Research on Autonomous Obstacle Avoidance for Quadrotor UAVs Based on Deep Learning

DOI: https://doi.org/10.62517/jike.202604221

Author(s)

Qiyan Song

Affiliation(s)

Engineering&Technical College, Chengdu University of Technology, Hebei, China

Abstract

To address single‑sensor limitations and real‑time constraints in quadrotor UAV obstacle avoidance, this paper proposes a closed‑loop framework integrating cross‑modal RGB‑D/radar fusion, PPO‑based dual‑stream decision‑making with online incremental learning, and differential‑flatness trajectory generation. Under ideal conditions, detection accuracy reaches 95.4%, avoidance success rate 91.8%, and end‑to‑end latency 109 ms; under dark and fully occluded conditions, accuracy is 89.2% and avoidance success rate drops to 82.0%, indicating a performance boundary. These results validate the framework's potential in simulated unstructured dynamic environments while revealing areas for improvement under extreme perception degradation.

Keywords

UAV Autonomous Obstacle Avoidance; Deep Learning; Multi-Sensor Fusion; Trajectory Planning

References

[1] B. Fan and G. Zeng, “Design of autonomous obstacle avoidance scheme for quadrotor UAV in cruise flight mode,” Electronic Design Engineering, no. 2, pp. 180–184, 2020. [2] P. Gao, “Binocular vision obstacle avoidance algorithm for multi-rotor UAV based on deep learning,” D. thesis, National University of Defense Technology, 2017. [3] P. Shang, “Object detection and obstacle avoidance for quadrotor UAV,” D. thesis, Xi’an Shiyou University, 2025. [4] Y. Yan, “Research on trajectory tracking control algorithm for quadrotor UAV,” D. thesis, Shenyang Ligong University, 2024. [5] Y. Liang, “Research on visual obstacle avoidance technology for quadrotor UAV based on convolutional neural network,” D. thesis, Nanjing University of Aeronautics and Astronautics, 2023. [6] X. Ma, “Trajectory planning for wheeled mobile robot on mobile platform based on differential flatness,” D. thesis, Nanjing University of Information Science and Technology, 2018. [7] X. Wang, K. Xia, and D. Song, “Multi-source heterogeneous sensor fusion mapping method for UAV,” Robot, 2024. [8] J. Hu and Q. Chen, “Path planning optimization method for unmanned vehicle based on dual-stream spatio-temporal fusion network,” Intelligent IoT Technology, no. 4, pp. 73–77, 2025. [9] Z. Li, L. Li, and X. Tao, “Dual-stream graph convolutional network for dynamic traffic flow prediction,” Journal of Computer Science and Exploration, no. 2, pp. 384–394, 2022. [10] H. Wang, J. Huang, W. Wang, K. Zeng, N. Wang, and H. Hong, “Multi-UAV formation obstacle avoidance control method based on PPO algorithm,” Ordnance Industry Automation, no. 2, pp. 108–112, 2026. [11] W. Zhang, W. Zhang, F. Song, and L. Long, “Monocular vision obstacle avoidance method for quadrotor UAV based on deep learning,” Journal of Computer Applications, vol. 39, no. 4, pp. 1001–1005, 2019. [12] C. Zhu, “UAV obstacle avoidance algorithm based on deep reinforcement learning,” D. thesis, Dalian University of Technology, 2025. [13] J. Li, “Research on object detection and autonomous obstacle avoidance for multi-rotor UAV,” D. thesis, Changchun Institute of Optics, Fine Mechanics and Physics, Chinese Academy of Sciences, 2020.