IMPROVEMENTS TO THE FEASIBILITY RANGE OF RENDEZVOUS

MANEUVERS USING AERODYNAMIC FORCES

M. Walther* a, C. Trauba, G. Herdricha, S. Fasoulasa
aInstitute of Space Systems (IRS), University of Stuttgart, Pfaffenwaldring 29, 70569 Stuttgart, Germany

Abstract
Several small, unconnected satellites flying in formation are a frequently discussed option to replace large satellites due to
advantages such as increased flexibility, reduced costs and enhanced reliability. Using differential aerodynamic forces as a means
to maintain the formation despite present perturbations, to perform formation reconfiguration or to initiate a rendezvous maneuver
is a promising propellant-less alternative to chemical and/or electric propulsion systems. Subjected to uncertainties, dynamic
variations and non-linearities its implementation in a real mission is, however, challenging. A common practice to either design
suitable reference trajectories for robust control strategies or to gain deeper insights into the methodology is to use linearized
relative motion models and a constant density assumption. Following this practice, multiple previous studies developed and
enhanced analytic control algorithms to zero out the relative position and velocity of two spacecraft utilizing differential
aerodynamic forces. Even though aerodynamic lift expands the controllability to all three translational degrees of freedom, it has
been frequently neglected due to the low lift coefficients experienced in orbit so far. And even if considered, analysis has shown that
the feasibility range of using differential lift for the eccentricity control of the in-plane motion is extremely limited compared to
using differential drag. However, two different possible options, namely applying enhanced algorithms to decrease the eccentricity
before even initializing the eccentricity control phase as well as using advanced satellite surface materials targeting specular or
quasi-specular reflection and thereby increasing the differential acceleration forces, have been suggested to increase the feasibility
domain of the methodology. In this paper, the effectiveness of the proposed adjustments is analyzed in a Monte-Carlo approach by
applying them to the analytic control algorithms introduced in literature. Since these are based on the linearized Schweighart-
Sedwick equations, so is the presented analysis. Moreover, additional modifications to the algorithms are made to reduce the
maneuver time of one of the respective control phases. The results show that the analyzed options noticeably improve the success
rates of the maneuvers. Especially the feasibility range of the lift-based control algorithms is enlarged considerably. In addition,
the implemented modifications are shown to consistently reduce the maneuver time of the respective control phase. Consequently,
the presented analysis improves the current state of the art of the analytically designed rendezvous trajectories and provides
valuable new insights into the methodology of using differential aerodynamic forces as a means of relative motion control.
Keywords
Satellite aerodynamic; differential lift; differential drag; formation flight; rendezvous; energy accommodation;

1. INTRODUCTION
In the recent past, there is a tendency of replacing large satellites
with several smaller ones flying in formation due to possible
advantages such as increased flexibility, reduced costs and
enhanced reliability for military, commercial and scientific
missions. An up-to-date review of impending small satellite
formation missions can be found in [1]. In order to maintain the
formation despite present perturbations, to perform formation
reconfiguration or to initiate a rendezvous maneuver, the spacecraft
need to be able to exert control forces. The state-of-the-art solution
to fulfill this requirement is to equip the satellites with chemical
and/or electric propulsion systems. While this method has proven
itself to be successful, it comes with several disadvantages. For
example, equipping a small satellite with a dedicated propulsion
system might cause an exceedance of its limited mass, volume and
power budgets. Therefore, utilizing the residual atmosphere in Low
Earth Orbits (LEO) to generate differential aerodynamic forces
poses a viable propellant-less alternative for relative satellite motion
control. Using aerodynamic forces as a means to perform satellite
rendezvous, an orbital maneuver during which two spacecraft
approach each other, has been subject of a variety of scientific
research publications.
For a rendezvous maneuver to be successful, the relative position
and velocity of the satellites need to be fully zeroed out. Different
methods for achieving rendezvous using differential drag (and later
on differential lift) have been developed and refined by many
authors over the years. An up-to-date review can be found in [2, 3].
The option of using differential drag to perform a rendezvous
maneuver of two spacecraft orbiting in the same orbital plane was
first introduced by Leonard in her Master’s thesis [4]. Horsley et al.
[5, 6] then expanded the control authority to all three translational
degrees of freedom by exploiting differential lift to control out-of-
plane motion. However, due to the low lift coefficients experienced
in orbit so far, the use of differential lift has been frequently
neglected since then. And even if considered, analysis has shown
that the feasibility domain of using differential lift for the
eccentricity control of the in-plane motion is extremely limited
compared to using differential drag [7]. However, there exist
options to increase the success rate of the maneuver sequence as
well as its feasibility range. Two of them, namely:
• Using the sawtooth pattern, initially proposed by Leonard et al.
[4, 8], to decrease the in-plane oscillating motion (later on

*
Corresponding author, st142890@stud.uni-stuttgart.de
 referred to as eccentricity) before initializing the eccentricity 𝑦𝐿𝑉𝐿𝐻
control phase,
• Using advanced satellite surface materials with decreased
levels of energy accommodation, resulting in increased
available differential acceleration2,
𝛽
have been proposed by the authors in a previous publication [2, 3].
This paper investigates the influence of the proposed methods on
the feasibility range via a Monte-Carlo based approach similarly to
the analysis by Smith et al. [7]. Furthermore, additional 𝑦̅
modifications to the algorithms are made to reduce the maneuver
times required for one of the control phases. Consequently, the
major contributions of this paper are threefold:

• Increasing the success rate of the maneuver by implementing 𝑥𝐿𝑉𝐿𝐻

𝑥̅ 𝛼
the sawtooth pattern,
• Analyzing the effect of spacecraft surface materials on the Figure 1: Reference target coordinate system
maneuver time and feasibility range in form of different levels
of energy accommodation, with 𝑥̅ , 𝑦̅, 𝛼 and 𝛽 being defined as:
• Presenting a faster maneuver trajectory for zeroing out the out-
of-plane relative motion. 𝐴 (4)
𝑥̅ = 𝑥̅0 + 𝑎 𝑡
𝜔 𝑦
2. BACKGROUND
𝐴 𝐴𝐵 (5)
2.1. Equations of motion 𝑦̅ = 𝑦̅0 + 𝐵𝜔𝑥̅0 𝑡 − 𝑎𝑥 𝑡 + 𝑎 𝑡²
The equations of motion describing the movement of a deputy 𝜔 2 𝑦
spacecraft (also referred to as chaser) relative to a chief spacecraft
(also referred to as target) utilized within the presented analysis are 𝐴𝑎𝑥 2𝑐 𝐴𝑎𝑥
the Schweighart-Sedwick (SS) equations [9], which include 𝐽2 𝛼 = (𝛼0 − ) cos (√ 𝜔𝑡) +
2𝑐𝜔 2 𝐴 2𝑐𝜔² (6)
perturbation influence of the non-uniform gravity field caused by
the Earth’s oblateness. The SS equations are expressed in the local- 𝛽0 𝐴2 𝑎𝑦 2𝑐
vertical/local-horizontal (LVLH) coordinate system centered at the +( − √
2 ) sin ( 𝐴 𝜔𝑡)
chief. The x-axis points from the Earth’s center to the chief’s center √2𝑐𝐴 2𝜔
of mass. The y-axis is orthogonal to the x-axis and points towards
the direction of motion. The z-axis is orthogonal to the x- and y-
axes and completes the coordinate system following the right-hand 𝛽 𝛽0 𝐴 2 𝑎𝑦 2𝑐 𝐴²𝑎𝑦
=( − √
2 ) cos ( 𝐴 𝜔𝑡) + 2𝜔²
rule [10]. √2𝑐𝐴 √2𝑐𝐴 2𝜔 (7)

In the following, Leonard et al.’s suggested approach of separating 𝐴𝑎𝑥 2𝑐

the chaser’s relative in-plane movement (𝑥, 𝑦) into an average +( − 𝛼0 ) sin (√ 𝜔𝑡)
2𝑐𝜔 2 𝐴
component (𝑥̅ , 𝑦̅) as well as an oscillating component (𝛼, 𝛽) is used.
These are exemplarily visualized in Figure 1, in which the deputy’s
position oscillates around a certain average offset. In its linearized The so-called SS coefficient 𝑐 which takes the 𝐽2 influence into
form, the relative motion in the out-of-plane direction is completely account, the auxiliary variables 𝐴, 𝐵 and 𝐷 and ω, which is the
decoupled from the in-plane motion; a property which is used in the angular velocity of the chief, are defined as follows:
subsequently following maneuver algorithms. For this paper, the
solutions to the SS equations including differential accelerations are
3𝐽2 𝑅𝐸2 (8)
expressed, as suggested by Smith et al. [7]. A slight correction to 𝑐 = √1 + (1 + 3cos(2𝑖𝑐 ))
these equations was made in [2, 3] and the corrected version is used 8𝑟𝑐2
in the following.
2𝑐 2−5𝑐² (9)
𝐴= , 𝐵= and 𝐷 = √3𝑐 2 − 2
𝑥 = 𝑥̅ + 𝛼 (1) 2−𝑐² 2𝑐

𝑦 = 𝑦̅ + 𝛽 (2) µ𝐸 (10)
𝜔=√
𝑎𝑧 𝑧̇0 𝑎𝑧 (3) 𝑟𝑐 3
𝑧 = (𝑧0 − ) cos(𝐷𝜔𝑡) + sin(𝐷𝜔𝑡) + 2 2
𝐷2 𝜔2 𝐷𝜔 𝐷 𝜔 Here 𝑅𝐸 is the Earth’s mean radius, 𝜇𝐸 is its gravitational parameter,
𝑖𝑐 is the chief’s orbital inclination and 𝑟𝑐 its orbital radius. The

2
As the lift forces increase more drastically than the drag forces, the lift
to drag ratio is increased.
 𝑧̇ 0 𝛽0 Table 3 shows the resulting maneuver times from the simulation for
parameters 𝑥̅0 , 𝑦̅0 , 𝑧0 , , 𝛼0 and are the initial conditions at
𝐷𝜔 √2𝑐𝐴 each control phase. The drag-based algorithm was used for Phase
the time 𝑡 = 0, 𝑎𝑥 , 𝑎𝑦 and 𝑎𝑧 are the accelerations of the chaser 2𝑆ℎ𝑎𝑜 . The subphases of phase 3𝑆ℎ𝑎𝑜 are defined as in [7]. The times
relative to the chief in the respective LVLH coordinate directions are very much in accordance with the times published by Smith et
which are induced by the differential aerodynamic drag and lift al. and prove the validity of the implementation.
forces. To do so, each spacecraft is equipped with so-called
aerodynamic plates and is able to control their orientation so that
differential lift and drag accelerations can be created purposefully.
2.3. Collision avoidance
More information can be found in [11]. Shao et al.’s control sequence will inevitably lead to collisions of
the spacecraft, as Smith et al. justly pointed out [7]. Therefore, the
order of the control phases is changed like Smith et al. suggested
2.2. Verification and used as follows throughout the rest of the paper:
The original analytic rendezvous maneuver algorithm is divided
into three separate control phases, as suggested by Horsley et al. [6] 1. Phase 1 zeroes out the average in-plane position (𝑥̅ , 𝑦̅).
and later on adapted by Shao et al. [11] using the SS equations: 2. Phase 2 controls the out-of-plane motion (formerly phase
3𝑆ℎ𝑎𝑜 ). This control phase does not jeopardize the average in-
1. Phase 1𝑆ℎ𝑎𝑜 zeroes out the average in-plane position (𝑥̅ , 𝑦̅) of plane position since the in- and out-of-plane motion is
the deputy satellite with respect to the chief using differential decoupled.
drag (𝑎𝑦 ). 3. Phase 3 deals with the in-plane motion eccentricity in the
2. Phase 2𝑆ℎ𝑎𝑜 zeroes out the in-plane eccentricity (see Equation (𝛼, 𝛽)-plane (formerly phase 2𝑆ℎ𝑎𝑜 ) by either using
(11)) in the (𝛼, 𝛽)-plane by either using differential drag (𝑎𝑦 ) differential drag (𝑎𝑦 ) or differential lift (𝑎𝑥 ).
or differential lift (𝑎𝑥 ).
3. Phase 3𝑆ℎ𝑎𝑜 zeroes out the out-of-plane relative motion (𝑧 and 3. MODIFICATIONS
𝑧̇ ) by applying differential lift (𝑎𝑧 ) in the 𝑧-direction. The algorithms introduced in the previous subchapter lead to
successful rendezvous for a certain range of initial conditions, the
In order to verify the implementation of the algorithm, the maneuver so-called feasibility range. As pointed out by Smith et al. [7],
as published by Shao et al. [11] (for initial conditions and simulation especially the conditions prior to phase 3 are critical for the overall
parameters see Table 1 and Table 2) was re-simulated and the results maneuver success. However, different options to increase the
compared to the ones published in [11] and [7]. success rate of the maneuver sequence as well as its feasibility range
have been proposed by the authors in previous publications [2, 3].
𝒙𝟎 [𝒎] 𝒚𝟎 [𝒎] 𝒛𝟎 [𝒎] 𝒙̇ 𝟎 [𝒎⁄𝒔] 𝒚̇ 𝟎 [𝒎⁄𝒔] 𝒛̇ 𝟎 [𝒎⁄𝒔]
In this paper, the effectiveness of the two options are analyzed via
82.50 -930.46 55.27 -0.17 -0.04 0.29
Monte-Carlo simulations. Firstly, the sawtooth pattern, originally
Table 1: Shao et al.'s [11] initial relative conditions proposed by Leonard et al. [4, 8] with the intention of decreasing
the in-plane eccentricity during phase 1 is implemented. Secondly,
Parameter Value the advantage of using advanced surface materials targeting
𝑟𝑐 6.778137 ∙ 106 𝑚 specular or quasi-specular reflections is analyzed in form of
𝑖𝑐 10° different levels of energy accommodation. In addition, phase 2 is
𝐽2 0.0010826267 [−] modified so that its resulting maneuver time is reduced.
𝑎𝑥,𝑚𝑎𝑥 0.9 ∙ 10−5 𝑚 ⁄ 𝑠²
3.1. Sawtooth algorithm
𝑎𝑦,𝑚𝑎𝑥 4.0 ∙ 10−5 𝑚 ⁄ 𝑠² Differently from the algorithms presented in [6, 11], in which phase
𝑎𝑧,𝑚𝑎𝑥 0.9 ∙ 10−5 𝑚 ⁄ 𝑠² 1 uses the time-optimal solution of regulating a double integrator
along one parabola towards the switch curve3 (SC), the sawtooth
Table 2: Simulation parameters [11] algorithm aims to decrease the in-plane eccentricity in the (𝛼, 𝛽)-
plane by detouring from the time-optimal solution while still
Maneuver time [s]: Shao et al. Smith et This work: eventually zeroing out the average in-plane position (𝑥̅ , 𝑦̅). During
[11]: al. [7]: phase 1, the in-plane eccentricity, defined as:
Phase 1𝑆ℎ𝑎𝑜 10586.72 10586.72 10586.73
Phase 2𝑆ℎ𝑎𝑜 4778.56 Not given 4819.80
𝛽2
Phase 3.1𝑆ℎ𝑎𝑜 1585.91 1544.68 1544.71 𝑒 = √𝛼 2 + (11)
2𝑐𝐴
Phase 1791.66 1791.65 1791.65
3.2𝑆ℎ𝑎𝑜 /3.3𝑆ℎ𝑎𝑜
can be reduced by switching the sign of the acceleration 𝑎𝑦
∑ Phase 3𝑆ℎ𝑎𝑜 55046.23 Not given 55008.06
whenever 𝛼 switches signs, which is twice per orbit. The rate of
Overall maneuver 70414.51 70414.55 70414.59
Table 3: Comparison of control phase times

3
The switch curve consists of two half parabolas leading into origin in
the 𝑥̅ , 𝑦̅-plane. They result from the reaction of the average position,
when a positive or negative drag acceleration 𝑎𝑦 is applied (see [11]).
change of eccentricity can be expressed for the SS equations, as
shown in [2, 3]:
𝑑(𝑒 2 ) √2𝑐𝐴3 (12)
=− ∙ 𝑎𝑦 ∙ 𝛼
𝑑𝑡 𝜔
Equation (12) holds until the (𝛼, 𝛽) position gets too close to the
origin. Hence, the value of the eccentricity is checked whenever 𝛼
switches signs. As soon as the following condition for the
eccentricity (taken from [12]):
𝐴 2
𝑒 < 𝑎𝑦 ∙ ( ) (13)
𝜔
is fulfilled, the algorithm switches to the time-optimal solution and
moves directly towards the SC.
As the main objective of phase 1 still is to zero out the average
position differences, switching the acceleration is only permitted
when it does not counteract the goal of zeroing out the average
position (𝑥̅ , 𝑦̅). This results in the condition that the sawtooth
pattern is only allowed to be initiated when the state (𝑥̅ , 𝑦̅) is
between the switch curves and the 𝑦̅-axis. Otherwise the average
position of the deputy constantly moves away from the chief, as any
trajectory in quadrant II and IV leads away from the origin in 𝑦̅-
direction. In addition, allowing the sawtooth pattern to be initiated
too early (i.e. too close to the 𝑦̅-axis) causes the state to spend most
of the time of the sawtooth pattern in quadrant II or IV, thus moving
away from the origin. This scenario is avoided by adding another
condition to the algorithm, which only allows the sawtooth pattern
1
to be initiated if the deputy spends at least (𝑘 > 1) of its time
𝑘
during the sawtooth pattern within quadrant I or III (and thus
moving towards the origin). Mathematically, this condition can be
described as follows:
1 𝐴𝜋𝑎𝑦.𝑚𝑎𝑥
|𝑥̅ | ≥ ∙ (14)
𝑘 𝜔2
The parameter 𝑘 is user defined and was set to 3/2 throughout this
publication. The combination of these conditions results in
designated areas, the so-called sawtooth areas, in which the
sawtooth pattern is allowed (see Figure 2).
3.2. Reduced levels of energy accommodation
The residual atmosphere above 200 km is so rarefied that it must not
be considered a continuous fluid but a free molecular flow. In this
regime, the forces and torques occurring on a free body are a result
of the energy exchange taking place between the incident gas
particles and the external surfaces. These gas-surface interactions
(GSI) are dominated by the material chemistry, with the
predominant gas species in the Very Low Earth Orbit (VLEO) range
(< 450 km), atomic oxygen, adsorbing to and possibly eroding the
surface. The adsorbed molecule layer is known to increase energy
accommodation and to broaden the angular distribution of
molecules reemitted from the surface [3], explaining the small lift
coefficients experienced in orbit so far. Contrariwise, the magnitude
4
Disruptive Technologies for Very Low Earth Orbit Platforms,
https://discoverer.space/
of the available lift forces could strongly be increased by using
surfaces materials which encourage specular or quasi-specular
reflections, as currently being identified within the Horizons 2020
funded DISCOVERER4 project [13], in which the Institute of Space
Systems (IRS) of the University of Stuttgart is participating.
𝑦̅
1 𝐴𝜋𝑎𝑦.𝑚𝑎𝑥
𝑥̅ = − ∙
𝑘 𝜔2
𝐼𝐼 𝐼
𝐼𝐼𝐼 𝐼𝑉 𝑥̅
1 𝐴𝜋𝑎𝑦.𝑚𝑎𝑥
𝑥̅ = ∙
𝑘 𝜔2
Figure 2: Visualization of sawtooth areas
In a previous publication by the authors [14], the influence of energy
accommodation on a robust spacecraft rendezvous maneuver using
differential aerodynamic forces was analyzed. Energy
accommodation describes how close the kinetic energy of the
incoming molecules is adjusted to the thermal energy of the surface
and is expressed via the dimensionless energy/thermal
accommodation coefficient 𝛼 defined as:
𝐸𝑖 − 𝐸𝑅
𝛼= (15)
𝐸𝑖 − 𝐸𝑊
where 𝐸𝑖 is the kinetic energy of the incident molecules, 𝐸𝑅 is the
kinetic energy of the reemitted molecules and 𝐸𝑊 is the kinetic
energy the reemitted molecules would have if they left the surface
at the surface (wall) temperature. During the full maneuver
sequence, the drag and lift coefficients of the satellites were (for
each time step) dynamically calculated using Sentman’s Gas-
Surface Interaction (GSI) model [15], which assumes that all the
incident particles colliding with the surface are adsorbed and later
reemitted diffusely with a Maxwell–Boltzmann distribution of
velocities. In total, the maneuver sequence was simulated for three
different levels of energy accommodations:
Two constant values of 1 and 0.7 were used in order to simulate the
completely accommodated case as well as to indicate with a
reasonable assumption what might be possible using advanced
surface materials. In a third case, the energy accommodation
coefficient was dynamically calculated using Pillinski et al.’s semi
empirical model for satellite energy accommodation coefficients
(SESAM) [16] resulting in a timely average of 0.91.
In this publication, the influence of the different levels of energy 𝑡𝑖𝑛𝑖𝑡𝑖𝑎𝑙 can be calculated by solving Equation (18), which is
accommodation on the feasibility range as well as the maneuver established from the derivative of Equation (3):6
time is analyzed by using the average differential drag and lift
values published in [14] as the available (constant) differential
acceleration values for the analytic maneuver algorithm. For the 𝑎𝑧,𝑚𝑎𝑥 ∙ 𝑠𝑖𝑔𝑛(𝑧̇2.1,𝑚 )
sake of comparability, in both cases the orbit parameters stated in (𝑧2.1,𝑚 + ) ∙ sin(𝐷𝜔𝑡𝑖𝑛𝑖𝑡𝑖𝑎𝑙 ) (18)
Table 2 were used. The values for the three different cases can be (𝐷𝜔)2
found in the following Table 4. For additional information the 𝑧̇2.1,𝑚
= ∙ cos(𝐷𝜔𝑡𝑖𝑛𝑖𝑡𝑖𝑎𝑙 )
reader is referred to the original publication [14]. 𝐷𝜔

label 𝒂𝒚,𝒎𝒂𝒙 𝒂𝒙,𝒎𝒂𝒙 = 𝒂𝒛,𝒎𝒂𝒙 The radius 𝑅2.1,𝑚 of the circle, on which the state will initially move
𝑧̇
𝛼=1 high −5
5.7 ∙ 10 m/s² 4.94 ∙ 10−6 m/s² along in the (𝑧, )-plane, can be calculated by Equation (19):
−5 𝐷𝜔
𝛼 = 0.91 medium 6.6 ∙ 10 m/s² 1.18 ∙ 10−5 m/s²
−5
𝛼 = 0.7 low 7.4 ∙ 10 m/s² 2.19 ∙ 10−5 m/s²
Table 4: Differential acceleration values for the three different 2 𝑧̇2.1,𝑚 2 (19)
𝑅2.1,𝑚 = √(𝑧2.1,𝑚 + 𝑠𝑖𝑔𝑛(𝑧̇2.1,𝑚 ) ∙ 𝑘𝑧 ) + ( )
levels of energy accommodation 𝐷𝜔

3.3. Out-of-plane algorithm modification As the time for completing an (half-)orbit is well-known, 𝑡𝑜𝑟𝑏𝑖𝑡𝑠 can
In this subchapter, a modification to Horsley et al.’s [6] algorithm be calculated by determining how many orbits are performed,
to zero out the out-of-plane relative motion is presented. Still, it is before |𝑧| ≤ 2𝑘𝑧 (when 𝑧̇ = 0). The out-of-plane eccentricity is
based on the same ideas and mechanics used by Horsley et al. reduced by 2𝑘𝑧 per half-orbit by the switching algorithm, thus the
Readers are advised to familiarize themselves with Horsley et al.’s number of orbits 𝑛𝑧 amounts to:
algorithm [6] in order to better understand the following
elucidations.
𝑅2.1,𝑚 − 3𝑘𝑧 (20)
𝑛𝑧 = ⌈ ⌉⁄2
2𝑘𝑧
Generally, the required sign of the differential acceleration 𝑎𝑧 to
reduce the out-of-plane eccentricity defined as: Therefore:
2𝑛𝑧 𝜋 (21)
𝑡𝑜𝑟𝑏𝑖𝑡𝑠 =
𝐷𝜔
2
𝑧̇ (16)
𝑒𝑧 = √𝑧 2 + ( )
𝐷𝜔 If 𝑛𝑧 < 0, which occurs if the initial conditions (𝑧2.1,𝑚 , 𝑧̇2.1,𝑚 ) are
close to the origin, subphase 2.1𝑚 is skipped. In this case: 𝑡2.1,𝑚 =
is dictated by the sign of 𝑧̇ . Therefore, in the modified out-of-plane 𝑡𝑖𝑛𝑖𝑡𝑖𝑎𝑙 = 𝑡𝑜𝑟𝑏𝑖𝑡𝑠 = 0.
algorithm the eccentricity reducing algorithm starts immediately by
applying a positive 𝑎𝑧 if 𝑧̇ < 0 and a negative 𝑎𝑧 if 𝑧̇ > 0. It keeps 3.3.2. Subphase 𝟐. 𝟐𝒎
switching the signs of the lift acceleration whenever 𝑧̇ switches
In subphase 2.2𝑚 the state moves from its position (𝑧2.2,𝑚 , 0) on the
signs until |𝑧| ≤ 2𝑘𝑧 (at 𝑡2.1,𝑚 ).5 This is when the acceleration 𝑎𝑧 is
z-axis until it intersects one of the semicircles passing through the
kept constant for 𝑡2.2,𝑚 seconds until the state intersects one of the origin. In order to express with which of the two semicircles the
𝑧̇
two semicircles leading into origin in the (𝑧, )-plane. Then the intersection occurs, it is defined:
𝐷𝜔
sign is switched one last time to follow the semicircle for 𝑡2.3,𝑚
seconds into the origin, where the acceleration is set to zero. An 𝑠 = (−1)2𝑛𝑧 +1 ∙ 𝑠𝑖𝑔𝑛(𝑧̇2.1,𝑚 ) (22)
example of this algorithm can be seen in Figure 3.
The intersection occurs in the position (𝑧2.3,𝑚 , 𝑧̇2.3,𝑚 ). Both
The required times can be calculated beforehand via the equations positions can be calculated from the initial position of subphase
presented in the following subchapters: 2.1𝑚 :

𝑧2.2,𝑚 = (𝑘𝑧 − (𝑅2.1,𝑚 − 4𝑛𝑧 𝑘𝑧 )) ∙ 𝑠 (23)

3.3.1. Subphase 𝟐. 𝟏𝒎
The first subphase 2.1𝑚 of the algorithm consists of the state 2
moving from its initial position (𝑧2.1,𝑚 , 𝑧̇2.1,𝑚 ) to the z-axis and then 𝑘𝑧2 − (𝑅2.1,𝑚 − 4𝑛𝑧 𝑘𝑧 ) (24)
𝑧2.3,𝑚 =
performing 𝑛𝑧 orbits until |𝑧| ≤ 2𝑘𝑧 (when 𝑧̇ = 0), therefore the 4𝑘𝑧 𝑠
time required 𝑡2.1,𝑚 can be expressed as:
2 (25)
𝑧̇2.3,𝑚 = 𝑠𝐷𝜔 ∙ √𝑘𝑧2 − (𝑧2.3,𝑚 + 𝑘𝑧 𝑠)
𝑡2.1,𝑚 = 𝑡𝑖𝑛𝑖𝑡𝑖𝑎𝑙 + 𝑡𝑜𝑟𝑏𝑖𝑡𝑠 (17)

|𝑎𝑧,𝑚𝑎𝑥 | 6
5
With 𝑘𝑧 = The signum function as used throughout this paper is defined as:
(𝜔𝐷)2 𝑥
𝑠𝑖𝑔𝑛(𝑥) ∶= |𝑥|
Figure 3: Example case for modified control phase 2: Out-of-plane trajectory (left) and commanded control (right)

To reach the position (𝑧2.3,𝑚 , 𝑧̇2.3,𝑚 ) it takes the time 𝑡2.2,𝑚 , which 1 and lift for phase 2), phase 3 can be controlled via either lift or
is calculated by solving the following Equation (26): drag. Therefore, two different feasibility ranges need to be
distinguished: one in which the maneuver is feasible using drag and
𝑠𝑎𝑦,𝑚𝑎𝑥 a second in which the maneuver is feasible using lift.
𝑧2.3,𝑚 = (𝑧2.2,𝑚 − ) ∙ 𝑐𝑜𝑠(𝐷𝜔𝑡2.2,𝑚 )
(𝐷𝜔)2 (26)
𝑠𝑎𝑦,𝑚𝑎𝑥 In this chapter, the influences of the proposed modifications on the
+
(𝐷𝜔)2 feasibility range are analyzed via a MATLAB based Monte-Carlo
simulation approach based on the one proposed by Smith et al. [7].
3.3.3. Subphase 𝟐. 𝟑𝒎
In subphase 2.3𝑚 of the modified algorithm, the state is controlled Each simulation case consists of 10,000 individual rendezvous
from the position (𝑧2.3,𝑚 , 𝑧̇2.3,𝑚 ) into origin along the semicircle. maneuvers (referred to as runs) for randomly generated initial
This requires the time 𝑡2.3,𝑚 , which is calculated by solving the conditions within the range given in Table 5:
following Equation (27):
̅𝟎 [𝒎]
𝒙 ̅𝟎 [𝒎]
𝒚 𝒛𝟎 [𝒎] 𝜶𝟎 [𝒎] 𝜷𝟎 [𝒎] 𝒛̇ 𝟎 [𝒎⁄𝒔]
𝑠𝑎𝑦,𝑚𝑎𝑥 𝑠𝑎𝑦,𝑚𝑎𝑥 [±1800] [±4000] [±250] [±1000] [±1500] [±0.25]
= (𝑧2.3,𝑚 − ) ∙ 𝑐𝑜𝑠(𝐷𝜔𝑡2.3,𝑚 )
(𝐷𝜔)2 (𝐷𝜔)2 (27) Table 5: Range of initial conditions
𝑧̇2.3,𝑚
+ ∙ 𝑠𝑖𝑛(𝐷𝜔𝑡2.2,𝑚 )
𝐷𝜔 For each run, three different outcomes are possible:
• The rendezvous maneuver is not successful. This case is
The overall phase 2 maneuver time using the modified algorithm indicated by a red dot.
amounts to: • The rendezvous maneuver is feasible using differential drag
but not using differential lift. This case is indicated by a green
𝑡2,𝑚 = 𝑡2.1,𝑚 + 𝑡2.2𝑚 + 𝑡2.3,𝑚 (28) dot.
• The rendezvous maneuver is feasible for both, differential drag
and lift. This case is indicated by a blue dot.
3.3.4. Example case
An example case for the initial conditions of 𝑥̅ = 𝑦̅ = 0𝑚, 𝛼 = From the large quantity of results, deep insights concerning the
40𝑚, 𝛽 = −50𝑚, 𝑧 = −15𝑚, 𝑧̇ = 0.067 𝑚⁄𝑠 and the acceleration feasibility of rendezvous using only aerodynamic forces can be
value given by Shao et al. (see Table 2) can be seen in Figure 3. The derived. Within the course of this paper, three different performance
respective times result to: 𝑡𝑖𝑛𝑖𝑡𝑖𝑎𝑙 = 1504.2𝑠, 𝑡2.1,𝑚 = 9817.54𝑠, indicators are used to compare the influence of the modifications:
𝑡2.2,𝑚 = 338.71𝑠, 𝑡2.3,𝑚 = 1685.32𝑠 and therefore 𝑡2,𝑚 =
11841.57𝑠. 1. The success rates of the maneuver, which is defined as the ratio
of the number of successful maneuvers over the total simulated
4. RESULTS maneuver number and expressed as a percentage.
As stated by Smith et al. [7], the critical phase for the overall 2. The feasibility ranges for drag and for lift, which already have
maneuver success is phase 3, where the algorithm fails if the initial been introduced beforehand.
𝛽 3. The required maneuver times for phase 2 and 3.
(𝛼, )-position is too far away from the origin. Consequently,
√2𝑐𝐴
the terminology feasibility range, as it is used throughout this Since it represents the state-of-the-art in terms of maneuver
publication, refers to the initial condition range of phase 3 for which algorithm and satellite surfaces, the case assuming full
the algorithm presented by Shao et al. [11] leads to a successful accommodation (𝛼 = 1) using the time-optimal phase 1 algorithm
rendezvous. Different from both other phases, for which only one is taken as a reference. All other analyzed cases are subsequently
differential aerodynamic force direction can be used (drag for phase
compared relative to the reference case. Throughout this thesis, the
modified algorithm for phase 2 is used.
4.1. Reference case
In order to be able to analyze the influence of the modifications
made to the algorithms, the initial conditions at the beginning of
control phase 1 are depicted in Figure 4 and Figure 5.
Figure 4: Reference case: Initial (𝒙 ̅)-conditions
̅, 𝒚
The green points in Figure 4 are evenly distributed, indicating no
interdependence of initial (𝑥̅ , 𝑦̅)-conditions and rendezvous
feasibility with the used algorithms. The rarity of blue points
suggests a low success rate for the lift-based phase 3 maneuver. The
SC is depicted as a black dashed line.
𝜷
Figure 5: Reference case: Initial (𝜶, )-conditions
√𝟐𝒄𝑨
In Figure 5 the green points accumulate around the origin of the
𝛽
(𝛼, )-plane. This suggests a higher likeliness of feasible
√2𝑐𝐴
𝛽
rendezvous for small initial (𝛼, )-conditions.
√2𝑐𝐴
7 The color scale to encode the times stays constant throughout
this publication, in order to make the results comparable.
Maneuver success
The maneuver algorithm was successful in 23.06% of all runs when
using the drag-based and 0.05% when using the lift-based phase 3
algorithm.
Feasibility range
The feasibility range for both lift and drag are displayed in Figure 6
and Figure 8. The green area inside of which all drag-based
maneuvers lead to successful rendezvous is a circle with a radius of
about 465 meters. As the lack of sufficing blue points prevents a
precise estimation of the size of the feasibility range in case of lift,
the simulation was repeated with initial conditions closer to the
origin. The results of this simulation are only used for the
visualization of the feasibility range in Figure 8 and are not further
taken into account for the analysis. The blue area is determined to
be an ellipse with semi-axes of about 20×18 meters.
Maneuver time phase 2
The required time to execute phase 2 strongly depends on the initial
out-of-plane position of phase 2, as shown in Figure 7 by encoding
the time with colors.7 In 100% of the runs the modified phase 2
algorithm is faster than the original phase 2 algorithm suggested by
Horsley et al. [6]. On average, the modified maneuver is 3.457%
faster than the one originally proposed.
Maneuver time phase 3
In those cases in which both lift- and drag-based phase 3 control
laws lead to successful rendezvous, the lift-based maneuver is faster
in 0 out of 5 times and on average is slower by 82.95%.
Figure 6: Reference case: Phase 3 drag feasibility range
 𝒛̇
Figure 7: Reference case: Phase 2 time for different (𝒛, )-conditions at the beginning of phase 2
𝑫𝝎

𝛽
spread out meaningfully over the (𝛼, )-plane. This suggests a
√2𝑐𝐴
less strict interdependence of feasible rendezvous and initial
𝛽
(𝛼, )-conditions.
√2𝑐𝐴

Figure 8: Reference case: Phase 3 lift feasibility range

4.2. Reference case with sawtooth

In the second simulation case, the sawtooth algorithm was added to
phase 1, whereas the energy accommodation coefficient and
consequently also the resulting relative accelerations, remained
unchanged.
Maneuver success
The rendezvous maneuver was successful in 70.20% of the
iterations when using the drag-based and 0.27% when using the lift-
based phase 3 algorithm. The initial conditions at the beginning of
control phase 1 are depicted in both in-plane systems in Figure 9
and Figure 10.
Differently to the reference case without applying sawtooth, the
green points in Figure 9 are not evenly distributed in the (𝑥̅ , 𝑦̅)-plane
anymore. A feasible maneuver becomes more likely, the further the
initial (𝑥̅ , 𝑦̅)-conditions are away from the SC. The small number of
blue points suggests a low success rate for the lift-based phase 3
maneuver; still the number of blue points slightly increases
compared to the reference case without sawtooth. In Figure 10 the
green points, while still concentrated towards the origin, have
Figure 9: Reference case with sawtooth: Initial (𝒙 ̅)-conditions
̅, 𝒚
Feasibility range
Due to the control accelerations staying constant, the diagram
depicting the feasibility ranges is very similar to Figure 6 and is not
included. The aforementioned dimensions of the feasibility ranges
for 𝛼 = 1 are confirmed.
Maneuver time phase 2
For the same reason, the figure showing the resulting phase 2 times
is very similar to Figure 7 and is not included. Again, the modified
algorithm was faster for 100% of the runs; this time by 3.43% on
average.
Maneuver time phase 3
In those cases, in which both options lead to successful rendezvous
the lift-based maneuver is faster in 3 out of 27 times, but on average
it is slower than the drag-based one by 89.52%.
 𝜷 Figure 11: 𝜶 = 𝟎. 𝟗𝟏: Phase 3 drag feasibility range
Figure 10: Ref. case with sawtooth: Initial (𝜶, )-conditions
√𝟐𝒄𝑨

4.3. Medium levels of energy accommodation

In the following simulation cases, the values of the aerodynamic
acceleration are increased as a consequence of reduced levels of
energy accommodation (𝛼 = 0.91).

4.3.1. Without sawtooth

The time-optimal phase 1 algorithm is used.

Maneuver success
The rendezvous maneuver was successful in 30.33% of the runs
when using the drag-based and 0.2705% of the runs when using the
lift-based phase 3 algorithm. The figures showing the feasibility of
the maneuver depending on the initial conditions at the beginning
of control phase 1 are not included as, besides a slightly increased
number of green and blue dots, no new insight is offered.

Feasibility range
The feasibility ranges are shown in Figure 11. The green area inside
of which all drag-based maneuvers lead to successful rendezvous is
a circle with a radius of about 540 meters. The dimensions of the
blue area inside of which all lift-based maneuvers lead to successful
rendezvous cannot precisely be identified due to an insufficient
number of blue points; this will be done in the next subchapter.
Maneuver time phase 2
The modified phase 2 algorithm is always faster than the original
one; on average by 7.494%. The required time to execute phase 2
𝑧̇
depending on the position (𝑧, ) at the beginning of phase 2 is
𝐷𝜔
shown in Figure 12.
Maneuver time phase 3
In those cases, in which both lift- and drag-based phase 3 control
laws lead to successful rendezvous the lift-based maneuver is faster
in 2 out of 27 times, but on average it is slower by 47.44%.
Figure 12: 𝜶 = 𝟎. 𝟗𝟏: Required phase 2 time
4.3.2. With sawtooth
In this simulation case, phase 1 is executed with the sawtooth
algorithm.
Maneuver success
The rendezvous maneuver was successful in 69.89% of the
iterations when using the drag-based and 0.7807% of the iterations
when using the lift-based phase 3 algorithm. Again, the initial
conditions at the beginning of control phase 1 are not depicted as,
besides a slightly increased number of green and blue dots, no new
insight is offered.
Feasibility range
The feasibility range of the drag-based maneuver is confirmed to
have a radius of about 540 meters for this level of energy
accommodation. The feasibility range for lift base maneuvers is
depicted in Figure 13. It is an ellipse with semi-axes of
approximately 48×45 meters.
 𝑧̇
execute phase 2 depending on the position (𝑧, ) at the beginning
𝐷𝜔
of phase 2 is shown in Figure 15.

Figure 13: 𝜶 = 𝟎. 𝟗𝟏: Phase 3 lift feasibility range

Maneuver time phase 2

Again, the modified phase 2 algorithm is faster in 100% of the runs. Figure 14: 𝜶 = 𝟎. 𝟕: Phase 3 drag feasibility range
On average it is faster by 7.536%. The diagram showing the phase
2 times is not included, as it is very similar to Figure 12 and offers
no new insights.

Maneuver time phase 3

In those cases, in which both options lead to successful rendezvous,
the lift-based maneuver is faster than the drag-based one in 6 out of
78 times but on average is slower by 77.87%.

4.4. Low levels of energy accommodation

For these simulation cases, the values of the aerodynamic
acceleration are increased once more as a consequence of the now
low levels of energy accommodation (𝛼 = 0.7).

4.4.1. Without sawtooth

The time-optimal phase 1 algorithm is used.

Maneuver success
The rendezvous maneuver was successful in 37.24% of the
iterations when using the drag-based and 0.6637% of the iterations
when using the lift-based phase 3 algorithm. Once again, the figures
showing the feasibility of the maneuver depending on the initial
conditions at the beginning of control phase 1 are not included as,
besides a furtherly increased number of green and blue dots, no new
insight is offered.
Figure 15: 𝜶 = 𝟎. 𝟕: Required phase 2 time
Maneuver time phase 3
In those cases, in which both lift- and drag-based phase 3 control
laws lead to successful rendezvous the lift-based maneuver is faster
in 6 out of 66 times, but on average it is slower by 42.13%.

Feasibility range 4.4.2. With sawtooth

The feasibility range for both drag and lift is shown in Figure 14.
The green area inside of which all drag-based maneuvers lead to
successful rendezvous is a circle with a radius of about 605 meters.
The dimensions of the blue area inside of which all lift-based
maneuvers lead to successful rendezvous cannot precisely be
identified because of a lack of enough blue points; this will be done
in the next subchapter.
Maneuver time phase 2
In 100% of the iterations the modified phase 2 algorithm is faster
than the original one; on average by 12.60%. The required time to
In the last simulation case, phase 1 is executed with the sawtooth
algorithm.
Maneuver success
The rendezvous maneuver was successful in 70.39% of the
iterations when using the drag-based and 2.380% of the iterations
when using the lift-based phase 3 algorithm. Again, the initial
conditions at the beginning of control phase 1 are not depicted as,
besides furtherly increased number of green and blue dots, no new
insight is offered.
 sizes for the feasibility ranges could be identified and their
implications are discussed in the following.

5.1.1. Feasibility range of lift

The lift-based phase 3 algorithm will lead to rendezvous if and only
𝛽
if the initial (𝛼, )-position at the beginning of phase 3 is inside
√2𝑐𝐴
of the blue ellipses depicted in Figure 8, Figure 13 and Figure 16. If
this is the case, the eccentricity is small enough to be zeroed out by
the lift-based algorithm. The dimensions of said ellipses directly
depend on the magnitude of the maximally achievable differential
lift acceleration 𝑎𝑥,𝑚𝑎𝑥 = 𝑎𝑧,𝑚𝑎𝑥 , which is why the semi-major and
semi-minor axes increase as the energy accommodation coefficient
is reduced in chapters 4.3 and 4.4 compared to the reference case. A
comparison of the increase in the size of the semi-major axes from
the reference case to chapters 4.3 (and 4.4 respectively) with the
increase of the maximum lift accelerations suggests an
approximately linear correlation: As the acceleration increases by
Figure 16: 𝜶 = 𝟎. 𝟕: Phase 3 lift feasibility range 138.9% (343.3%), the semi-major axis approximately increases by
140% (330%). The following table summarizes the main results on
Feasibility range the feasibility range of using lift:
The feasibility range of drag is confirmed to be about 605 meters
for this level of energy accommodation. The feasibility range for lift
Energy Differential lift Feasibility range
is depicted in Figure 16. It is an ellipse with semi-axes of accommodation acceleration using lift
approximately 86×81 meters. coefficient
𝛼=1 4.94 ∙ 10−6 m/s² 20×18 m
Maneuver time phase 2 𝛼 = 0.91 1.18 ∙ 10−5 m/s² 48×45 m
The modified phase 2 algorithm is faster 100% of the runs. On 𝛼 = 0.7 2.19 ∙ 10−5 m/s² 86×81 m
average it is faster by 12.63%. The diagram showing the phase 2 Table 6: Increase of feasibility range using lift for different
times is not included, as it is very similar to Figure 15 and offers no
levels of energy accommodation
new insights.
5.1.2. Feasibility range of drag
Maneuver time phase 3 The drag-based phase 3 algorithm will be successful only for all
In those cases, in which both options lead to successful rendezvous, 𝛽
the lift-based maneuver is quicker than the drag-based one in 19 out initial phase 3 (𝛼, )-positions inside of the green circles
√2𝑐𝐴
of 237 times but on average is slower by 55.96%. (including the blue ellipses) depicted in Figure 6, Figure 11 and
Figure 14. Inside this area, the eccentricity is small enough to be
5. DISCUSSION zeroed out by the drag-based algorithm. The radius of these circles
Before the influences of the individual modifications are discussed also increases with reduced levels of energy accommodation as the
in detail, general outcomes of the results presented can be value of the maximally achievable differential drag acceleration
summarized: 𝑎𝑦,𝑚𝑎𝑥 increases in chapters 4.3 and 4.4 compared to the reference
case. A comparison of the increase in the size of the radius from the
1. Applying the sawtooth pattern during phase 1 leads to a reference case to chapters 4.3 (and 4.4 respectively) with the
significantly increased maneuver success since more initial increase of the maximum drag acceleration suggests an
conditions before phase 3 lie within the respective feasibility approximately linear correlation: As the acceleration increases by
ranges. The size of the feasibility ranges, however, is not 15.79% (29.82%), the radius approximately increases by 16.13%
influenced by the sawtooth pattern. (30.11%).
2. Reducing the energy accommodation coefficient of the
satellite’s surface materials considerably enlarges the Energy Differential drag Feasibility range
feasibility domain, especially in the case of lift. Consequently, accommodation acceleration using drag
coefficient
the maneuver success rates are increased, as well. In addition,
the phase 2 maneuver time to zero out the out-of-plane relative 𝛼=1 5.7 ∙ 10−5 m/s² 465×465 m
motion is reduced significantly. 𝛼 = 0.91 6.6 ∙ 10−5 m/s² 540×540 m
3. The proposed modified control algorithm for phase 2 was able 𝛼 = 0.7 7.4 ∙ 10−5 m/s² 605×605 m
to reduce the maneuver time compared to the originally Table 7: Increase of feasibility range using drag for different
proposed algorithm in all analyzed cases. levels of energy accommodation

5.1. Feasibility range

The size of the feasibility ranges strongly depends on the level of
energy accommodation, especially in the case of lift. Very distinct
5.2. Influence of the sawtooth pattern (Table 4). As a consequence, the required times for the second
The sawtooth pattern is not able to modify the feasibility ranges. maneuver phase strongly reduces, too. Whereas differential
However, it raises the number of initial conditions of phase 3 to lie aerodynamic forces are most certainly not the method of choice for
within the feasible range and therefore considerably increases the time critical maneuvers, shorter maneuver times are beneficial as
maneuver success. Differently from the time-optimal phase 1 thereby the unwanted orbital decay can be reduced. The following
control law, where neither the initial (𝑥̅ , 𝑦̅)-position nor the initial table sums up the average times of all runs for all different cases
𝑧̇
(𝑧, )-position have any direct effect on the maneuver success, a using the modified phase 2 maneuver algorithm:
𝐷𝜔
dependence of the rendezvous success on the initial (𝑥̅ , 𝑦̅)-position Energy Maneuver time phase 2
could be revealed (see Figure 9): If the initial (𝑥̅ , 𝑦̅)-position is in accommodation
an area too close to the SC (black dashed line), the state will Without sawtooth With sawtooth
coefficient
intersect the SC before the sawtooth pattern is initiated. This results 𝛼=1 65451.96s 64960.50s
in the time-optimal algorithm, which does not actively reduce the 𝛼 = 0.91 27540.71s 27306.87s
eccentricity. Consequently, the red dots accumulate close to the SC. 𝛼 = 0.7 14954.04s 14935.75s
If the distance between the initial (𝑥̅ , 𝑦̅)-position and the switch Table 10: Average phase 2 maneuver time for all analyzed cases
curve is large enough for the sawtooth pattern to be initiated, the
chance for a successful maneuver increases. In these cases, the Finding an explanation as to why the phase 2 times slightly decrease
eccentricity is reduced sufficiently during phase 1 so that the when the sawtooth algorithm is used is left for future work.
𝛽
(𝛼, )-position at the beginning of phase 3 is inside the
√2𝑐𝐴
aforementioned feasibility range. Consequently, the success rates of 5.4. Phase 3 maneuver time
the maneuver increase meaningfully when the sawtooth algorithm For those cases, in which both the drag- and the lift-based
is used. The main results are summarized in the following table: algorithms lead to successful rendezvous, the maneuver times
required to complete phase 3 are compared:8 For 𝛼 = 1, the lift-
Energy Maneuver success Maneuver success based maneuver is faster in 7.08% of the cases, but on average the
accommodation without sawtooth [%] with sawtooth [%] drag-based maneuvers are faster by 49.02%.
coefficient Drag Lift Drag Lift
𝛼=1 23.06 0.05 70.20 0.27 Due to the resulting increased differential lift accelerations, reduced
𝛼 = 0.91 30.33 0.2705 69.89 0.7807 levels of energy accommodation lead to shorter maneuver times for
𝛼 = 0.7 37.24 0.6637 70.39 2.38 lift-based maneuvers: For 𝛼 = 0.91, the lift-based maneuver is
Table 8: Maneuver success for all analyzed cases faster in 7.619% of the cases, and the drag-based maneuvers are
faster by 41.19% on average. For 𝛼 = 0.7, the lift-based maneuver
Noticeably, the trade-off between the effectivity of the eccentricity is faster in 8.251% of the cases, and the drag-based maneuvers are
reduction and the increase of maneuver time during phase one is faster on average only by 34.62%.
controllable by the parameter 𝑘 introduced in subchapter 3.1. As a
consequence, the results from Table 8 are only valid for 𝑘 = 3/2. Energy Ratio of lift-based Average time
accommodation phase 3 maneuvers benefits using drag
5.3. Phase 2 maneuver time coefficient faster than drag-based for phase 3
5.3.1. Benefits of the modifications 𝛼=1 7.08% 49.02%
In all simulation cases and regardless of the initial conditions, the 𝛼 = 0.91 7.619% 41.19%
modified phase 2 maneuver algorithm reduced the maneuver time 𝛼 = 0.7 8.251% 34.62%
compared to using the original algorithm as proposed by Horsley et Table 11: Comparison of phase 3 maneuver time using lift and
al. [6]. In addition, this time reduction increases with the available drag
lift acceleration:
Based on the gained insights, no general assertion on which
Energy Differential lift Average time algorithm leads to shorter maneuver times can be given, as this is
accommodation acceleration reduction phase heavily dependent on the conditions at the beginning of phase 3. On
coefficient two average, the drag-based maneuver is faster, but especially with
𝛼=1 4.94 ∙ 10−6 m/s² 3.443% increased lift values, there is a non-negligible amount of cases, in
𝛼 = 0.91 1.18 ∙ 10−5 m/s² 7.515% which the lift-based maneuver leads to faster rendezvous.
𝛼 = 0.7 2.19 ∙ 10−5 m/s² 12.61% Consequently, the decision of which algorithm is used should be
Table 9: Average phase 2 time reduction by modified algorithm made on a case-to-case basis.

5.3.2. Benefits of reduced levels of energy 5.5. Comparison to literature

accommodation The feasibility ranges explained in 5.1 confirm the results presented
Whereas the differential drag accelerations only increase by Smith et al. [7]. Even though being simulated with a decreased
moderately for lower levels of energy accommodation (see Table orbital radius of the chief of 𝑟𝑐 = 6678.1363𝑘𝑚, an increased
4), the increase of the differential lift accelerations is enormous inclination of 𝑖𝑐 of 30° and slightly different differential

8 4.1), which was used for the visualization in Figure 8, are included here for
Because of the low number when the lift-based algorithm was feasible in
the reference cases, the results from the additional simulation (see chapter increased validity.
accelerations of 𝑎𝑥 = 𝑎𝑧 = 1 ∙ 10−5 𝑚/𝑠² and 𝑎𝑦 = 5 ∙ 10−5 𝑚/
𝑠², with a stated radius of the feasibility ranges of 35𝑚 for lift as
well as 396𝑚 for drag, the results closely match the results
presented herein (feasibility radius for lift of 46𝑚 with a lift
acceleration of 1.18 ∙ 10−5 𝑚/𝑠² and for drag of 465𝑚 with a drag
acceleration of 5.7 ∙ 10−5 𝑚/𝑠²).
Furthermore, Smith et al.’s assertion on the time required for phase
3 were confirmed: The lift-based algorithm is only faster for a
limited domain of initial phase 3 conditions and in the vast majority
of cases, the drag-based algorithm leads to decreased maneuver
times for phase 3. As the total time for the first two phases is the
same in either case, it is the maneuver time of phase 3 which
ultimately decides which maneuver sequence leads to the shortest
overall maneuver time.
5.6. Using differential lift or drag? A critical discussion
The presented analysis aimed to validate originally proposed
methods to increase the feasibility range and the maneuver success
rates of differential lift-based rendezvous maneuvers, both of which
was successful: the semi-axes of the feasibility domain of using lift
could be increased by more than a factor of 4 compared to the
reference case by using satellite surfaces with reduced levels of
energy accommodation. By additionally adding the sawtooth
pattern to phase 1, the maneuver success could even be increased by
a factor of 47.
However, even with the performed modifications, the statements of
Smith et al. [7] according to which differential lift for the in-plane
relative motion control of a rendezvous scenario is an inferior
approach compared to differential drag, still holds since in the best
case:
• The ratio of the semi-axes of the feasibility range for lift over
drag is 1/7 (see Table 6 and Table 7),
• The ratio of the maneuver success for lift over drag is 1/30 (see
Table 8),
• The time for phase 3 using lift is only faster in 8.351% of the
cases but in average longer by 34.62% (see Table 10).
In addition, possible downsides of the implemented strategies need
to be discussed, too. The most vital one is the increased orbital decay
due to the sawtooth pattern: The analytically designed trajectories
are of bang-bang nature, meaning that either one of the satellites
consistently maximizes its area perpendicular to the flow. As a
consequence, the orbital decay during the maneuver is maximized,
too. If, as being the case for the sawtooth pattern, the maneuver time
is drastically increased, so is the resulting orbital decay. Future work
will be targeting to include a parameter expressing the resulting
orbital decay of the maneuver in the overall analysis.
5.7. Topics for future work
Besides the just discussed inclusion of a parameter expressing the
resulting orbital decay, two other possible topics for future work are
planned:
1. Whether a maneuver is successful or not was, throughout the
presented analysis, fully dependent on the maneuver algorithm
developed by Horsley et al. [6] and Shao et al. [11],
respectively. However, new control algorithms which bypass
this ‘non-feasible’ case by e.g. simply repeating phase 3 until
rendezvous is accomplished, can be developed.
2. This paper proved that the modified phase 2 algorithm
presented leads to shorter maneuver times for its respective
phase. Whether or not this results in an overall faster
rendezvous maneuver, however, still needs to be answered.
6. CONCLUSION
In the presented analysis, improvements to the feasibility ranges and
success rates of differential lift- and drag-based rendezvous
maneuver algorithms presented in literature were made by
analyzing the effects of two modifications, namely:
• Using the sawtooth pattern, initially proposed by Leonard et al.
[4, 8], to decrease the eccentricity before initializing the
eccentricity control phase,
• Using advanced satellite surface materials with decreased
levels of energy accommodation, resulting in increased
available differential lift forces.
In addition, a modified algorithm to zero out the out-of-plane
motion (phase 2) is presented.
The analysis shows that implementing sawtooth significantly
increases the maneuver success but does not influence the feasibility
range. Reduced levels of energy accommodation, on the other hand,
are able to significantly increase the feasibility range and with it the
resulting maneuver success. The highest success rate is reached
when both modifications are applied simultaneously: in this case,
the success rates are increased by a factor of 3.05 for the drag-based
maneuver and 47.6 for the lift-based maneuver with respect to a
reference case. In addition, the ratio of runs in which the lift-based
maneuver leads to a shorter maneuver time could be increased while
the average time benefit of using drag was reduced.
However, even with the performed modifications, the statements of
Smith et al. [7], according to which differential lift for the in-plane
relative motion control of a rendezvous scenario is an inferior
approach compared to differential drag, still holds: differential drag
is the method of choice to control all sections of the in-plane relative
motion, due to:
1. Shorter resulting maneuver times,
2. Higher maneuver success rates,
3. Larger feasibility range.
However, the reduced levels of energy accommodation strongly
increase the control authority of differential lift, which drastically
reduces the required time for the out-of-plane formation control and
consequently the resulting orbital decay during the maneuver. These
results are very much in accordance with the results presented in
[14].
The time for the out-of-plane control phase could consistently be
decreased by applying the modified algorithm. With the reference
lift acceleration value, the phase 2 time was decreased by 3.443%.
This effect was amplified for medium (low) levels of energy
accommodation where a time reduction of 7.515% (12.61%) could
be achieved.
Consequently, the presented analysis improves the current state of 12. Bevilacqua, R., Romano, M.: Rendezvous Maneuvers of
the art of analytically designed rendezvous trajectories and provides Multiple Spacecraft Using Differential Drag Under J2
valuable new insights into the methodology of using differential Perturbation. Journal of Guidance, Control, and Dynamics
aerodynamic forces as a means of relative motion control. 31(6), 1595–1607 (2008)
13. Roberts, P.C.E., Crisp, N.H., Edmondson, S., Haigh, S.J.,
References Lyons, R.E., Oiko, V.T.A., Macario Rojas, A., Smith, K.L.,
1. Bandyopadhyay, S., Subramanian, G.P., Foust, R., Morgan, Becedas, J., González, G., Vázquez, I., Braña, Á., Antonini,
D., Chung, S.-J., Hadaegh, F.Y.: A Review of Impending K., Bay, K., Ghizoni, L., Jungnell, V., Morsbøl, J., Binder,
Small Satellite Formation Flying Missions. In: 53rd AIAA T., Boxberger, A., Herdrich, G.H., Romano, F., Fasoulas, S.,
Aerospace Sciences Meeting, Kissimmee, Florida, USA Garcia-Almiñana, D., Rodriguez-Donaire, S., Kataria, D.,
(2015) Davidson, M., Outlaw, R., Belkouchi, B., Conte, A., Perez,
2. Traub, C., Romano, F., Binder, T., Boxberger, A., Herdrich, J.S., Villain, R., Heißerer, B., Schwalber, A.:
G.H., Fasoulas, S., Roberts, P.C.E., Smith, K., Edmondson, DISCOVERER - Radical Redesign of Earth Observation
S., Haigh, S., Crisp, N.H., Oiko, V.T.A., Lyons, R., Worrall, Satellites for Suistained Operation at Significantly Lower
S.D., Livadiotti, S., Becedas, J., González, G., Domínguez, Altitudes. In: 68th International Astronautical Congress,
R., González, D., Ghizoni, L., Jungnell, V., Bay, K., Adelaide, Australia (2017)
Morsbøl, J., Garcia-Alminana, D., Rodriguez-Donaire, S., 14. Traub, C., Herdrich, G.H., Fasoulas, S.: Influence of energy
Sureda, M., Kataria, D., Outlaw, R., Villain, R., Perez, J.S., accommodation on a robust spacecraft rendezvous maneuver
Conte, A., Belkouchi, B., Schwalber, A., Heißerer, A.: A using differential aerodynamic forces. CEAS Space J 87(5),
Review and Gap Analysis of Exploiting Aerodynamic 427 (2019)
Forces as a Means to Control Satellite Formation Flight. In: 15. Sentman, L.H.: Free molecule flow theory and its
67th Deutscher Luft- und Raumfahrtkongress 2018, application to the determination of aerodynamic forces.
Friedrichshafen, Germany (2018) Technical Report, Lockheed Aircraft Corporation (1961)
3. Traub, C., Romano, F., Binder, T., Boxberger, A., Herdrich, 16. Pilinski, M.D., Argrow, B.M., Palo, S.E.: Semiempirical
G.H., Fasoulas, S., Roberts, P.C.E., Smith, K., Edmondson, Model for Satellite Energy-Accommodation Coefficients.
S., Haigh, S., Crisp, N.H., Oiko, V.T.A., Lyons, R., Worrall, Journal of Spacecraft and Rockets 47(6), 951–956 (2010)
S.D., Livadiotti, S., Becedas, J., González, G., Dominguez,
R.M., González, D., Ghizoni, L., Jungnell, V., Bay, K.,
Morsbøl, J., Garcia-Almiñana, D., Rodriguez-Donaire, S.,
Sureda, M., Kataria, D., Outlaw, R., Villain, R., Perez, J.S.,
Conte, A., Belkouchi, B., Schwalber, A., Heißerer, B.: On
the exploitation of differential aerodynamic lift and drag as a
means to control satellite formation flight. CEAS Space J
65((11–12)), 1537 (2019)
4. Leonard, C.L.: Formationkeeping of Spacecraft via
Differential Drag. Master Thesis, Massachusetts Institute of
Techology (1986)
5. Horsley, M.: An Investigation into using Differential Drag
for Controlling a Formation of CubeSats. In: Advanced
Maui Optical and Space Surveillance Technologies
Conference, Maui, HI, United States (2011)
6. Horsley, M., Nikolaev, S., Pertica, A.: Small Satellite
Rendezvous Using Differential Lift and Drag. Journal of
Guidance, Control, and Dynamics 36(2), 445–453 (2013)
7. Smith, B., Boyce, R., Brown, L., Garratt, M.: Investigation
into the Practicability of Differential Lift-Based Spacecraft
Rendezvous. Journal of Guidance, Control, and Dynamics
40(10), 2682–2689 (2017)
8. Leonard, C.L., Hollister, W., Bergmann, E.: Orbital
formationkeeping with differential drag. Journal of
Guidance, Control, and Dynamics 12(1), 108–113 (1987)
9. Schweighart, S.A., Sedwick, R.J.: High-Fidelity Linearized
J2 Model for Satellite Formation Flight. Journal of
Guidance, Control, and Dynamics 25(6), 1073–1080 (2002)
10. Vallado, D.A., McClain, W.D.: Fundamentals of
astrodynamics and applications, 4th edn. Space Technology
Library. Published by Microcosm Press, Hawthorne, CA
(op. 2013)
11. Shao, X., Song, M., Zhang, D., Sun, R.: Satellite rendezvous
using differential aerodynamic forces under J2 perturbation.
Aircraft Eng & Aerospace Tech 87(5), 427–436 (2015)