This work was supported by National Natural Science Foundation of China (Grant Nos. 61473099, 61333001). The Titan Xp used for the RNNs training is donated by the NVIDIA Corporation.
[1] Yao Y, Zheng T Y, He F H, et al. Several hot issues and challenges in terminal guidance of flight vehicles. Acta Aeronaut Et Astronaut Sin, 2015, 36: 2696--2716. Google Scholar
[2] Zhai C, He F, Hong Y. Coverage-Based Interception Algorithm of Multiple Interceptors Against the Target Involving Decoys. J Guidance Control Dyn, 2016, 39: 1647-1653 CrossRef ADS Google Scholar
[3] Lu P. Asymptotic Analysis of Quasi-Equilibrium Glide in Lifting Entry Flight. J Guidance Control Dyn, 2006, 29: 662-670 CrossRef ADS Google Scholar
[4] Lu P. Entry Guidance: A Unified Method. J Guidance Control Dyn, 2014, 37: 713-728 CrossRef ADS Google Scholar
[5] Jorris T R, Cobb R G. Three-Dimensional Trajectory Optimization Satisfying Waypoint and No-Fly Zone Constraints. J Guidance Control Dyn, 2009, 32: 551-572 CrossRef ADS Google Scholar
[6] Liu X, Shen Z, Lu P. Entry Trajectory Optimization by Second-Order Cone Programming. J Guidance Control Dyn, 2016, 39: 227-241 CrossRef ADS Google Scholar
[7] Zhao J, Zhou R. Reentry trajectory optimization for hypersonic vehicle satisfying complex constraints. Chin J Aeronautics, 2013, 26: 1544-1553 CrossRef Google Scholar
[8] Rong Li X, Jilkov V P. Survey of maneuvering targettracking . part I: dynamic models. IEEE Trans Aerosp Electron Syst, 2003, 39: 1333-1364 CrossRef ADS Google Scholar
[9] Singer R. Estimating Optimal Tracking Filter Performance for Manned Maneuvering Targets. IEEE Trans Aerosp Electron Syst, 1970, AES-6: 473-483 CrossRef ADS Google Scholar
[10] Kumar K S P, Zhou H. A 'current' statistical model and adaptive algorithm for estimating maneuvering targets. J Guidance Control Dyn, 1984, 7: 596-602 CrossRef ADS Google Scholar
[11] X Rong Li , Jilkov V P. Survey of maneuvering target tracking. part v: multiple-model methods. IEEE Trans Aerosp Electron Syst, 2005, 41: 1255-1321 CrossRef ADS Google Scholar
[12] Li S, Lei H, Shao L. Multiple Model Tracking for Hypersonic Gliding Vehicles With Aerodynamic Modeling and Analysis. IEEE Access, 2019, 7: 28011-28018 CrossRef Google Scholar
[13] Vasuhi S, Vaidehi V. Target tracking using Interactive Multiple Model for Wireless Sensor Network. Inf Fusion, 2016, 27: 41-53 CrossRef Google Scholar
[14] Zhou Y, Li J X, Wang D L. Target tracking in wireless sensor networks using adaptive measurement quantization. Sci China Inf Sci, 2012, 55: 827-838 CrossRef Google Scholar
[15] Yu Y, Peng S, Dong X. UIF-based cooperative tracking method for multi-agent systems with sensor faults. Sci China Inf Sci, 2019, 62: 010202 CrossRef Google Scholar
[16] García J, Besada J A, Molina J M. Model-based trajectory reconstruction with IMM smoothing and segmentation. Inf Fusion, 2015, 22: 127-140 CrossRef Google Scholar
[17] Yuan Y, Cheng L, Wang Z. Position tracking and attitude control for quadrotors via active disturbance rejection control method. Sci China Inf Sci, 2019, 62: 010201 CrossRef Google Scholar
[18] Chen Y S. A distributed estimator for on-road target tracking with lane estimation and identification. Sci China Inf Sci, 2010, 53: 2495-2505 CrossRef Google Scholar
[19] Lan J, Li X R, Jilkov V P. Second-Order Markov Chain Based Multiple-Model Algorithm for Maneuvering Target Tracking. IEEE Trans Aerosp Electron Syst, 2013, 49: 3-19 CrossRef ADS Google Scholar
[20] Elzoghby M, Li F, Arif U, et al. Small UAV localization based strong tracking filters augmented with interacting multiple model. In: Proceedings of the 15th International Bhurban Conference on Applied Sciences and Technology (IBCAST), 2018, 310--317. Google Scholar
[21] Zhang Y, Xiao M, Wang Z. Robust three-stage unscented Kalman filter for Mars entry phase navigation. Inf Fusion, 2019, 51: 67-75 CrossRef Google Scholar
[22] Lipton Z C, Berkowitz J, Elkan C. A critical review of recurrent neural networks for sequence learning. 2015,. arXiv Google Scholar
[23] Coskun H, Achilles F, DiPietro R, et al. Long short-term memory kalman filters: recurrent neural estimators for pose regularization. In: Proceedings of International Conference on Computer Vision, 2017. 5524--5532. Google Scholar
[24] Karkus P, Hsu D, Lee W S. Particle filter networks with application to visual localization. In: Proceedings of Conference on Robot Learning, 2018. 169--178. Google Scholar
[25] Vinh N X, Busemann A, Culp R D. Hypersonic and Planetary Entry Flight Mechanics. NASA Sti/Recon Technical Report A, 1980. Google Scholar
[26] Vinh N X, Ma D M. Optimal multiple-pass aeroassisted plane change. Acta Astronaut, 1990, 21: 749-758 CrossRef ADS Google Scholar
[27] Gers F A, Schmidhuber E. LSTM recurrent networks learn simple context-free and context-sensitive languages.. IEEE Trans Neural Netw, 2001, 12: 1333-1340 CrossRef PubMed Google Scholar
[28] Sunahara Y. An Approximate Method of State Estimation for Nonlinear Dynamical Systems. J Basic Eng, 1970, 92: 385 CrossRef Google Scholar
[29] Jorris T R. Common Aero Vehicle Autonomous Reentry Trajectory Optimization Satisfying Waypoint and No-fly Zone Constraints. Technical Report, 2007. Google Scholar
[30] Kingma D P, Ba J. Adam: a method for stochastic optimization. 2014,. arXiv Google Scholar
Figure 1
(Color online) Data preprocessing.
Figure 2
(Color online) The architecture of the recurrent neural networks.
Figure 3
Active switching estimation and fusion algorithm.
Figure 4
Training results. (a) $\mathrm{RNN}_{\mathrm{t}}$; (b) $\mathrm{RNN}_{1,\mathrm{QEG}}$; (c) $\mathrm{RNN}_{2,\mathrm{QEG}}$; (d) $\mathrm{RNN}_{1,\mathrm{SG}}$; (e) $\mathrm{RNN}_{2,\mathrm{SG}}$; (f) $\mathrm{RNN}_{3}$.
Figure 5
(Color online) Recognition results. (a) Scenario 1; (b) scenario 2; (c) scenario 3.
Figure 6
(Color online) Estimation results of different model selection strategies. (a) Position estimation of scenario 1; (b) velocity estimation of scenario 1; (c) position estimation of scenario 2; (d) velocity estimation of scenario 2; (e) position estimation of scenario 3; (f) velocity estimation of scenario 3.
Figure 7
(Color online) Model usage of different model selection strategies. (a) Scenario 1; (b) scenario 2; (c) scenario 3.
Figure 8
(Color online) Comparison between ASMM and IMM. (a) Position estimation of scenario 1; (b) velocity estimation of scenario 1; (c) position estimation of scenario 2; (d) velocity estimation of scenario 2; (e) position estimation of scenario 3; (f) velocity estimation of scenario 3.
Labels | QEG | SG |
$L_{1}~=~0$ | $\lambda_{\mathrm{QEG1}}~\in~\left[0.25,0.625\right)$ | $\lambda_{\mathrm{SG1}}~\in~\left[0.25,0.625\right)$ |
$L_{1}~=~1$ | $\lambda_{\mathrm{QEG1}}~\in~\left[0.625,1.125\right)$ | $\lambda_{\mathrm{SG1}}~\in~\left[0.625,1.125\right)$ |
$L_{1}~=~2$ | $\lambda_{\mathrm{QEG1}}~\in~\left[1.125,1.625\right)$ | $\lambda_{\mathrm{SG1}}~\in~\left[1.125,1.625\right)$ |
$L_{1}~=~3$ | $\lambda_{\mathrm{QEG1}}~\in~\left[1.625,2\right]$ | $\lambda_{\mathrm{SG1}}~\in~\left[1.625,2\right]$ |
$L_{2}~=~0$ | $\lambda_{\mathrm{QEG2}}~\in~\left[2,2.75\right)$ | $\lambda_{\mathrm{SG2}}~\in~\left[0.5,0.875\right)$ |
$L_{2}~=~1$ | $\lambda_{\mathrm{QEG2}}~\in~\left[2.75,3.25\right)$ | $\lambda_{\mathrm{SG2}}~\in~\left[0.875,1.125\right)$ |
$L_{2}~=~2$ | $\lambda_{\mathrm{QEG2}}~\in~\left[3.25,4.25\right)$ | $\lambda_{\mathrm{SG2}}~\in~\left[1.125,1.625\right)$ |
$L_{2}~=~3$ | $\lambda_{\mathrm{QEG2}}~\in~\left[4.25,5\right]$ | $\lambda_{\mathrm{SG2}}~\in~\left[1.625,2\right]$ |
/* Initialize the preprocessed measurement. |
Get preprocessed measurements $\tilde{z}_k$ using ( |
/* Trajectory estimation. |
Get preprocessed measurements $\tilde{z}_k$ using ( |
Sequence of preprocessed measurements $\tilde{z}^{N}_{k}~=~[\tilde{z}_{k-N+1},\tilde{z}_{k-N+2},\ldots,\tilde{z}_{k}]$; |
Recognize label $L_{\mathrm{t},k}$, using ( |
Recognize label $L_{3,k}$, using ( |
Recognize label $L_{1,k}$, using ( |
Recognize label $L_{2,k}$, using ( |
Calculate probability of each motion behavior using ( |
Select models to be used in the estimation (get $I^{\mathrm{s}}_{k}$) according to the probabilities of motion behaviors; |
|
Estimate $\hat{x}^{i}_{k|k}$ and $P^{i}_{k|k}$ using ( |
|
Get the final estimate results $\hat{x}_{k|k}$ and $P_{k|k}$ using ( |
Accuracy on the training set (%) | Accuracy on the test set (%) | |
$\mathrm{RNN}_{\mathrm{t}}$ | 94.92 | 95.11 |
$\mathrm{RNN}_{1,\mathrm{QEG}}$ | 96.78 | 88.25 |
$\mathrm{RNN}_{1,\mathrm{SG}}$ | 93.18 | 83.83 |
$\mathrm{RNN}_{2,\mathrm{QEG}}$ | 93.10 | 90.39 |
$\mathrm{RNN}_{2,\mathrm{SG}}$ | 90.08 | 79.41 |
$\mathrm{RNN}_{3}$ | 97.23 | 92.40 |
Item | Value |
Initial height, $h_0$ (m) | 60000 |
Initial longitude, $\theta_0$ (rad) | 0 |
Initial latitude, $\phi_0$ (rad) | 0 |
Initial velocity, $v_0$ (m/s) | 7000 |
Initial flight-path angle, $\gamma_0$ (rad) | $-$1 |
Initial heading angle, $\psi_0$ (rad) | 90 |
End up velocity, $v_{\mathrm{f}}$ (m/s) | 2000 |
Variance of height, $\mathrm{Var}_h$ (m$^2$) | $4.9\times10^4$ |
Variance of longitude, $\mathrm{Var}_\theta$ (rad$^2$) | $1\times10^{-5}$ |
Variance of latitude, $\mathrm{Var}_\phi$ (rad$^2$) | $1\times10^{-5}$ |
Scenario | All (ms/step) | Certainty (ms/step) | Top-5 (ms/step) | CP (ms/step) |
1 | 68.794 | 32.933 | 34.480 | 33.765 |
2 | 69.459 | 32.898 | 34.603 | 33.478 |
3 | 69.690 | 33.897 | 35.471 | 34.806 |