Reinforcement learning UAV navigation system based on improved ACO in 3D rescue scene

• Published in: IEEE International Symposium on Broadband Multimedia Systems and Broadcasting pp. 1 - 6
• Main Authors: Li, He, Dong, Chuang, Sun, Shixian, Ma, Xiaopu, Yu, Peng, Qi, Qinglei
• Format: Conference Proceeding
• Language: English
• Published: IEEE 11.06.2025
• Subjects: Autonomous aerial vehicles; Benchmark testing; Broadband communication; Broadcasting; DQN; improved ant colony algorithm; Multimedia communication; Navigation; Reinforcement learning; RSS; Three-dimensional displays; three-dimensional mountain rescue; UAV
• ISSN: 2155-5052
• DOI: 10.1109/BMSB65076.2025.11165581

In the three-dimensional mountain rescue scene, unmanned aerial vehicle (UAV) is the main search and rescue tool. How to quickly locate the distress source and plan the subsequent material distribution route has become the key to solve such problems. Received signal strength (RSS) is used as the benchmark for path finding, and the navigation process is divided into two stages, inside and outside the signal field, by setting a threshold. Outside the signal field, an improved ant colony algorithm is designed. The local and global pheromones are set, and the algorithm parameters are adaptively adjusted according to the progress to accelerate the UAV into the signal field. After entering the signal field, the DQN dominates the navigation of the UAV, and the reward mechanism is formulated according to the signal strength change and the mapping pheromone distribution to improve the rate of the UAV reaching the distress source. At the same time, the experience exchange between the UAV and the environment was realized to solve the optimal rescue path.