臺灣博碩士論文加值系統

無人機以其靈活的飛行特性、低廉的運維成本成為人們日常生活的重要組成部分。嗯，這些無人機正在從玩具到威脅武器的許多應用中發展。導航是自主無人機時代最重要的角色。獲得一條可行且最安全的路徑是近期研究中最具挑戰性的任務。我們專注於在人口稠密的城市環境中安全可靠的無人機導航，旨在高效地跟隨道路。在本文中，我們提出了一種深度學習 (DL) 方法，通過在動態城市環境中跟隨道路自主導航無人機。無人機配備了前後兩個攝像頭，用於感知環境。卷積神經網絡 (CNN) 通過注意力和殘差機制從無人機的原始視覺輸入中提取高級特徵表示。 CNN 的並發管道，用於三個原始視覺輸入，輸出高級特徵，然後遵循並發特徵融合 (CFF) 到封裝動態和靜態信息的自我頭部注意 (SHA)，以預測偏航和線速度.這些預測被調製成最終控制命令，以便為無人機提供平穩和連續的輸入。此外，實驗評估表明，所提出的方法學習了無人機的導航策略以進行道路跟隨，訓練後的導航策略在不同場景的模擬中具有高性能。此外，我們提出的方法也優於最先進的 ImageNet 模型和現有模型。

Drones are vital part in people's daily lives because of its flexible nature of flying, low cost of operation and maintenance. Well, these UAVs are developing in many applications from toys to threaten weapons. Navigation is the most important role in era of autonomous drones. To attain a feasible and safest path is most challenging task in recent research. We focused on safe and reliable UAV navigation in highly populated city environment by aiming to follow road efficiently. In this article, we presented a Deep Learning (DL) method to navigate drone autonomously by following road in dynamic urban environment. The drone is equipped with two cameras, facing front and down, to perceive the environment. A Convolutional Neural Network (CNN) extracts the high-level feature representation through Attention and residual mechanism from the drone's raw visual inputs. Concurrent pipelines of the CNN, for three raw visual inputs, outputs the high-level features and then follow Concurrent Feature-Fusion (CFF) into Self-Head Attention (SHA) that encapsulate both dynamic and static information to predict the Yaw and Linear velocity. These predictions modulated into final control commands in order to have smooth and continuous inputs to drone. In addition, experimental evaluation shows that, the proposed method learns the drone's navigation strategy for road-following, high performance of the trained navigation policy in simulations with different scenarios. Moreover, our proposed approach outperforms the existing models and state-of-the-art ImageNet models as well.

Chinese Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i
English Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
Acknowledgement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1 GPS or Waypoint Based Navigation . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Sensor Based Navigation with Various Techniques . . . . . . . . . . . . . . . 8
2.3 Existing Methods with Optimized Technique . . . . . . . . . . . . . . . . . . 11
3 System Model and Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.1 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.1.1 Road Follow Navigation Model . . . . . . . . . . . . . . . . . . . . . 14
3.1.2 Collision Avoidance Model . . . . . . . . . . . . . . . . . . . . . . . 15
3.1.3 Energy Consumption Model . . . . . . . . . . . . . . . . . . . . . . . 16
3.2 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.2.1 Drone Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.2.2 Deep Learning Framework . . . . . . . . . . . . . . . . . . . . . . . . 20
4 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.1 Proposed Method RF-CFFA . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.2 Control Systems Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
5 Dataset and Model Implementation . . . . . . . . . . . . . . . . . . . . . . . . . 36
5.1 Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
5.2 Balanced Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
5.3 Network Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
6 Experiments and Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
6.1 Evaluation Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.1.1 Training Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.1.2 Test Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
6.1.3 Flight Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
6.2 Training and Validation Performance . . . . . . . . . . . . . . . . . . . . . . . 49
6.3 Evaluation Performance over Testing Data . . . . . . . . . . . . . . . . . . . . 51
6.4 Qualitative Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
6.5 Experimental Performance in AirSim Simulator . . . . . . . . . . . . . . . . . 58
6.5.1 Simulation Results at 30 m Altitude for Navigation Path of 2000m . . . 62
6.5.2 Simulation Results at 50 m Altitude for Navigation Path of 2000m . . . 65
6.5.3 Simulation Results at 80 m Altitude for Navigation Path of 2000m . . . 67
6.5.4 Simulation Results at 30 m Altitude for Navigation Path of 1200m . . . 70
6.5.5 Simulation Results at 50 m Altitude for Navigation Path of 1200m . . . 72
6.5.6 Simulation Results at 80 m Altitude for Navigation Path of 1200m . . . 75
6.5.7 Simulation Results at 30 m Altitude for Navigation Path of 700m . . . 77
6.5.8 Simulation Results at 50 m Altitude for Navigation Path of 700m . . . 80
6.5.9 Simulation Results at 80 m Altitude for Navigation Path of 700m . . . 83
6.6 Ablation Study on Proposed Method RF-CFFA with Different Input Image Size 85
7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
Extended Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
External References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

[1] P. Kopardekar, J. Rios, T. Prevot, M. Johnson, J. Jung, and J. E. Robinson, “Unmanned
aircraft system traffic management (utm) concept of operations,” in AIAA Aviation and
Aeronautics Forum (Aviation 2016), no. ARC-E-DAA-TN32838, pp. 1–16, Jun. 2016.
[2] P. Chhikara, R. Tekchandani, N. Kumar, V. Chamola, and M. Guizani, “Dcnn-ga: A deep
neural net architecture for navigation of uav in indoor environment,” IEEE Internet of
Things Journal, vol. 8, no. 6, pp. 4448–4460, Sep. 2020.
[3] M. A. Arshad, S. H. Khan, S. Qamar, M. W. Khan, I. Murtza, J. Gwak, and A. Khan, “Drone
navigation using region and edge exploitation-based deep cnn,” IEEE Access, vol. 10, pp.
95 441–95 450, Sep. 2022.
[4] B. G. Maciel-Pearson, S. Akçay, A. Atapour-Abarghouei, C. Holder, and T. P. Breckon,
“Multi-task regression-based learning for autonomous unmanned aerial vehicle flight control within unstructured outdoor environments,” IEEE Robotics and Automation Letters,
vol. 4, no. 4, pp. 4116–4123, July. 2019.
[5] YilinLiu, KeXie, and HuiHuang, “Vgf-net: Visual-geometric fusion learning for simultaneous drone navigation and height mapping,” Graphical Models, vol. 116, p. 101108,
May. 2021.
[6] A. Loquercio, A. I. Maqued, C. R. del Blanco, and D. Scaramuzza, “Dronet: Learning to
fly by driving,” IEEE Robotics and Automation Letters, vol. 3, no. 2, pp. 1088–1095, Jan.
2018.
[7] C. Tang, Y. Wang, L. Zhang, and Y. Zhang, “Gnss/inertial navigation/wireless station fusion uav 3-d positioning algorithm with urban canyon environment,” IEEE Sensors Journal, vol. 22, no. 19, pp. 18 771–18 779, Aug. 2022.
90
[8] X. Liu, X. Liu, W. Zhang, and Y. Yang, “Interacting multiple model uav navigation algorithm based on a robust cubature kalman filter,” IEEE Access, vol. 8, pp. 81 034–81 044,
Apr. 2020.
[9] H. Bai and C. N. Taylor, “Future uncertainty-based control for relative navigation in gpsdenied environments,” IEEE Transactions on Aerospace and Electronic Systems, vol. 56,
no. 5, pp. 3491–3501, Feb. 2020.
[10] H. Bai and R. W. Beard, “Relative heading estimation and its application in target handoff
in gps-denied environments,” IEEE Transactions on Control Systems Technology, vol. 27,
no. 1, pp. 74–85, Nov. 2017.
[11] E. Vitali, D. Gadioli, G. Palermo, M. Golasowski, J. Bispo, P. Pinto, J. MartinoviČ,
K. SlaninovÁ, J. M. Cardoso, and C. Silvano, “An efficient monte carlo-based probabilistic time-dependent routing calculation targeting a server-side car navigation system,”
IEEE Transactions on Emerging Topics in Computing, vol. 9, no. 2, pp. 1006–1019, July.
2019.
[12] Y. Yang, J. Khalife, J. J. Morales, and Z. M.Kassas, “Uav waypoint opportunistic navigation in gnss-denied environments,” IEEE Transactions on Aerospace and Electronic
Systems, vol. 58, no. 1, pp. 663–678, Aug. 2021.
[13] B. Huang, P. Feng, J. Zhang, D. Yu, and Z. Wu, “A novel positioning module and fusion
algorithm for unmanned aerial vehicle monitoring,” IEEE Sensors Journal, vol. 21, no. 20,
pp. 23 006–23 023, Aug. 2021.
[14] Ó. Gil, A. Garrell, and A. Sanfeliu, “Social robot navigation tasks: Combining machine
learning techniques and social force model,” Sensors, vol. 21, no. 21, p. 7087, Oct. 2021.
[15] S. J. Haddadi and E. B. Castelan, “Visual-inertial fusion for indoor autonomous navigation
of a quadrotor using orb-slam,” in 2018 Latin American Robotic Symposium, 2018 Brazilian Symposium on Robotics (SBR) and 2018 Workshop on Robotics in Education (WRE),
pp. 106–111, Dec. 2018.
[16] Y. Wu and J. Zhao, “A robust and precise lidar-inertial-gps odometry and mapping method
for large-scale environment,” IEEE/ASME Transactions on Mechatronics, vol. 27, no. 6,
pp. 5027–5036, May. 2022.
91
[17] V. Roberge, M. Tarbouchi, and G. Labonté, “Comparison of parallel genetic algorithm
and particle swarm optimization for real-time uav path planning,” IEEE Transactions on
Industrial Informatics, vol. 9, no. 1, pp. 132–141, May. 2012.
[18] K. Zhu and T. Zhang, “Deep reinforcement learning based mobile robot navigation: A
review,” Tsinghua Science and Technology, vol. 26, no. 5, pp. 674–691, Apr. 2021.
[19] O. Doukhi and D. J. Lee, “Deep reinforcement learning for autonomous map-less navigation of a flying robot,” IEEE Access, vol. 10, pp. 82 964–82 976, Mar. 2022.
[20] B. Li and Y. Wu, “Path planning for uav ground target tracking via deep reinforcement
learning,” IEEE Access, vol. 8, pp. 29 064–29 074, Feb. 2020.
[21] D. Li, W. Yin, W. E. Wong, M. Jian, and M. Chau, “Quality-oriented hybrid path planning
based on a* and q-learning for unmanned aerial vehicle,” IEEE Access, vol. 10, pp. 7664–
7674, Dec. 2021.
[22] C. Wang, J. Wang, J. Wang, and X. Zhang, “Deep-reinforcement-learning-based autonomous uav navigation with sparse rewards,” IEEE Internet of Things Journal, vol. 7,
no. 7, pp. 6180–6190, Feb. 2020.
[23] A. Singla, S. Padakandla, and S. Bhatnagar, “Memory-based deep reinforcement learning
for obstacle avoidance in uav with limited environment knowledge,” IEEE Transactions
on Intelligent Transportation Systems, vol. 22, no. 1, pp. 107–118, Nov. 2019.
[24] D. Hong, S. Lee, Y. H. Cho, D. Baek, J. Kim, and N. Chang, “Energy-efficient online
path planning of multiple drones using reinforcement learning,” IEEE Transactions on
Vehicular Technology, vol. 70, no. 10, pp. 9725–9740, Aug. 2021.
[25] V. J. Hodge, R. Hawkins, and R. Alexander, “Deep reinforcement learning for drone navigation using sensor data,” Neural Computing and Applications, vol. 33, no. 6, pp. 2015–
2033, Jun. 2021.
[26] S. D. Morad, R. Mecca, R. P. Poudel, S. Liwicki, and R. Cipolla, “Embodied visual navigation with automatic curriculum learning in real environments,” IEEE Robotics and Automation Letters, vol. 6, no. 2, pp. 683–690, Jan. 2021.
92
[27] F. Schilling, J. Lecoeur, F. Schiano, and D. Floreano, “Learning vision-based flight in
drone swarms by imitation,” IEEE Robotics and Automation Letters, vol. 4, no. 4, pp.
4523–4530, Aug. 2019.
[28] J. Jagannath, A. Jagannath, S. Furman, and T. Gwin, “Deep learning and reinforcement
learning for autonomous unmanned aerial systems: Roadmap for theory to deployment,”
in Deep Learning for Unmanned Systems, vol. 984, pp. 25–82, Oct. 2021.
[29] A. Kouris and C.-S. Bouganis, “Learning to fly by myself: A self-supervised cnn-based
approach for autonomous navigation,” in 2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 1–9, Jan. 2018.
[30] N. Smolyanskiy, A. Kamenev, J. Smith, and S. Birchfield, “Toward low-flying autonomous
mav trail navigation using deep neural networks for environmental awareness,” in 2017
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 4241–
4247. IEEE, Dec. 2017.
[31] S. S. Khodaparast, X. Lu, P. Wang, and U. T. Nguyen, “Deep reinforcement learning based
energy efficient multi-uav data collection for iot networks,” IEEE Open Journal of Vehicular Technology, vol. 2, pp. 249–260, Jun. 2021.
[32] X. Wang, M. C. Gursoy, T. Erpek, and Y. E. Sagduyu, “Learning-based uav path planning
for data collection with integrated collision avoidance,” IEEE Internet of Things Journal,
vol. 9, no. 17, pp. 16 663–16 676, Sep. 2022.
[33] M. Labbadi and M. Cherkaoui, “Robust adaptive backstepping fast terminal sliding mode
controller for uncertain quadrotor uav,” Aerospace Science and Technology, vol. 93, p.
105306, July. 2019.
[34] Z. Weidong, Z. Pengxiang, W. Changlong, and C. Min, “Position and attitude tracking
control for a quadrotor uav based on terminal sliding mode control,” in 2015 34th Chinese
Control Conference (CCC), pp. 3398–3404, Sep. 2015.
[35] K. Wu, H. Wang, M. A. Esfahani, and S. Yuan, “Learn to navigate autonomously through
deep reinforcement learning,” IEEE Transactions on Industrial Electronics, vol. 69, no. 5,
pp. 5342–5352, May. 2021.
93
[36] L. Chen, H. Zhang, J. Xiao, L. Nie, J. Shao, W. Liu, and T.-S. Chua, “Sca-cnn: Spatial and
channel-wise attention in convolutional networks for image captioning,” in Proceedings
of the IEEE conference on computer vision and pattern recognition, pp. 5659–5667, July.
2017.
[37] O. K. Oyedotun, K. A. Ismaeil, and D. Aouada, “Why is everyone training very deep neural
network with skip connections?” IEEE Transactions on Neural Networks and Learning
Systems, pp. 1–15, Jan. 2022.
[38] A. Howard, M. Sandler, G. Chu, L.-C. Chen, B. Chen, M. Tan, W. Wang, Y. Zhu, R. Pang,
V. Vasudevan, Q. V. Le, and H. Adam, “Searching for mobilenetv3,” in Proceedings of the
IEEE/CVF international conference on computer vision, pp. 1314–1324, Oct. 2019.
[39] J. Hu, L. Shen, and G. Sun, “Squeeze-and-excitation networks,” in Proceedings of the
IEEE conference on computer vision and pattern recognition, pp. 7132–7141, Jun. 2018.
[40] A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly, J. Uszkoreit, and N. Houlsby, “An image
is worth 16x16 words: Transformers for image recognition at scale,” arXiv preprint
arXiv:2010.11929, Oct. 2020.
[41] J. Raj, K. S. Raghuwaiya, and J. Vanualailai, “Novel lyapunov-based autonomous controllers for quadrotors,” IEEE Access, vol. 8, pp. 47 393–47 406, Mar. 2020.
[42] S. Shah, D. Dey, C. Lovett, and A. Kapoor, “Airsim: High-fidelity visual and physical
simulation for autonomous vehicles,” in Field and service robotics, pp. 621–635, Nov.
2018.
[43] K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image
recognition,” arXiv preprint arXiv:1409.1556, Sep. 2014.
[44] C. Szegedy, V. Vanhoucke, S. Ioffe, J. Shlens, and Z. Wojna, “Rethinking the inception
architecture for computer vision,” in Proceedings of the IEEE Conference on Computer
Vision and Pattern Recognition, pp. 2818–2826, Jun. 2016.