Advanced Sensor Integration for Enhanced Flight Control in UAV-Based Construction Automation

Abstract

In construction automation, several tasks such as surface finishing, material transport, and inspections typically rely on labor-intensive manual methods, which are inefficient, susceptible to human error, and expose workers to hazardous conditions. Drones have emerged as a solution to automate these processes, but maintaining precise navigation and stable attitude control in dynamic construction environments remains challenging. This paper presents a sensor fusion control framework to enhance UAV performance in dynamic and complex construction environments. The framework integrates multi-dimensional data from diverse sensing modalities, including geometric, visual, and inertial information, to achieve robust situational awareness and adaptive control. A Simultaneous Localization and Mapping (SLAM) model fuses multi-modal sensor inputs, including LiDAR point clouds, camera visual features, and inertial measurements, to generate precise environmental maps and perform accurate localization. The SLAM framework employs advanced loop closure techniques to minimize trajectory drift and optimize navigation paths in real-time. Additionally, flight stability under external disturbances is managed by a Deep Reinforcement Learning (DRL) algorithm, which models the UAV's control system as a Markov Decision Process (MDP). The DRL agent dynamically adjusts the UAV's flight parameters, such as thrust, roll, pitch, and yaw, based on real-time feedback from fused sensor data. The control strategy leverages a stability-based reward function to minimize deviations in position and orientation, ensuring rapid recovery from external disturbances. Simulation tests conducted in a Robotic Operation System and Gazebo simulation environment demonstrated mapping accuracy of 95.4%, localization errors under 2 cm, and rapid recovery from disturbances within 0.8 seconds.