Astrodynamics
Vol. 2, No. 1, 25–37, 2018
https://doi.org/10.1007/s4206401700092
Hongwei Yang ^{1} , Gao Tang ^{1}^{,}^{2} , and Fanghua Jiang ^{1} ( )
1. School of Aerospace Engineering, Tsinghua University, Beijing 100084, China
2. Department of Mechanical Engineering and Materials Science, Duke University, Durham, NC 27708, USA
ABSTRACT
This paper presents the crucial method for obtaining our team’s results in the 8th Global Trajectory Optimization Competition (GTOC8). Because the positions and velocities of spacecraft cannot be completely determined by one observation on one radio source, the branch and bound method for sequence optimization of multiasteroid exploration cannot be directly applied here. To overcome this diﬃculty, an optimization method for searching the observing sequence based on nominal lowthrust trajectories of the symmetric observing conﬁguration is proposed. With the symmetric observing conﬁguration, the normal vector of the triangle plane formed by the three spacecraft rotates in the ecliptic plane periodically and approximately points to the radio sources which are close to the ecliptic plane. All possible observing opportunities are selected and ranked according to the nominal trajectories designed by the symmetric observing conﬁguration. First, the branch and bound method is employed to ﬁnd the optimal sequence of the radio source with thrice observations. Second, this method is also used to ﬁnd the optimal sequence of the left radio sources. The nominal trajectories are then corrected for accurate observations. The performance index of our result is 128,286,317.0 km which ranks the second place in GTOC8.
KEYWORDS
branch and bound optimization symmetric observing
conﬁguration
GTOC
low thrust
Research Article Received: 16 December 2016 Accepted: 20 June 2017 2017 Tsinghua University Press
1
Introduction
In the 8th Global Trajectory Optimization Competition (GTOC8), an innovative problem, which is related to interplanetary trajectory design, was released by the Jet Propulsion Laboratory. The theme for this new problem is “highresolution mapping of radio sources in the universe using spacebased verylongbaseline interferometry (VLBI) [1]. During the competition, we proposed an eﬀective design method for this new problem. Using this method, all computations were ﬁnished eﬃciently on our personal computers. The authors’ team, Tsinghua University, ranks the second place in GTOC8 [1]. Our team achieved that all observed radio sources are near the ecliptic plane. These observed radio sources are preferable targets according to the performance index of GTOC8. The goal of this paper is to introduce the optimization method of the observing sequence as well as the nominal trajectories of
jiangfh@tsinghua.edu.cn
a symmetric observing conﬁguration for the sequence optimization. The reason that we used the symmetric observing conﬁguration and proposed the observing sequence optimization based on nominal trajectories is because of the particularity and complexity of GTOC8 compared with the previous GTOCs. For the most of the previous GTOCs, the problems concern lowthrust trajectory optimization for asteroid missions, such as asteroid defense, asteroid sample return, multiple asteroid rendezvous, and ﬂyby, etc. (http://sophia.estec.esa.int/gtoc portal/). Extensive papers have been published in the ﬁelds of low thrust interplanetary trajectory optimization. The optimization methods can be classiﬁed into two categories: indirect methods [2–8] and direct methods [9]. Gao and Kluever [10] also proposed a hybrid method which combines these two kinds of methods. As for improving the solution eﬃciency, there are two
26
H. Yang, G. Tang, F. Jiang
levels of techniques The ﬁrst level is about global searching. One of the most popular approaches is to approximate lowthrust trajectories by impulse trajectories and then search the optimal asteroid sequence through the branch and bound method [5]. Because impulse trajectories are not equivalent to lowthrust trajectories, fast methods for evaluating lowthrust trajectories are presented as well [6, 7]. Additionally, there are studies about determining optimal gravity assist sequences [8,11]. The second level is about lowthrust trajectory optimization. Amount of powerful techniques have been proposed, such as smoothing techniques [2, 4], parallel computation [3], a shapebased trajectory technique [9], etc. However, two new features appear in the problem of GTOC8 compared with the previous GTOCs, leading the problem to be more complex. Three spacecraft need to cooperate for observations. More importantly, the directions of the radio sources cannot be used to determine the position and/or velocity of any of the spacecraft. Although the problem of GTOC7 is also about a multiple spacecraft mission [12], searching of target sequences of each spacecraft can still be regarded independently. Besides, the position and velocity of each spacecraft are equal to that of the target asteroid at each rendezvous time. A symmetric observing conﬁguration is found by the authors during GTOC8. With this conﬁguration, the observing direction rotates periodically in the ecliptic plane, which is preferable for repetitive observations of the radios close to the ecliptic plane. This conﬁguration is also taken as an eﬃcient tool by some other teams such as the team of State Key Laboratory of Astronautic Dynamics and the team of Chinese Academy of Sciences [13, 14]. These teams [13, 14] both got excellent results. One of main diﬀerences between our method and their method is that the optimization of the observing sequence. In this paper, we take the idea of using the branch and bound method to search the optimal observing sequence. Diﬀerent from the application of the branch and bound method in sequence searching of the multiple asteroid rendezvous mission [5], the searching in the current paper is based on the predesigned nominal trajectories so that no lambert problems need to be solved. Our method for GTOC8 has been brieﬂy introduced in our previous conference paper [15]. Compared with that paper [15], this paper makes two contributions.
First, the symmetric observing conﬁguration and its nominal trajectories are further analyzed. Second, the algorithms for optimization of the observing sequence are given in detail. The rest of the paper is organized as follows. In Section 2, the problem of GTOC8 is brieﬂy introduced. Section 3 presents and analyzes the symmetric observing conﬁguration. In Section 4, the designed nominal trajectories of the symmetric observing conﬁguration is introduced and analyzed. Section 5 presents the branch and bound method for searching the optimal observing sequence. Section 6 concludes the paper.
2
Background
In this section, the problem of GTOC8 [1] is brieﬂy introduced. Three lowthrust propelled spacecraft are launched to formulate a series of triangles to observe radio sources. The equation of motion of each spacecraft is
r¨ + µ
r
r ^{3}
=
T max u
m (1)
α
where r is the position vector of the spacecraft in the Earth Mean Ecliptic and Equinox of J2000 frame, µ is the gravitational constant of the Earth, T _{m}_{a}_{x} is the maximal thrust magnitude, and m is the mass of the spacecraft. The maximum magnitude of the thrust is 0.1 N. The control variable consists of the unit vector of thrust direction α and the engine thrust ratio u ∈[0, 1]. The mass variation is computed by
m˙
= −
T max u I sp g 0
(2)
where I _{s}_{p} is the speciﬁc impulse, and g _{0} is the standard acceleration of gravity at the sea level. The three spacecraft are launched from a 400km altitude circular orbit around the Earth. The whole mission is no more than 3 years. The range of the spacecraft relative to the Earth must obey the following condition:
6578.14 km r 1,000,000.0 km
(3)
An observation is recorded when the normal direction of the observing triangle is aligned with the direction of a radio source. The direction error must be less than 0.1 deg. The time intervals between two adjacent observations must be no less than 15 days. The performance index J to be maximized is [1]:
J = ^{} Ph ^{} 0.2 + cos ^{2} δ ^{} (4)
Optimization of observing sequence based on nominal trajectories of symmetric observing conﬁguration
27
where P is the weighting factor, h is the smallest of the three altitudes of the observing triangle, and δ is the declination of the observed radio sources. An altitude of a triangle is the perpendicular distance from a vertex to the opposite side (or its extension) [1]. The observing altitude h must be larger than 10,000 km for eﬀective observations. The weighting factor P takes on the values of 1, 3, 6, and 0 for repeat observations according to the rules. The most attractive case is to observe the same radio source for three times and the three observing altitudes satisfy h _{2} /h _{1} > 3 and h _{3} /h _{2} > 3.
3 Analysis of the symmetric observing conﬁguration
3.1
Reason
of
using
the
observing conﬁguration
symmetric
According to Eq. (4), radio sources near the ecliptic plane are preferable because a low declination δ greatly contributes to the performance index. Therefore, an observing conﬁguration with its observing direction rotating in the ecliptic plane is considered in the preliminary design. In GTOC8, 420 radio sources are given by the organizer. Because the time interval between two adjacent observations should be no less than 15 days, no more than 16 diﬀerent radio sources, if each one is observed thrice, can be observed in two years. Actually, 37 radio sources, whose declinations are from −3 to 3 deg, can be selected from all 420 given radio sources, as shown in Fig. 1. Therefore, the radio sources near the ecliptic plane are enough for designing a competitive observing mission. Because the number of radio sources is reduced to 37, the complexity of computations is greatly reduced. Additionally, it contributes greatly to the performance index if the weighting factor P takes the maximum value. But, it requires that a radio source is observed thrice. In order to repeatedly observe many radio sources, it is beneﬁcial to rotate the observing direction periodically. Therefore, we propose to use the symmetric observing conﬁguration which enables the observing direction rotating periodically in the ecliptic plane. The geometry of the desired conﬁguration and its properties will be given below.
3.2
Geometry of the symmetric observing conﬁguration
The geometry of the symmetric observing conﬁguration
is shown in Fig. 2. As for this conﬁguration, the following
four conditions are satisﬁed:
(1) two of the three
spacecraft, denoted as S _{1} and S _{2} , orbit symmetrically about the ecliptic plane; (2) the other one of the three spacecraft, denoted as S _{3} , stays in the ecliptic plane; (3) the orbits are all circular with the same radius; (4) the arguments of latitude of S _{1} and S _{2} are the same (the ascending nodes are opposite) while the diﬀerence in the
(a) All given radio sources
(b) The selected radio sources
Fig. 1
Selection of the radio sources.
Fig. 2
Geometry of the symmetric observing conﬁgurations.
28
H. Yang, G. Tang, F. Jiang
arguments of latitude of S _{1} and S _{3} is π. Denote the inclination and the argument of latitude of S _{1} as i and θ, respectively. Then, the position vectors of the three spacecraft are derived as follows:
r _{1} = 
r[ 
cos θ 
sin θ cos i 
sin θ sin i ] 

r _{2} = 
r[ 
cos θ 
sin θ cos i 
− sin θ sin i 
] 
r _{3} = r[ 
− cos θ 
− sin θ 
0 ] 
(5)
(6)
(7)
The normal direction of the observing triangle is
n = ±
(r _{1} − r _{2} ) × (r
1
−
r 3
)
(r _{1} − r _{2} ) × (r _{1} − r _{3} )
where (r _{1} − r _{2} ) × (r _{1} − r _{3} ) = r ^{2} [ −2sin ^{2} θsini(cosi+1)
4sinθsinicosθ
0
]
^{(}^{8}^{)}
(9)
According to Eqs. (8) and (9), the normal direction of the observing triangle keeps parallel to the ecliptic plane, which is also shown in Fig. 2. Moreover, the observing altitude is low if S _{1} and S _{2} are close to the ecliptic plane but it becomes higher if S _{1} and S _{2} move away from the ecliptic plane. To obtain the maximum h, the following conditions must be satisﬁed: (1) the inclination of S _{1} is 60 deg; (2) the three spacecraft consist of an equilateral triangle; (3) the radius of each spacecraft’s orbit is 1×10 ^{6} km. Once these conditions are satisﬁed, h is equal to 1.5×10 ^{6} km. The conﬁguration with i = 60 deg is chosen as the ﬁnal conﬁguration in our strategy.
3.3 Properties of the symmetric observing conﬁguration
Because n is always parallel to the ecliptic plane, the ﬁrst property of the symmetric observing conﬁguration is that the observing direction can be regarded as staying in the ecliptic plane. The second property is about the periodicity of the observing direction. According to Eqs. (8) and (9), the right ascension of the observing direction is
α _{1} = 
arctan 
2 ^{} 4sinθsinicosθ, −2sin ^{2} θsini(cosi+1) ^{} 

(10) 

or 

α _{2} = arctan 
2 
^{} −4sinθsinicosθ, 2sin ^{2} θsini(cosi+1) ^{} 

(11) 
Because the diﬀerence between α _{1} and α _{2} keeps
π, only the variation of α _{1}
is analyzed herein. Both
directions will be considered in searching the optimal
sequence. Besides, the considered range for α is from 0
to 2π in the rest of this paper.
If α is less than zero,
then α = α + 2π. The history of α with θ for diﬀerent i is shown in Fig. 3. According to this ﬁgure, α has
periodicity for each inclination.
It also
can be found
that the curve of α to θ becomes straight when the
inclination decreases.
Nevertheless, the eﬀect of i on
α is small according to the shapes of the curves.
4 Nominal trajectories of the symmetric observing conﬁguration
In this paper, we only concern the lowthrust trajectories after the Moon gravity assists because there is no eﬀective observation in our design result before gravity assists. Before the Moon gravity assists, we use the analytical control law which maximizes the variation of the apogee to design the multiplerevolution trajectories. In this way, the orbits of spacecraft can be raised to the Moon rapidly. At the end stage, the trajectory optimization is conducted to obtain the desired ﬁnal state before the Moon gravity assists. More details for designing the lowthrust trajectory before the Moon gravity assists can be found in Ref. [15]. Additionally, we use the homotopy method [4] with the logarithmic homotopy [2] for lowthrust trajectory optimization. And the equation of motion is described by the equinoctial elements [10]. More details on the lowthrust trajectory optimization can also be found in Ref. [15]. The state variables of the three spacecraft are shown in Table 1. After the Moon gravity assists, the orbits of all spacecraft are ellipses with high eccentricities. Besides, the orbital planes of S _{1} and
inclinations.
Optimization of observing sequence based on nominal trajectories of symmetric observing conﬁguration
29
Table 1
State variables of spacecraft immediately after the Moon gravity assists
Spacecraft Epoch (MJD) 
x (km) 
y (km) 
z (km) 
v _{x} (km) 
v _{y} (km) 
v _{z} (km) 
Mass (kg) 
59077.839 
372202.105 
0 
−0.672 
−0.900 
0.028 
1994.145 

59077.839 
372202.105 
0 
−0.672 
−0.900 
−0.028 
1995.612 

59090.705 
−385500.332 
0 
0.876 
−0.653 
0 
1994.060 
S _{2} are very close to the ecliptic plane. To achieve the ﬁnal observing conﬁguration as well as adjust the size of the conﬁguration for obtaining desired observing altitudes, both circularization and adjustments of orbital inclination are essential. The designed nominal trajectories in our methods belong to four stages. In the ﬁrst stage, the orbits of the three spacecraft are circularized and adjusted to construct the symmetric observing conﬁguration. The target radius is set to 0.99×10 ^{6} km. In the second stage, the inclinations of S _{1} and S _{2} are changed to ±13 deg and the time spent on this stage is 2 orbital periods. In order to achieve observations with high performance, a radio source needs to be observed three times and the three observing altitudes should satisfy h _{m}_{a}_{x} /h _{m}_{i}_{d} 3 and h _{m}_{i}_{d} /h _{m}_{i}_{n} > 3 [1]. The h for each i takes its maximum when θ = π. The variation of h with respect to i when θ = π is shown in Fig. 4. The inclination of 13 deg is chosen to make the maximum observing height at the last coast stage as about 1/3 of the largest observing height. The length of this period is approximate to those of the ﬁrst stage and the last coast stage. The reason that this ratio is not exactly 1/3 is to make the period of the third stage as integral multiples of the half orbital period. In the third stage, the inclinations of S _{1} and S _{2} are changed to ±60 deg. In the fourth stage, all spacecraft coast until the mission ends.
Fig. 4
Variation of h with respect to i when θ = π.
The nominal trajectories in each stage are presented and analyzed below.
Nominal trajectories in stage 1
The designed nominal trajectories for stage 1 are shown in Fig. 5. According to the ﬁgure, the semimajor
axis of each spacecraft’s orbit increases and all orbits are circularized during this stage. The history of the observing altitude and direction
for stage 1 are shown in Fig. 6.
The reason that there
are two color lines in Fig. 6(b) is because α has two solutions. This explanation also holds for the ﬁgures in other stages. According to the ﬁgure, the observing altitude and direction are not periodic in this stage. This is because the symmetric observing conﬁguration has not been constructed before this stage. Nominal trajectories in stage 2 The designed nominal trajectories for stage 2 are shown in Fig. 7. According to the ﬁgure, both the inclinations of S _{1} and S _{2} change during this stage and then stay at the desired values. The history of the observing altitude and direction for stage 2 are shown in Fig. 8. The ﬁrst wave in Fig. 8(a) is not the same as other three waves because the inclination adjustment is conducted. After that, both the observing altitude and direction becomes periodic. And the range of the observing direction is from 0 to
2π.
Fig. 5
Nominal trajectories in stage 1.
30
H. Yang, G. Tang, F. Jiang
Fig. 7
Nominal trajectories in stage 2.
Nominal trajectories in stage 3
The designed nominal trajectories for stage 3 are shown in Fig. 9. According to the ﬁgure, both the inclinations of S _{1} and S _{2} change continuously during this stage.
Fig. 9
Nominal trajectories in stage 3.
The history of the observing altitude and direction for stage 3 are shown in Fig. 10. According to the ﬁgure, the peaks of the wave increases. This is because of the inclination adjustment. But, the observing direction is almost periodic. And the range of the observing
Optimization of observing sequence based on nominal trajectories of symmetric observing conﬁguration
31
Fig. 10
History of the observing altitude and direction in stage 3.
direction is from 0 to 2π. Nominal trajectories in stage 4
The designed nominal trajectories for stage 4 are shown in Fig. 11. According to the ﬁgure, all spacecraft
keep coasting during this stage.
Fig. 11
Nominal trajectories in stage 4.
The history of the observing altitude and direction for stage 4 are shown in Fig. 12. According to the ﬁgure, both the observing altitude and direction are periodic. And the range of the observing direction is from 0 to 2π. Compared Fig. 12 with Fig. 8, we see that the shapes of the waves of the observing altitude are similar and the histories of the observing direction are similar as well. Because of this phenomenon, we predict that if an observing sequence is found in stage 4 then the same observing sequence can be found in stage 2 with the same order it appears. Therefore, we search the observing sequence in stage 2 the same as that in stage 4. It greatly reduces the computation eﬀort when the branch bound method is applied.
Fig. 12
History of the observing altitude and direction in stage 4.
32
H. Yang, G. Tang, F. Jiang
trajectories. With these nominal trajectories, we propose an algorithm to select all possible observing opportunities. After that, a global optimization method, the branch bound method, is employed for optimization of the observing sequence.
5.1 
Selecting 
possible 
observing 
opportunities 
The following algorithm is used for selecting possible observing opportunities:
Step 1 Initialize the discrete time points {t _{1}
,. . .
,
t _{N} } between t _{s} and t _{f} with the time interval of one day.
Step
2
Compute
h
and
α
at
the
discrete
time
points.
Step
3
Find all observing possibilities with the
selected radio sources between
t _{k}
and t _{k}_{+}_{1} (k =
1,. . .
,
N − 1) based on α. Use linear interpolations to obtain
observing time points and h at these time points. Step 4 Sort all possible observations according to observing time. The time interval for the discrete time points is set to 1 h. The positions of the three spacecraft at these time points are calculated by a cubic spline method. Because the declinations of the selected radio sources are small, a radio source is regarded as observed if its right ascension is the same as that of the observing direction. The declination error will be corrected after the sequence has been optimized. The observing time and the observing altitude between two adjacent discrete time points are computed by the linear interpolation. With the algorithm mentioned above, we select all the possible observing opportunities as well as the corresponding observing altitudes. The results
are shown in Fig. 13, where the stars denote the observing opportunities and the dashed lines represent the boundaries of the four stages. As shown in the ﬁgure, the widths of stage 1, stage 2, and stage 4 are almost the same. Moreover, the maximum altitudes of these three stages satisfy h _{m}_{a}_{x} /h _{m}_{i}_{d} > 3 and h _{m}_{i}_{d} /h _{m}_{i}_{n} > 3. We will search the optimal sequence of the thrice observed radio sources in these stages. Because the thrice observed radio sources dominate the performance index, we conducted this search ﬁrst. After the optimal thrice observing sequence has been determined, we conducted optimization of the sequence with twice and once observed radio sources.
5.2
Optimization of the sequence of the thrice observed radio sources
The branch and bound method is used for the optimization of the sequence with thrice observed radio sources. Because the observations with P = 6 dominate the performance index, we conducted the searching in stage 4 ﬁrst. The searching procedure is shown in Fig. 14. In this ﬁgure, P _{N}_{−}_{k} represents the number of the radio sources. As illustrated in Fig. 14, we take last 30 points of the observing opportunities in Fig. 13 as the starting point in the searching procedure. In this way, the observing opportunities near the last wave crest are all considered. Besides, the opportunities with high observing altitude are preferred. The points whose observing altitude h is less than 7.0×10 ^{5} km are discarded, as indicated by the dashed boxes in the ﬁgure. Then, we increase the sequence step by step. As for the point P _{N}_{−}_{k} , the points from P _{N}_{−}_{k}_{−}_{1} to P _{N}_{−}_{k}_{−}_{5}_{0} are considered and the point
Fig. 13
History of the observing altitude and all possible observing opportunities.
Optimization of observing sequence based on nominal trajectories of symmetric observing conﬁguration
33
Fig. 14 Searching procedure of the branch and bound algorithm 
stage 2 as {Q _{N} illustrated in Fig. 
,Q _{1} }. The searching procedure is 

for observations with P = 6 (dashed box: h is not satisﬁed; dashed line: the time interval is not satisﬁed). 
varies from Q _{N} to Q _{1} . If its number matches T _{k} , then the following two conditions are checked: (1) the time interval between this radio source and any other selected radio source is no less than 15 days; (2) its observing altitude is less than 1/3 of that of the same radio source in stage 4. If these two conditions are satisﬁed, this radio source is regarded as feasible and the observing sequence increases. After the branch and bound searching, we obtain 50,872 observing sequences. The ten sequences with the largest performance indexes are shown in Table 3. In this table, ∆J represents the increment to the performance index because of the observations with P = 3. We ﬁnd that the given performance indexes are all larger than 115×10 ^{6} km. Because the number 

is chosen if the observing time interval between this point and P _{N}_{−}_{k} is no less than 15 days. A termination condition of increasing sequences is either no new radio source can be found or a sequence’s length reaches 15. With the branch and bound method mentioned above, we obtain 158,112 observing sequences and all these sequences will be used for the next searching. According to the performance index, the top ten of the obtained observing sequences are found and listed in Table 2. The lengths of these observing sequences are all 12 and the performance indexes are all over 100×10 ^{6} km. Then, we search the observing sequences in stage 2 with P = 3. Because the observing sequence in stage 3 must have the same radio sources, the searching procedure is not the same as that shown in Fig. 14. Denote a observing sequence found in stage 4 as 
of the obtained observing sequences is large, only those with J > 114 × 10 ^{6} km will be used in the next step. Last, we search the observing sequences in the part of stage 2 where no radio sources are selected and in stage 

{T _{1} 
,T _{1}_{2} } and the numbers of the radio sources in 
Fig. 15 Searching procedure of the branch and bound algorithm for observations with P = 3. 

Table 2 Top ten of the obtained observing sequences with P = 6 

Rank 
Observing sequence 
J (km) 


101830796.8 


101830790.5 


101830785.8 


101830785.1 


101830440.2 


101830429.8 


101830429.2 


101830426.8 


101828333.3 

10 
223199193207220214206218212224201213 
101828325.2 
34
H. Yang, G. Tang, F. Jiang
Table 3
Top ten of the obtained observing sequences with P = 6 and 3
Rank 
Observing sequence 
∆J (km) 
J (km) 

13288434.3 
115119220.1 


13288618.4 
115119047.5 


13288184.9 
115116506.1 


13288369.0 
115116333.6 


13283593.2 
115099630.1 


13283777.3 
115099457.6 


13283343.9 
115096916.2 


13283527.9 
115096743.6 


13281708.2 
115050641.6 

10 
223199193206200212207220420224201195 
13281892.3 
115050469.1 
1 with P = 1. The searching procedure is similar to that for stage 2. After the branch and bound searching, we obtain 1,384,288 observing sequences. Then, the best observing sequence is taken as the optimal observing sequence. The result is shown in Table 4. The increment to the performance index in this stage is small because P = 1 and the observing altitudes are small. The distribution of the optimal observations with three times is shown in Fig. 16. As the prediction stated in Section 4, the observing sequence in stage 4 has the same appearing order as that in stage 2.
5.3
Optimization of the sequence with twice and once observed radio sources
After the optimal sequence with thrice observations is obtained, the branch and bound method is used for optimization of the sequence with twice and once observations. Because these observations contribute
minor to the performance index, we introduce the approach brieﬂy. First, radio sources that will be repeatedly observed twice (P = 1, 3) is searched. The searching procedure is similar to that for thrice observations, but the searching area for P = 3 are in stage 3 and stage 1 while searching area for P = 1 is in stage 1. Only the radio sources which have not been selected for thrice observation are considered. And the selected radio sources in this section must satisfy the constraint of time intervals. Second, all other radio sources that can be observed once (P = 1) is searched. The result of the observing sequence with two observations is listed in Table 5 and the result of the observing sequence with one observation is listed in Table 6. The increments due to twice observation and once observation is much smaller compared with that of the thrice observation. The total performance index is 126,298,367.8 km which is a little smaller than our ﬁnal
Table 4
Optimal observing sequence with three times of observations
Fig. 16
Distribution of the optimal observations with three times.
Optimization of observing sequence based on nominal trajectories of symmetric observing conﬁguration
35
Table 5 
Result of the observing sequence with two observations 

Observing sequence ∆J (km) 
J (km) 

204209190 4259296.0 
120058103.3 

Table 6 
Result of the observing sequence with one observation 

Observing sequence ∆J (km) 
J (km) 

211198203195218205211216 6240264.5 
126298367.8 
result. This is because the observations designed by the nominal trajectories are not accurate. After orbit corrections for accurate observations, the performance index will change. The distribution of the optimal observations with one and two times is shown in Fig. 17. According to this ﬁgure, all these observations happen in stage 1 and stage
3.
5.4 Results with accurate observations
After the observations are approximately designed, orbit corrections are conducted to obtain accurate observations. The introduction on the orbit corrections
can be found in Ref. [15]. The trajectories of the three spacecraft for accurate observations are shown in Fig.
18.
The summary of the optimal observing sequence submitted is shown in Table 7. The ﬁnal performance index is J = 128 286 317.0 km.
6
Conclusions
In this paper, the symmetric observing conﬁguration is used for designing lowthrust trajectories of three
Table 7
Result of the observing sequences submitted
Category
Number of points
Radio source
_{R}_{s}_{p}_{1}_{3}_{6}
Rsp13
Rsp11
Rsp1
_{1}_{2}
3
1
6
223, 199, 193, 206, 200, 212, 207, 220, 214, 224, 201, 213
204, 209, 190
211
198, 203, 195, 218, 205, 216
Rsp136 = radio sources observed exactly thrice with P =1, 3, and 6. Rsp13 = radio sources observed exactly twice with P =1, 3. Rsp11 = radio sources observed exactly twice with P = 1 both times. Rsp1 = radio sources observed exactly once with P = 1.
spacecraft for observing radio sources. With the observing conﬁguration, the normal vector of the triangle plane formed by the three spacecraft rotates in the ecliptic plane periodically and approximately points to the radio sources which are close to the ecliptic plane. The searching procedures using the branch and bound method based on nominal trajectories for optimal observing sequence is then proposed. The obtained optimal sequence has 50 observations and 22 unique radio sources. Speciﬁcally, 12 radio sources observed exactly thrice with P = 1, 3, and 6; 3 radio sources observed exactly twice with P = 1, 3; 1 radio source observed exactly twice with both P = 1; and 6 radio sources observed exactly once with P = 1. The performance J of the optimized lowthrust trajectories for accurate observations is 128,286,317.0 km. These results are exactly the same as that we submitted in GTOC8. The second place ranking of our performance in GTOC8 implies advantage and eﬀectiveness of our method.
Fig. 17
Distribution of the optimal observations with one and two times.
36
H. Yang, G. Tang, F. Jiang
accurate observations.
Acknowledgements
This work is supported by the National Natural Science Foundation of China (Grant Nos. 11672146 and 11432001). The authors thank the organizer of GTOC8.
References
[1] Petropoulos, A. E. GTOC8: Problem description and
[2] 
summary of the results. In: Proceedings of the 26th AAS/AIAA Space Flight Mechanics Meeting, 2016. Bertrand, R., Epenoy, R. New smoothing techniques for solving bang–bang optimal control problems— numerical results and statistical interpretation. Optimal Control Applications and Methods, 2002, 23(4): 171– 

197. 

[3] 
Olympio, J. T. Optimal control problem for lowthrust 

multiple asteroid tour missions. Journal of Guidance, 

[4] 
Control, and Dynamics, 2011, 34(6): 1709–1720. Jiang, F., Baoyin, H., Li, J. Practical techniques for 

lowthrust trajectory optimization with homotopic approach. Journal of Guidance, Control, and 

[5] 
Dynamics, 2012, 35(1): 245–258. Jiang, F., Chen, Y., Liu, Y., Baoyin, H., Li, J. GTOC5: 

[6] 
Results from the Tsinghua University. Acta Futura, 2014, 8: 37–44. Casalino, L. Approximate optimization of lowthrust transfers between loweccentricity close orbits. Journal 

of Guidance, Control, and Dynamics, 2014, 37(3): 

1003–1008. 

[7] 
Gatto, 
G., 
Casalino, 
L. 
Fast 
evaluation and 
optimization of lowthrust transfers to multiple targets. Journal of Guidance, Control, and Dynamics, 

[8] 
2015, 38(8): 1525–1530. Yang, H., Li, J., Baoyin, H. Lowcost transfer between 
asteroids with distant orbits using multiple gravity assists. Advances in Space Research, 2015, 56(5): 837–
[9]
847.
Petropoulos,
A. E., Longuski, J. M. Shapebased
algorithm for the automated design of lowthrust,
gravity assist trajectories. Journal of Spacecraft and Rockets, 2004, 41(5): 787–796.
[10] 
Gao, 
Y., 
Kluever, 
C. A. Lowthrust interplanetary 

orbit transfers using hybrid trajectory optimization method with multiple shooting. In: Proceedings of the AIAA/AAS Astrodynamics Specialist Conference and 

Exhibit, Guidance, Navigation, and Control and Co 

[11] 
located Conferences, 2004: AIAA 20045088. Petropoulos, A. E., Longuski, J. M., Bonﬁglio, E. P. 

Trajectories to Jupiter via gravity assists from Venus, Earth, and Mars. Journal of Spacecraft and Rockets, 2000, 37(6): 776–783. 

[12] Casalino, 
L., 
Colasurdo, 
G. Problem description 

for the 7th Global Trajectory Optimisation Competition. 2014. Available at http://sophia.estec. esa.int/gtoc portal. 

[13] 
Shen, H.X., 
Huang, A.Y., Zhang, Z.B., Li, H.N. 

GTOC8: Results and methods of State Key Laboratory 

of Astronautic Dynamics. In: Proceedings of the AAS/AIAA Space Flight Mechanics Meeting, 2016: 

AAS 16384. 

[14] 
He, 
S., 
Zhu, 
Z., 
Gao, 
Y. 
GTOC8: 
Results and 
methods of TEAM 8/Chinese Academy of Sciences. In:
Proceedings of the AAS/AIAA Space Flight Mechanics Meeting, 2016: AAS 16552.
[15]
Tang, G.,
Yang,
H.,
Jiang, F.,
Baoyin,
H.,
Li,
J.
GTOC8: Results and methods of TEAM 3–Tsinghua
University. In: Proceedings of the AAS/AIAA Space Flight Mechanics Meeting, 2016: AAS 16248.
Hongwei Yang received his Ph.D. degree in aerospace engineering from Tsinghua University, China, in 2017, and he was a visiting Ph.D. student of Rutgers University, USA, in 2015–2016. His current research interests include dynamics and control near asteroids, interplanetary trajectory design, and optimization. Email: yang.hw.thu@gmail.com.
Gao
Tang received his bachelor
degree and master degree in aerospace
engineering from Tsinghua University, China, in 2010 and 2012, respectively.
He is currently pursuing his Ph.D. degree in the Department of Mechanical Engineering and Material Science at Duke University, USA. His current
include
dynamics
and
control
in
Optimization of observing sequence based on nominal trajectories of symmetric observing conﬁguration
37
robotics, interplanetary trajectory optimization, and global optimization methods. Email: gao.tang@duke.edu.
Fanghua Jiang received his B.S. degree in engineering mechanics and Ph.D. degree in mechanics from Tsinghua University, China, in 2004 and 2009, respectively. Since 2009, he has worked in the School of Aerospace Engineering at Tsinghua University, China, including two years of postdoctor, three years of research assistant, and two years of associate professor up to now. Currently, he is an AIAA senior member. His
current research interests include astrodynamics, spacecraft formation ﬂying, and interplanetary trajectory optimization. Email: jiangfh@tsinghua.edu.cn.