The proposed system integrates a lightweight MCU based object detection model within a custom built UAV platform and a 5G-enabled multi agent coordination infrastructure.This architecture facilitates real time obstacle detection and dynamic path planning in construction environments.The system comprises:
Custom UAV Platform: A hexacopter equipped with a monocular camera and an onboard microcontroller for real time image processing.
Multi-Agent Coordination: Utilizing a central coordinator connected via 5G, the system manages task distribution and vehicle status monitoring.
Ground Control Station (GCS): A modular software architecture using React/Redux for the front end and ROS based back end, supporting task planning and path optimization.
Object Detection and Path Planning
The UAV employs a lightweight object detection model optimized for microcontroller units (MCUs), enabling real time processing of camera frames. For path planning, the system utilizes Voronoi graph based A* algorithms, ensuring efficient navigation and obstacle avoidance in dynamic construction sites.
Dataset and Model Optimization
To support MCU based edge applications, the researchers developed the TUBITAK EdgeDrone dataset, comprising approximately 25,000 training and 5,000 testing images.Model optimization techniques, such as quantization, were applied to compress the object detection model, making it suitable for deployment on resource constrained devices.
Hardware and Software Integration
UAV Hardware: The hexacopter integrates a Pixhawk based flight control unit, six BLDC motors, a carbon fiber frame, an onboard computer, an STM32F769i Discovery board, a 5G router, and a separate power supply for sensors.
Software Stack: The system utilizes ROS Noetic with MAVROS for flight control and MQTT for communication. A vision node processes camera frames for object detection, while the central coordinator manages task distribution and path planning.
Results and Implications
Field experiments demonstrated the practical viability of the system, highlighting its scalability and computational efficiency advantages over existing UAV solutions.The integration of Edge AI enables real time decision making, reducing latency and enhancing the responsiveness of autonomous operations in construction environments.