Autonomous drone hunter operating by deep learning and all-onboard computations in GPS-denied environments

Area of Science:

• Robotics and Autonomous Systems
• Artificial Intelligence
• Aerospace Engineering

Background:

• The increasing prevalence of small unmanned aerial vehicles (UAVs) necessitates effective counter-UAV (C-UAV) strategies.
• Operation in Global Positioning System (GPS)-denied environments presents significant challenges for autonomous navigation and target engagement.

Purpose of the Study:

• To develop and validate an autonomous UAV platform capable of detecting, tracking, and engaging other small UAVs in GPS-denied conditions.
• To investigate the efficacy of machine learning-based detection algorithms and visual-servoing for close-range drone interception.

Main Methods:

• A dataset of 58,647 images was collected and utilized to train a Tiny YOLO object detection algorithm.
• A physical UAV platform was equipped with the trained detection model and a visual-servoing system for autonomous operation.
• The system was tested for its ability to detect, track, and follow a target drone in real-world conditions.

Main Results:

• The autonomous UAV platform successfully tracked a target drone at an estimated speed of 1.5 m/s.
• The Tiny YOLO detection algorithm achieved 77% accuracy in cluttered environments.
• System performance was constrained by detection accuracy, camera frame rate (8 frames per second), and field of view.

Conclusions:

• The proposed UAV platform demonstrates feasibility for autonomous C-UAV operations in GPS-denied environments.
• Machine learning and visual-servoing are viable components for developing autonomous drone interception systems.
• Further improvements in detection algorithms and sensor capabilities are required to enhance performance in complex scenarios.