performs the first pair of radial/along-track
maneuvers at 16:52 and 17:42 UTC separated by 50 minutes (50% of one orbit). Each in-plane maneuver
performs half the desired correction of aδey with δvR1 = 2.48 cm/s and δvR2 = -2.55 cm/s located
respectively at 0 deg and 180 deg mean argument of latitude (cf. Fig. 3, middle-right). These results

6
match very closely the delta-v computed from Eq. (4) for a desired variation aδey = 50 m. It is noteworthy
that the relative mean argument of latitude achieved after the second in-plane maneuver matches very
well the targeted value despite the undesired perturbation caused by the preceding cross-track
maneuver. This is made possible by the specific real-time implementation of Eq. (3) and (4) which
computes the size of the second maneuver based only on the residual error on aδa or aδλ shortly before
its execution. The effectiveness of the CL-R control approach based on Eq. (4) is finally demonstrated by
the removal of the errors on the semi-major axis (cf. Fig. 3, top-left) realized by the second in-plane
maneuver thanks to the execution of an along-track delta-v given by δvT2 = -0.92 mm/s together with the
execution of the radial maneuver. Full convergence to the nominal relative orbital elements is only
achieved after a second pair of in-plane maneuvers to absorb residual errors caused by maneuver
execution errors.

A comprehensive summary of the achieved formation keeping accuracy, convergence time and delta-v
usage is provided in the next sections for all formations listed in Table 1 during the SAFE (AFC2)
experiment.

5.1 Autonomous Formation Keeping

Table 2 lists the time frame (start and end time), duration (in orbital revolutions, 100 min), mean and
maximum control errors as well as the average delta-v commanded per orbital revolution (DV/rev) during
all formation keeping phases (A to P) of the SAFE (AFC2) experiment. For a description of each formation
flying configuration the reader is referred to Table 1 in Section 4. The average duration of fine formation
keeping for each configuration is 8.3 orbits (13.8 h) for a total of 133.5 orbits (9.3 d) or ca. 48.8% of the
19-days experiment slot. The maneuver cycle for in-plane control or the distance between consecutive
pairs of maneuvers is two orbital revolutions in average. This is mainly driven by the soft control box
threshold for the relative e-vector which is set to 0.5 m for most of the experiment. According to Eqs. (3-
4), the transfer of the relative e-vector to the opposite side of the control box requires a total δvT = δvT1 +
δvT2 = naδe/2 = 0.52 mm/s in CL-T mode or a total δvR = δvR1 + δvR2 = naδe = 1.05 mm/s in CL-R mode
for aδe = 1 m. As shown in the last column of Table 1 these results are basically confirmed in orbit. The
delta-v consumption nearly doubles when AFC is operated in CL-R mode (cf. Table 1).

Table 2. Autonomous formation keeping results during SAFE (AFC2).

Time frame Duration Control errors DV/rev
[dd/mm hh] [rev] abs(mean/max) in [m] [mm/s]
ID Start End R T N RT N
A 17/03 24 18/03 16 9.0 1.8/30.6 6.2/38.0 0.0/1.4 0.48 -
B 19/03 03 19/03 17 6.5 0.7/23.4 0.2/33.2 0.0/1.3 0.60 -
C 20/03 03 20/03 15 6.0 0.3/12.5 2.7/17.1 0.0/1.2 0.46 -
D 21/03 04 22//03 16 20.1 0.0/4.9 2.7/14.2 0.1/0.9 0.71 0.04
E 22/03 22 23/03 14 8.5 0.0/2.2 3.7/10.4 0.0/0.9 0.54 0.12
F 24/03 22 25/03 13 7.6 0.0/1.6 0.5/3.8 0.1/0.6 1.02 -
G 25/03 24 26/03 16 8.6 0.0/1.4 0.2/3.8 0.0/1.3 1.52 0.45
H 27/03 01 27/03 16 8.4 0.0/1.4 0.3/2.6 0.0/0.3 0.77 0.02
I 27/03 24 28/03 16 6.6 0.1/2.3 1.0/6.9 0.1/0.6 1.37 0.08
J 28/03 23 29/03 12 3.7 0.0/1.3 1.2/4.9 0.1/0.4 0.54 0.08
K 29/03 23 30/03 15 8.5 0.0/2.3 1.5/5.9 0.2/0.7 1.04 -
L 30/03 24 31/03 13 7.4 0.0/1.8 0.9/7.2 0.2/0.9 0.87 -
M 01/04 01 01/04 15 7.7 0.0/2.2 0.0/4.6 0.2/1.1 0.97 0.07
N 02/04 09 02/04 17 4.0 0.0/1.9 0.1/5.2 0.0/0.2 1.03 0.05
O 02/04 23 04/04 06 17.9 0.0/2.7 1.6/6.2 0.0/1.0 0.81 0.15
P 04/04 10 04/04 15 3.0 0.1/1.1 0.4/1.7 0.0/0.1 0.47 0.03
Total (133.5 rev) 121.5

7
The irregularity of the commanded delta-v listed in Table 1 as compared to the nominal trend is due to the
tight control box adopted during the experiment for two main reasons. First of all the individual maneuvers
are of the same order of magnitude of the minimum impulse bit of the propulsion system during this
phase (ca. 0.4 mm/s) and are affected by relatively large execution errors (ca. 10%). Second the desired
corrections of the relative e-vector are affected by relatively large navigation errors (ca. 10%, cf. Section
5.3). Furthermore the nominal relative e-vector is gradually decreasing in magnitude from formation A to
P thus requiring gradually larger maneuver cycles to violate the control box. Due to the specific choice of
the nominal relative i-vector (i.e., aδix = 0) the delta-v required for out-of-plane control is nearly zero and
limited to sparse cross-track maneuvers which compensate accumulated cross-coupling and second-
order dynamics effects.
The control errors are defined as the difference between the actually obtained and the desired relative
position of Mango w.r.t. Tango. The true relative position is derived from the post-facto Precise Orbit
Determination (POD) reference available on ground as product of the PRISMA verification layer [12] (cf.
Section 5.3). Whereas the desired relative position is computed from the solution of the HCW equations
expressed in terms of relative orbital elements [9]. By definition the defined control tracking errors show
secular, short- and long-period effects which are caused by actual errors in the tracking of the nominal
relative orbital elements and in addition unmodelled second order effects in spacecraft separation,
eccentricity and Earth’s oblateness. Being the pattern dominated by the second order relative dynamics,
Table 1 lists the mean and maximum control tracking errors instead of standard deviations figures. As
expected the radial and along-track control errors decrease with the spacecraft separation, while the
cross-track control error is basically regular. Formations with reduced mean along-track separations (e.g.,
E to P, with aδu < 500 m) are characterized by maximum control errors below 2.7/10.4/1.3 m in R/T/N,
whereas formations with increased along-track separations (e.g., A to D, with aδu > 500 m) are
characterized by larger offsets and maximum errors of 30.6/38.0/1.4 m in R/T/N (cf. A). The accumulated
total delta-v commanded by AFC during formation keeping phases is 12.15 cm/s.

5.2 Autonomous Formation Reconfiguration

Table 3 lists the time frame (start and end time), convergence duration (in orbital revolutions, 100 min),
the total delta-v commanded by AFC, as well as the ideally required delta-v (from Eqs. (2-4)), during the
formation reconfiguration phases (A to P) of the SAFE (AFC2) experiment.

Table 3. Autonomous formation reconfiguration results during SAFE (AFC2) between 17th March and 4th April
2011. (*) Ideal delta-v evaluated over two orbits reconfiguration time.
Time frame Duration DV ideal DV real
[dd/mm hh] [rev] [cm/s] [cm/s]
ID Start End RT N RT N RT N
A 17/03 18 17/03 24 3.0 1.2 3.61 2.63 3.21 3.16
B 18/03 18 19/03 02 5.0 - 8.33* - 9.77 0.04
C 19/03 18 20/03 01 4.5 - 8.33* - 10.71 0.03
D 20/03 16 21/03 04 7.4 - 8.33* - 45.66 0.33
E 22//03 16 22/03 23 2.5 - 2.78* - 3.26 0.03
F 24/03 16 24/03 21 2.7 1.2 5.24 5.24 6.74 6.14
G 25/03 16 25/03 20 2.1 1.0 5.24 5.24 5.72 6.15
H 26/03 17 27/03 24 3.9 2.4 5.24 5.24 5.79 6.21
I 27/03 16 27/03 21 2.4 0.6 2.10 2.10 2.20 2.44
J 28/03 19 28/03 20 - 0.7 0.45 6.29 0.36 7.39
K 29/03 14 29/03 16 - 0.8 0.45 6.29 0.76 7.52
L 30/03 14 30/03 17 1.8 - 1.05 - 1.15 0.08
M 31/03 14 31/03 18 2.4 - 8.39 - 9.97 0.06
N 02/04 04 02/04 09 3.0 2.4 8.91 15.71 9.78 18.68
O 02/04 19 02/04 22 1.5 0.4 1.77 1.55 1.78 1.84
P 04/04 06 04/04 10 2.1 2.1 0.22 1.55 0.62 2.00
Total (44.3 rev) 179.58

8
For an evaluation of the desired correction of the relative orbital elements for each reconfiguration, the
reader is referred to Table 1 in Section 4. The convergence duration is defined as the time elapsed from
the reconfiguration telecommand (prescribing a new set of relative orbital elements) to the acquisition of
the desired formation with control tracking errors steadily below the maximum values listed in Table 2.
The average convergence time in cross-track direction is 1.3 orbits for the 10 commanded out-of-plane
reconfigurations, which means that two cross-track maneuvers are in general sufficient to obtain any
desired correction of the relative i-vector. The average convergence time in radial and along-track
directions is correlated and amounts to 2.4 orbits for the 9 commanded in-plane reconfigurations of the
relative e-vector. This means that typically 2-3 pairs of in-plane maneuvers are sufficient to enter the fine
formation keeping phase. Mean along-track reconfigurations (i.e., A to E) require longer convergence
times which amount to an average of 3.7 orbit revolutions. This value does not consider the
reconfiguration from C to D as affected by a contingency at initialization. According to the adopted control
scheme, the performance of the along-track reconfigurations is driven by the regularity of the maneuver
cycle and ultimately by the behaviour of the relative e-vector. In this case a degradation of the navigation
performance caused an unexpected violation of the control box for aδe and a too small maneuver cycle
after the first maneuver pair. The subsequent along-track overshooting required corrections of aδa larger
than allowed hich caused an overall degradation of the control performance in terms of convergence time
(7.4 orbits) and commanded delta-v (45.66 cm/s). This event is under investigation to improve the overall
robustness of the control algorithms. On the other hand, excluding formation D, the overall consistency
between actually commanded and ideal delta-v is reasonably good and shows a total discrepancy of
17.2% over the formation reconfigurations autonomously executed during SAFE.

5.3 Absolute and Relative Navigation

As discussed in Section 4, the AFC control software makes use of the GPS navigation output to compute
the orbital elements used for maneuver planning. A comprehensive description of the GPS-based
navigation system and its design is provided in [6]. The meter level control accuracy demonstrated in the
previous sections is a clear but indirect indication of the quality of the navigation filter output. Anyway for
completeness this section analyses the accuracy of the on-board real-time navigation through a direct
comparison with the post-facto on-ground POD reference.

Table 4. On-board navigation accuracy results during SAFE (AFC2) between 17th March and 4th April 2011.
Positioning Nearly zenith S/C tumbling with Sporadic large attitude rotations
errors pointing antennas antenna switches with antenna switches
Absolute [m] Mean Std Mean Std Mean Std
Radial 0.0 2.0 0.0 2.4 0.0 2.0
Along-track 0.6 0.9 0.5 1.1 0.5 1.1
Cross-track 0.1 0.6 0.1 0.8 0.1 0.8
3D 1.9 1.4 2.1 1.8 2.0 1.5
Relative [mm] Mean Std Mean Std Mean Std
Radial -28.3 29.3 -44.4 322.4 -22.6 125.2
Along-track 5.2 25.2 9.8 274.9 11.9 151.4
Cross-track -10.4 15.8 -63.1 190.8 3.5 48.7
3D 47.4 22.5 167.0 440.7 80.6 189.2

This approach is justified since the POD layer has been verified, calibrated, and commissioned prior to
the conduction of the SAFE experiment in the time frame between June and September 2010 [12]. Table
4 lists the statistics of the GPS-based absolute and relative navigation errors obtained during SAFE
(AFC2). The navigation errors are computed through a comparison of the navigation state generated on-
board by the GNC subsystem with the POD products generated post-facto on-ground. The statistics are
split in three representative phases to show the dependency of the navigation accuracy from the adopted
attitude control mode. Continued nearly Zenith-pointing GPS antennas (±10 deg) are used for a total
duration of 8 days. This favourable scenario provides 3D relative positioning accuracies of 4.74 ± 2.25 cm
(cf. Table 4) which take advantage of an average of 10 GPS satellites commonly tracked by the Phoenix
GPS receivers on-board Mango and Tango. The accuracy degrades to 16.70 ± 44.07 cm during extended
periods of spacecraft tumbling with GPS antenna switches every half orbit. Such attitude mode is used for

9
 an accumulated time of 4 days to verify the robustness of the GNC system under unfavourable GPS
satellite visibility conditions. Here the navigation accuracy is dominated by the noise of the pseudorange
measurements due to the lack of commonly visible GPS satellites. The subsequent Target pointing
operations are limited to 2-3 orbital revolutions per day (for a total of 6 days) and show an accuracy of
8.06 ± 18.92 cm which is somewhat half way between the best and worst presented scenarios.

6 CONCLUSIONS

This paper has presented one of the first comprehensive demonstrations of autonomous formation control
technology in low Earth orbits based on relative orbital elements. The analysis of the flight results
gathered during the Spaceborne Autonomous Formation Flying Experiment (SAFE) conducted in the
frame of the Swedish PRISMA mission shows the effectiveness of the adopted approach together with its
limitations. If on one hand a parameterization in terms of relative orbital elements simplifies extremely the
implementation on-board of a deterministic and robust maneuver planning strategy, on the other side the
resulting control functionality can only be used to maintain or reconfigure passive relative orbits which
allow hold points in along-track direction only. The demonstrated control accuracies below 2.7/10.4/1.3m
in R/T/N for separations below 500m are well within the typical requirements of future distributed satellite
systems for Earth observations. On the other hand the control strategy does not exploit the full potential
of the demonstrated 3D relative positioning accuracies at 4.74±2.25cm available onboard. In fact the
formation keeping control accuracy as well as the formation reconfiguration convergence times are
fundamentally constrained by the analytical feedback control law itself which relies on pairs of maneuvers
every one-two orbital revolutions. Despite these intrinsic limitations SAFE has demonstrated spaceborne
autonomy capabilities with superior flexibility and reliability as compared to classical ground-in-the-loop
approaches. Ultimately this research shows how low cost commercial-off-the-shelf GPS hardware and
simple GNC techniques enable the transfer to the spacecraft of tasks traditionally performed on ground.

7 REFERENCES

1. D'Amico S., Ardaens J.S., De Florio S., Montenbruck O., Persson S., Noteborn R.; GPS-Based Spaceborne
Autonomous Formation Flying Experiment (SAFE) on PRISMA: Initial Commissioning; AIAA/AAS Astrodynamics
Specialist Conference, 2-5 August 2010, Toronto, Canada.
2. Persson S., D'Amico S., Harr J.; Flight Results from PRISMA Formation Flying and Rendezvous Demonstration
Mission; IAC-10-B4.2.9; 61st International Astronautical Congress, 27 Sep - 1 Oct 2010,Prague, CZ. (2010).
3. D'Amico S., Montenbruck O.; Proximity Operations of Formation Flying Spacecraft using an
Eccentricity/Inclination Vector Separation; AIAA Journal of Guidance, Control and Dynamics, 29/3 554-563
(2006).
4. Montenbruck O., Kahle R., D'Amico S., Ardaens J.-S.; Navigation and Control of the TanDEM-X Formation;
Journal of the Astronautical Sciences, 56(3):341-357 (2008).
5. Montenbruck O., Kirschner M., D'Amico S., Bettadpur S.; E/I-Vector Separation for Safe Switching of the GRACE
Formation; Aerospace Science and Technology 10/7, 628-635 (2006).
6. D’Amico S., Ardaens J.-S., Montenbruck O.; Navigation of Formation Flying Spacecraft using GPS: the PRISMA
Technology Demonstration; ION-GNSS-2009, 22 Sep. - 25 Sep. 2009, Savannah, USA (2009).
7. D’Amico S., De Florio S., Larsson R., Nylund M.; Autonomous Formation Keeping and Reconfiguration for
Remote Sensing Spacecraft; 21st International Symposium on Space Flight Dynamics, 28 Sep. -2 Oct. 2009,
Toulouse, France (2009).
8. Clohessy, W.H., and Wiltshire, R.S. (1960). Terminal guidance system for satellite rendezvous. Journal of the
Aerospace Sciences, Vol. 27, pp. 653–658.
9. D’Amico, S. Autonomous formation flying in low earth orbit, Ph.D. Dissertation, Technical University of Delft,
ISBN 978-90-5335-253-3 (2010).
10. Brouwer D. . Solution of the problem of artificial satellite theory without drag. Astronomical Journal,
64(1274):378–397, 1959.
11. D'Amico S., Ardaens J.-S., De Florio S.; Autonomous Formation Flying Based on GPS - PRISMA Flight Results;
6th International Workshop on Satellite Constellation and Formation Flying, 1-3 Nov. 2010, Taipei, Taiwan
(2010).
12. Ardaens J.-S., D'Amico S., Montenbruck O.; Final Commissioning of the PRISMA GPS Navigation System; 22nd
International Symposium on Spaceflight Dynamics; 28 Feb. - 4 March 2011, Sao Jose dos Campos, Brazil
(2011).
13. D'Amico S., Larsson R.; Navigation and Control of the PRISMA Formation: In-Orbit Experience; 18th IFAC World
Congress, 28 Aug.- 2Sep 2011, Milano, Italy (2011).

10

View publication stats