BIM-based indoor robot initialization in construction automation using object detection

• Main Author: Zhao, Xinge
• Other Authors: Cheah Chien Chern
• Format: Thesis-Master by Research
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/163030
• Institution: Nanyang Technological University

In recent years, there has been increasing interest for construction automation solutions to revolutionize the conventional construction industry and among them, robotics solutions are commonly used to automate construction tasks and to operate under dangerous conditions. To ensure the satisfactory performance of robotics solutions at construction sites, especially for applications using mobile robots, awareness of the location is required to initialize the robot before the navigation can be commenced. However, in indoor construction sites, GPS is not accessible, and infrastructure-based wireless networks such as WiFi, Bluetooth are not available yet, making automated initialization of mobile robots difficult. The traditional marker-based methods require manual deployment and calibration, and markers could be blocked by construction materials and components. Therefore, infrastructure-free and robust robot initialization methods are required for the complex construction environments. This research aims to develop an integrated Building Information Model (BIM)-based indoor robot initialization system using an object detector to initiate the robot location in an environment map built from BIM, given the robot deployed at an arbitrary location. CNN-based object detection techniques are used to recognize and locate the visual features, which are common and widely distributed building components at construction sites. A feature matching algorithm is proposed to correlate the acquired online information of detected features with geometric and semantic information retrieved from BIM, and the robot location in the BIM coordinate frame is estimated based on the feature association. Moreover, the proposed robot initialization system provides the functions of robot exploration and BIM-based local navigation that interacts with the visual recognition system to supervise the robot motion for active localization, making the entire system fully automatic. This system could ease the problem of devices or markers deployment and time-consuming environment configuration for robot initialization at construction sites. It can be integrated into the robot navigation system towards a completely automatic mobile robot navigation stack in construction automation. The proposed system was validated through experiments at various environments including real-world construction sites and the robustness and efficiency are illustrated.