This paper addresses the coordination problem of distributed spacecraft systems with time-invariant communication topologies. In order to estimate the neighbor spacecraft’ relative states which are not able to be measured directly, estimating processes are established on each spacecraft by utilizing an unscented Kalman filtering approach with measuring principles of both the pseudorange and the carrier phase. Based on such estimations, a distributed consensus protocol is proposed to address the coordinated motion problem of distributed spacecraft. The effectiveness of the proposed approach is demonstrated through numerical simulations, with a low-Earth-orbit satellites flying scenario.
Introduction
The coordination problem of distributed spacecraft systems has attracted much attention in recent years. Compared to solo systems, additional benefits such as high flexibility and great efficiency can be obtained by having a group of distributed spacecraft work cooperatively.
A fundamental approach to achieve such coordination is consensus, which enables all agents of a system to reach an agreement on certain quantities of interest. Consensus problems have been widely studied in recent years. Jadbabaie et al. (2003) analyzed the coordination behavior of a group of autonomous agents and provided a theoretical explanation for such behavior, and their results indicated that proper control models based only on neighbor information could motivate systems to achieve consensus. Ren et al. (2007) made an overview on information consensus of multiple vehicles by not only reviewing different consensus control laws but also describing several potential specific applications, such as rendezvous, formation stabilization, formation maneuvering, and flocking. Some more challenging results were obtained in Ren (2008) by extending consensus algorithms from first-order models to second-order ones, which made consensus approaches applicable for systems with second-order dynamics. Some practical conditions such as the actuator saturations and the lacks of relative velocity measurements were also discussed. In the context of spacecraft coordination, Ren (2007) considered consensus problems and showed the concept of coordinated motion. By utilizing a consensus control law with local information exchanges, the given formation configuration was maintained during the maneuvers, and all states of spacecraft were assumed available with respect to the Earth Centered Inertial (ECI) frame. In order to obtain accurate formation configurations, orbital reference frames were used to represent relative motions of spacecraft instead of the ECI frame (Chung et al. 2009; Li et al. 2010), and all relative states of spacecraft were also assumed available in this work.
However in practice, different from absolute states (i.e., positions and velocities), relative states (i.e., relative positions and relative velocities) are generally difficult to directly measure due to the limitation of sensors. Therefore, additional work should be done to obtain these relative quantities. Among varieties of methods to do so, although the direct method that simply subtracts between two irrelevant absolute states could be used to calculate the relative state, the results would be quite rough. In contrast, methods based on relative sensing are preferred. Carrier-phase-based differential global positioning system navigation is one such method that is applicable only for missions near the Earth [e.g., Mitchell (2004)]. For more-general missions where the global positioning system may not be available, such as deep-space missions, several alternative sensing technologies have been developed in recent years. Some common such technologies are based on lasers, charge coupled devices, and high-speed cameras. However, laser-based sensing is of high directionality, and image-based sensing requires complex image processing and target identification—with associated failures. Gunnam et al. (2002) developed a new vision-based navigation (VISNAV) sensor at Texas A&M University (College Station, Texas) for relative navigation. VISNAV uses at least four active beacons, which are at known positions in the body frame of one spacecraft, as light sources. It also uses a position-sensitive diode sensor, which is at a known position in the body frame of the other spacecraft, as the light receiver. Thus after gathering all the line-of-sight vectors generated by light sources and receiver, relative states between the two spacecraft can be estimated. Lau et al. (1996) provided an autonomous formation flying sensor (AFF) for relative navigation, which is built around commercially-available transmitter/receiver antennas at known positions in body frames of each spacecraft. By dealing with the measured pseudorange or phase observables, relative states can be solved to extremely high precision by the estimation. In the context of distributed spacecraft systems, since the vision-based method may be blocked by other spacecraft due to the complexity and flexibility of the formations, AFF is considered as the sensor whose sensing range can be from several meters to several kilometers.
This paper considers the coordinated motion problem of distributed spacecraft systems that consist of one free-flying spacecraft and several maneuvering spacecraft. All spacecraft are equipped with the AFF sensors, and besides, only the free-flying spacecraft has the ability to sense its absolute position, while the maneuvering spacecraft do not. The organization of this paper proceeds as follows. First, some background on algebraic graph theory, definition of coordinates, and AFF sensors are given. Then the measurement equations of the AFF sensor, which will be used for filtering, are derived in terms of both pseudorange and phase. Next, the equation of relative motion of spacecraft is reviewed and presented, and a set of unscented kalman filters are designed for relative estimation in a distributed way. During the design, the initial values of filters are given not by a priori assumption but by the solution provided by a nonlinear minimization algorithm. Next, a consensus control law is designed for the orbital coordinative motion. In the simulation, a low-Earth-orbit satellites flying scenario is taken as an instantiation, and subsequent results and analyses are presented. Finally, concluding remarks are made in the last section.
Backgrounds
Algebraic Graph Theory
The bidirectional information exchange among spacecraft can be modeled by undirected graphs. A weighted undirected graph describes the communication links of a finite nonempty set of nodes with being the edge set. In a weighted undirected graph, an edge denotes that spacecraft and can obtain information from each other, and they are called neighbors. The weighted adjacency matrix of a weighted undirected graph is defined such that is a positive weight if , while if (without loss of generality, it is considered that has been turned into the range of in this paper). As implies , it follows that , . The Laplacian matrix associated with the graph is defined such that , while , . It is clear that is symmetric, and is symmetric, positive, and semidefinite when the undirected graph is connected [see e.g., Ren et al. (2007) for more].
Coordinates
In this paper, the free-flying spacecraft is treated as a reference spacecraft. To describe the absolute position of the reference spacecraft, an ECI frame named J2000 is used, while to describe the relative motion between every two spacecraft, an orbital reference frame named the local vertical/local horizontal (LVLH) frame is used. As shown in Fig. 1, denotes the ECI frame with the origin attached to the mass center of the Earth; denotes the LVLH frame with the origin attached to the mass center of the reference spacecraft; Point denotes the mass center of the th maneuvering spacecraft; and vector denotes the relative position of the th maneuvering spacecraft with respect to the LVLH frame. The LVLH frame is defined such that the axis is outward along the radial (local vertical), is perpendicular to in the orbit plane in the direction of motion (local horizontal), and is along the orbit normal.
Fig. 1. Definition of the coordinates
AFF Sensor
AFF sensor was first devised for the Deep Space Mission 3 (DS-3) (Lau et al. 1996), where the stabilization of a fixed triangle formation is strictly required. In this mission, since the three spacecraft always face towards each other, AFF is configured with only a steradian coverage for the purpose of saving cost and mass. There is also an alternative full configuration for more flexible formations, and in this paper, this full configuration is employed as shown in Fig. 2 for a full steradian coverage. For simplicity, the shape of each spacecraft is simplified to a cube, and coordinate is the body fixed frame of spacecraft (). Each spacecraft is equipped with the AFF consisting of eight antennas, of which six are receivers and two are transmitters at known positions in the frame. Symbol is borrowed to indicate the receivers, while Symbol indicates the transmitters. Superscripts are identifiers of distinct transmitters/receivers, and subscripts indicates the owner spacecraft of these antennas (e.g., means Receiver 3 on spacecraft , and means the Transmitter 1 on spacecraft ). In a single-epoch measurement, ranging codes and carrier phases are generated upon each spacecraft, sent to the neighboring spacecraft, and processed in sequence.
Fig. 2. AFF sensor configuration
Relative Estimation
Measurement Equations
For convenience, the maneuvering spacecraft is labeled with number in turn, and to achieve a unified format, the reference spacecraft is labeled with number 0. Now consider the transmit–receive relations between two such distinct spacecraft and ( and ) that can communicate with each other. Since there are two transmit antennas mounted on opposing corners on one spacecraft, there will always exist at least one transmit antenna that can be received by the other spacecraft. And in the meanwhile, there will always exist at least three active receive antennas on one spacecraft that can receive signals from the other spacecraft. Therefore, the directed transmit–receive paths between spacecraft and are at least six, among which only six will be selected (can be randomly) to take part in the measurement. Such six transmit–receive paths are illustrated in Fig. 3, where the arrows represent the directions of the information flows.
Fig. 3. Illustration of six transmit–receive paths
Pseudorange measurement equations will now be derived. As shown in Fig. 4, without loss of generality, focusing on one general path where the active receive antenna on spacecraft is denoted as and the active transmit antenna on spacecraft is denoted as , , . According to geometric vector addition, one obtains
(1)
where the superscripts of vectors indicate the coordinate frames in which the vectors are expressed, and , , and denote the ECI, LVLH, , and frames, respectively.
Fig. 4. Pseudorange measurement
Denoting the transformation matrix from frame to as , Eq. (1) can be rewritten as
(2)
Then, taking measurement noises and clock errors into consideration, pseudorange measurement equations can be obtained as
(3)
where = 2-norm operator; = clock error between spacecraft and , which is converted to the form of distance; and = measuring noise, which can be modeled to follow the normal distribution. Also note that Eq. (3) represents three equations corresponding to the three paths from spacecraft to .
The paths from spacecraft to are similar, and equations can be obtained by simply switching the variables and in Eq. (3) as
(4)
where . In Eqs. (3) and (4), , , , and are known vectors, and with and are known as the attitude matrices of spacecraft and , is the known transformation matrix from LVLH frame to ECI frame, while and are unknown quantities to be solved.
As shown in Fig. 5, the differential phase measurement is abstracted. Denote and as the two receivers on spacecraft that receive signals from the same transmitter on spacecraft . Then difference the two received signals, the differential phase can be measured. Define line-of-sight vectors as
(5)
Fig. 5. Differential phase measurement
Then the differential phase measurement equations can be expressed as
(6)
where = integer ambiguity; = measurement noise; and = small distance, which can be calculated as
(7)
Since three receive antennas can only make two combinations, there are four differential phase measurements in a single epoch, and hence the equations are four. Combining six pseudorange measurement equations and four differential phase measurement equations yields the final measurement equations.
Relative Motion Equations
As defined previously, the LVLH frame attached to the free flying spacecraft is used as the orbital reference frame in this paper. Although the relative motion can be also described in similar orbital frames attached to any one of the maneuvering spacecraft, the maneuvering spacecraft will be maneuvering in general so each of them would not likely move in a fixed Keplerian orbit. Therefore, it is more accurate to use the LVLH frame attached to the free-flying spacecraft as opposed to the maneuvering spacecraft. Besides, although the relative motion is being studied, the rough orbital absolute information is needed due to the orbital characteristics, so in this paper it is assumed that the free-flying spacecraft is equipped with an absolute position sensor (e.g., the global positioning system) to provide a reference absolute position for the spacecraft cluster.
For simplicity, in this paper, the free-flying spacecraft is assumed to move in a nearly circular orbit. Using Kepler’s third law, the orbit angular velocity can be given by
(8)
where = orbit angular velocity; = geocentric gravitational constant (398, ); and = orbit radius. Denote the relative position of the th maneuvering spacecraft with respect to the reference spacecraft in the LVLH frame as . When the relative position between a maneuvering spacecraft and the reference spacecraft are small compared to the orbit radius, the relative dynamics, known as the CW equations, can be derived as [see, Clohessy and Wiltshire (1960) and Lee et al. (2014) for more]
(9)
where = mass of the th maneuvering spacecraft; , , and = control forces; and , , and = disturbance forces. Although Eq. (9) is an approximation and is not applicable for long-term orbital motion propagation, it can be used to predict the relative state in the next moment with enough precision. Furthermore, motion within the orbital plane (i.e., and axes) are coupled, while the motion perpendicular to the orbital plane (i.e., axis) is independent from that within the orbital plane, and actually it is periodic vibratory. Denote and define , then the equation of can be rewritten in the matrix form as
where is a zero matrix, and is a identity matrix. Furthermore, by defining , , and , the relative equations between two distinct maneuvering spacecraft and ( and ) can be written as
(13)
Fig. 6. The system matrix F in the state differential equation Eq. (10)
Estimation
In this subsection, the method for relative estimations among multiple spacecraft will be presented. First, how the estimation process works at the system level will be shown. Since the spacecraft system is distributed and a host spacecraft does not exist, the estimation could be processed by many small estimators working together in a distributed way. It is clear that the number of small estimators is equal to that of undirected edges in the topology. To describe the allocation of these estimators, the sign matrix is defined such that if an estimator for is running on spacecraft , while otherwise. Define , then the problem of allocating the estimators evenly is equivalent to the problem of trying to average , . Hence a method to get an averaging decision is proposed as follows:
1.
For each spacecraft , initialize , ;
2.
During one communication, for each pair of neighbor spacecraft and , a random decision is made such that either , or , ;
3.
During a next communication, for each pair of neighbor spacecraft and , an improved decision is made by
(14)
where () = moment;
4.
Repeat Step. 3 for a given while. And then by rounding to the nearest integers, one obtains the final decision.
It can be seen from Eq. (14) that if spacecrafts and are neighbors, there comes , so that when (which means the work of spacecraft is heavier than that of spacecraft ), will decrease (meanwhile will increase) to offload a little work from to , and when , then will increase to help share some work from to . Thus after some time, the amount of computations would be well offloaded to every spacecraft. Eq. (14) also constrains that , so after rounding towards the nearest integers, the final decision can be obtained. The gain “” was chosen to guarantee the convergence of the averaging iteration. Actually, by denoting the gain as , then the gain can be chosen as any values satisfying . This choosing principle can be proved as follows. Replace the gain with , and the first equation of Eq. (14) can be written as
(15)
One can obtain by summing Eq. (15) together. Integrate these equations and get the matrix form as
(16)
where , is the -order identity matrix. Denote by the th eigenvalue of in an increasing sequence and by the th eigenvalue of , and it is easy to know that . Choose the gain following the principle of and the graph is connected; then it follows from Ren et al. (2007) that there is a single 0 in the eigenvalues of while all the other eigenvalues are positive but no more than 1, or in formula form, . By doing so, the eigenvalues of satisfy , which implies the spectral radius of is 1. It follows from Horn and Johnson (2012) that, , where is the left eigenvector of corresponding to the eigenvalue . Therefore after enough loops, converges following , where denotes a constant. It means after convergence, which indicates that all the spacecraft finally reach consensus on the division of work.
After offloading computations to every spacecraft by the averaging decision, each spacecraft will then activate a corresponding estimator according to . These estimators are designed based on unscented Kalman filters as follows:
1.
Initialization: At moment 0, initialize the estimation of relative state , process noise matrix , measurement noise matrix , and error covariance matrix . Further analyses and principles that accomplish the initialization will be given following the design.
2.
Sampling weights assignment: The weight assignment follows the minimal skew sampling method (Julier and Uhlmann 2002). Since contains seven quantities, the number of sigma points should be . The weights are chosen as
(17)
with being a initial nonnegative weight, and in this paper .
3.
Unbiased sigma points construction of u: Since the covariance is already reflected in matrices and , the sigma points should be unbiased. Denote the nine sigma points as , respectively, then their values can be given by following iterations
(18)
with three priori values
(19)
4.
Global propagation: At moment , the purpose of the global propagation is to predict roughly based on , and the prediction is denoted as . Then following Eq. (13), the prediction is given by
(20)
where = time interval between estimations; and = state transition matrix. In practice, it is of sufficient precision to preserve the first moment only, and hence one obtains
(21)
5.
Error propagation: The purpose of error propagation is to evaluate the errors caused by uncertainties of during the global propagation. Since Eq. (20) is linear, the prediction itself can be omitted in this process. Based on the unbiased sigma points, perturbations before global propagation can be constructed as
(22)
which will generate propagation errors as
(23)
6.
Measurement prediction: The measurement equations Eqs. (3), (4), and (6) are all functions of relative state , hence for clarity, they are treated as a whole, which is referred to as the following function
(24)
Therefore, the predictions of measurements are given by
(25)
7.
Covariance prediction: Covariances are predicted as
(26)
where
(27)
8.
Update: Once the true measurements occur at moment , the estimation will be updated as
(28)
where
(29)
9.
Assignment for next moment: After the estimation at moment , the following assignments will be done to get ready for the estimation at moment , which are
(30)
The initializations of estimators are assumed to be priori known in general studies, however, this assumption is difficult to realize in practice. In this paper, practical principles of initialization will be presented. contains three quantities which are , , and respectively. Regarding , the work done by Lau et al. (1996) show that clock error could be measured by a built-in approach with the precision of 0.1 ns, which is equivalent 0.03 m after being multiplied by the speed of light. Therefore, can be initialized with the bound . Regarding , the cluster of distributed spacecraft are assumed have been already launched approximately to a common orbit such that they can communicate with neighbors, and hence, their initial relative velocities are bounded. The bound can be approximately estimated simply. Since with bigger than the Earth’s radius, smaller than the first cosmic velocity, and no bigger than several kilometers (in fact, the maximum effective distance of AFF is commonly 2 km) in general, there comes the inequality that . Therefore, can be initialized with the bound . Regarding , it can be solved roughly based on Eqs. (3) and (4) ignoring the small . Thus the initialization of becomes the optimization problem optimizing to make the results of Eqs. (3) and (4) best fit the pseudorange measurements. Since the number of equations (i.e., 6) is more than the number of variables (i.e., 3), this problem is a determined system. Therefore, many nonlinear minimization methods can be used for help. In this paper, the trust-region method is used to make a large-scale optimization, and then can be initialized to the optimization result with a confidence bound . Detailed explanations about the algorithm procedure are omitted here due to its universal applicability, which can be found in many good articles, e.g., Coleman and Li (1996) and Conn et al. (2000). So can be initialized until now, and subsequently , and will be initialized as
(31)
where , , and = confidence bound of the disturbance forces respectively; = confidence bound of pseudorange measurement noises; = confidence bound of differential phase measurement noises; and the constant factor 2 here means that corresponding covariances are doubled when differencing and .
Distributed Control Based on Estimations
Analyses on Design of Control Law
In this section, a distributed control law will be proposed for relative orbital motion. Suppose a formation configuration is desired for the spacecraft cluster, and it is described by a set of , , , which is to say, denotes the desired relative position between spacecraft and . Disturbance forces are generally considered as external disturbances, and hence the system dynamics used in design of the control law can be obtained from the governing dynamic Eq. (9) by ignoring the disturbance terms as
(32)
where
(33)
It can be seen that the dynamics presented by Eq. (32) contain the two orbital terms and , which are generated by uncontrolled orbital motion. And hence, he effects caused should be eliminated by these terms as much as possible. However, although the differential relative states and can be obtained by estimation, and themselves are not direct available to spacecraft due to the fact spacecraft will only communicate with its neighbors and could not know the global configuration. Hence, prior to the design of control law, a distributed method is proposed, which is inspired by Cao et al. (2010) and Meng et al. (2010), to help in obtaining the estimation of .
Observing Based on
The observing method is proposed as
(34)
where = signum function; = estimation of ; = estimation of and is already known after the process shown in the previous section; and = positive gain satisfying . The theoretical proof of the method is as follows. Denote , , and then Eq. (34) can be rewritten as
(35)
where the fact that is used to develop the equality. Defining and , then Eq. (35) can be expressed in the matrix form as
(36)
where is the Kronecker product operator. The subsequent proof can follow the process given by Meng et al. (2010). Since is positive and semidefinite, consider the Lyapunov function candidate
(37)
Then the derivation of along the trajectories of Eq. (36) is
(38)
where the Holders inequality is used to obtain the first Holder’s inequality and the fact is used to obtain the last inequality. It then follows from Lyapunov stability theory that Eq. (35) is asymptotically stable and as , and by Lasalle’s Invariance principle, it follows that , as . So there must exist such a time that satisfies , with the margin of errors very small for all . In fact, it can be concluded from Cao and Ren (2012) that .
Design of Control Law
The control law is proposed as follows:
(39)
where = positive gain of the position term; = positive gain of the velocity term; and = hyperbolic tangent function and is used here to limit the output since the control forces are bounded in practice [e.g., Meng et al. (2013) and Yang et al. (2014)]. Using control law Eq. (40) for system dynamics Eq. (33), then the system is asymptotically stable. The proof is shown in the following. For , it is clear that system is always bounded, and for , by combining Eqs. (32) and (39), one obtains
(40)
where term counteracts the term since for . Consider the Lyapunov function candidate
(41)
where = natural logarithm function; and = hyperbolic cosine function. Then the derivation of along the trajectories of Eq. (40) is
(42)
By switching the dummy variables and in the summation function, one obtains
(43)
(44)
also, it can be seen from the definition that , then Eq. (42) can be developed as
(45)
As and are odd functions, and then one can obtain
(46)
So the system is asymptotically stable and the equilibrium is achieved when and , as , and thus the formation configuration of the spacecraft system will be formed.
Simulation
A spacecraft system that contains one free-flying spacecraft and four maneuvering spacecraft is taken as an example. A general communication topology is set for the system as shown in Fig. 7, which is fixed, undirected, and connected with communication link edges weighted by 0.8. The five spacecraft are set such that they move in a line initially in a same nearly-circular orbit with orbital semi-major axis 7378.137 km, eccentricity , inclination 60°, and right ascension of ascending node 0°. The desired formation offsets and other parameters are given as shown in Table 1. Denote the time when estimation is started as time origin, then the simulation is scheduled as follows. The establishment of communication links and allocation of estimators are processed during , all estimators and all observing processes for rough work during , and the distributed control law works during . Simulation results are as follows.
At time , a random decision is made such that . Following the definition of matrix , it can be seen that Spacecrafts 1, 3, and 5 are idle, while Spacecrafts 2 and 4 both carry out two estimators. At time after the allocation, final decision is obtained and , which shows that the four estimators are successfully distributed evenly with Spacecraft 0 estimating , Spacecraft 1 estimating , Spacecraft 3 estimating , and Spacecraft 4 estimating .
Fig. 8 shows the 2D projections of trajectories of the four maneuvering spacecraft onto the plane of the LVLH frame, where the points of the trajectories represent the mass centers of the spacecraft. Fig. 9 shows the detailed 3D position data. From both Figs. 8 and 9, it can be seen that the four maneuvering spacecraft have been successfully controlled to the desired formation configuration. Furthermore, the configuration errors are no more than 1 cm. Controlling is based on estimation, and the estimation results are as follows. During time , the position initializations of the four estimators are obtained by the trust-region method, and results as shown in Table 2. Although the optimization processing does not take clock errors and measurement errors into consideration, and hence the results are rough, it still provide a good initialization (because errors are bounded by ) for further estimation. During time , the relative-position estimation errors and relative-velocity estimation errors are shown in Figs. 10 and 11 respectively, from which the following conclusion can be drawn. The errors of relative-position estimation are no more than 0.2 m throughout all the time after convergence and are much smaller when relative motion becomes slight, and the errors of relative-velocity estimation are no more than . The results of the estimation for clock errors are similar, and they are omitted to save space. The estimation errors are very small, showing that the estimation is effective and helpful to controlling. Meanwhile during time , there are also four observers trying to observe , , and the errors of the observing results are shown in Fig. 12, from which it can be seen that the errors have been decreased to no more than 10 m before . In the control law (40) will be multiplied by matrix and meanwhile the elements of are small, then it is clear and make little difference to the control law.
Fig. 8. 2D projections of motion trajectories of the four maneuvering spacecraft onto the plane
Fig. 9. 3D positions of the four spacecraft during maneuvering
Table 2. Optimization for Position Initialization
Relative position
Seeking origin
Optimization
True value
Fig. 10. Relative-position estimation errors
Fig. 11. Relative-velocity estimation errors
Fig. 12. Observation errors
Conclusion
In this paper, coordinated motion problem of distributed spacecraft systems that consist of one free-flying spacecraft and several maneuvering spacecraft is considered, and a relatively complete processing framework is proposed for the coordinated motion. In the upper level, a distributed algorithm is proposed to allocate the estimators, helping offloading computations to every spacecraft. In the lower level, estimators, observers, and controllers are all designed in a distributed way. Among them, estimators are designed based on Kalman fiters and solutions of initialization problems are also presented, observers work for rough positions of maneuvering spacecraft in order to provide good approximations for controlling, and controllers work based on consensus and only use the results of estimators and observers to achieve coordinative motion. Finally, simulations are carried out where a low-Earth-orbit satellite flying scenario is taken as an instantiation. Simulation results show that estimators, observers, and controllers work well with fixed, connected, and undirected communication topologies and configuration can be formed with centimeter-level accuracy, having demonstrated the effectiveness of proposed approach.
Acknowledgments
The authors would like to thank the support by the National High Technology Research and Development Program of China (Grant No. 2013AA122601).
References
Cao, Y., and Ren, W. (2012). “Distributed coordinated tracking with reduced interaction via a variable structure approach.” IEEE Trans. Automat. Contr., 57(1), 33–48.
Cao, Y., Ren, W., and Meng, Z. (2010). “Decentralized finite-time sliding mode estimators and their applications in decentralized finite-time formation tracking.” Syst. Control Lett., 59(9), 522–529.
Chung, S. J., Ahsun, U., and Slotine, J. J. E. (2009). “Application of synchronization to formation flying spacecraft: Lagrangian approach.” J. Guid. Control Dyn., 32(2), 512–526.
Gunnam, K. K., Hughes, D. C., Junkins, J. L., and Kehtarnavaz, N. (2002). “A vision-based DSP embedded navigation sensor.” Sens. J. IEEE, 2(5), 428–442.
Jadbabaie, A., Lin, J., and Morse, A. S. (2003). “Coordination of groups of mobile autonomous agents using nearest neighbor rules.” IEEE Trans. Automat. Contr., 48(6), 988–1001.
Julier, S. J., and Uhlmann, J. K. (2002). “Reduced sigma point filters for the propagation of means and covariances through nonlinear transformations.” Proc., American Control Conf., IEEE, 887–892.
Lau, K., Lichten, S., Young, L., and Haines, B. (1996). “An innovative deep space application of gps technology for formation flying spacecraft.” Proc., AIAA GN&C Conf., AIAA, Reston, VA, 1–9.
Lee, D., Bang, H., Butcher, E. A., and Sanyal, A. K. (2014). “Kinematically coupled relative spacecraft motion control using the state-dependent riccati equation method.” J. Aerosp. Eng., 04014099.
Li, Z., Duan, Z., and Chen, G. (2010). “Consensus of multiagent systems and synchronization of complex networks: A unified viewpoint.” IEEE Trans. Circuits Syst., 57(1), 213–224.
Meng, Z., Ren, W., and You, Z. (2010). “Distributed finite-time attitude containment control for multiple rigid bodies.” Automatica, 46(12), 2092–2099.
Meng, Z., Zhao, Z., and Lin, Z. (2013). “On global leader-following consensus of identical linear dynamic systems subject to actuator saturation.” Syst. Control Lett., 62(2), 132–142.
Ren, W., Beard, R. W., and Atkins, E. M. (2007). “Information consensus in multivehicle cooperative control: Collective group behavior through local interaction.” IEEE Contr. Syst. Mag., 27(2), 71–82.
Yang, T., Meng, Z., Dimarogonas, D. V., and Johansson, K. H. (2014). “Global consensus for discrete-time multi-agent systems with input saturation constraints.” Automatica, 50(2), 499–506.
Assistant Research Scientist, Qian Xuesen Laboratory of Space Technology, China Academy of Space Technology, No. 104 Youyi Rd., Beijing 100094, China; formerly, State Key Laboratory of Precision Measurement Technology and Instruments, Dept. of Precision Instrument, Tsinghua Univ., Beijing 100084, China (corresponding author). E-mail: [email protected]; [email protected]
Assistant Professor, State Key Laboratory of Precision Measurement Technology and Instruments, Dept. of Precision Instrument, Tsinghua Univ., Beijing 100084, China. E-mail: [email protected]
Professor, State Key Laboratory of Precision Measurement Technology and Instruments, Dept. of Precision Instrument, Tsinghua Univ., Beijing 100084, China. E-mail: [email protected]
If you have the appropriate software installed, you can download article citation data to the citation manager of your choice. Simply select your manager software from the list below and click Download.