Dynamic obstacle avoidance for UAVs using fuzzy logic control and neural network

Unmanned Aerial Vehicles (UAVs) operating in dynamic environments require advanced navigation systems capable of avoiding moving obstacles in real time. In this paper, we introduce a fuzzy logic-based control algorithm designed to enhance UAV autonomy. Although many researches utilize data from different sensors, in this work we tried to operate only with on-board camera. Every frame is handled by a neural network for Object Detection. In this case the obstacle subject is another drone. For this purpose YOLO 10 was additionally trained and used. Detected obstacle is analysed by many parameters: distance, horizontal centre, vertical centre, vertical approaching, horizontal approaching. Following the computation, the derived parameters are input into the Mamdani fuzzy inference algorithm. As it interprets linguistic rules, we define fuzzy sets and apply them to generate an output based on degrees of membership. For the output, horizontal move and vertical move are chosen. The system was tested in a virtual environment. It utilizes the Unity framework, in which a simple scene was created. As an obstacle the model of the drone was used. A camera acts as the unmanned aerial vehicle (UAV). Frames are sent to a python script on a local machine. It helps to simulate the structure of a system on edge device. And output parameters are delivered back to correct the path, if the controlled UAV has to avoid collision.