Autonomous Scene Understanding, Motion Planning, and Task Execution for Geometrically Adaptive Robotized Construction Work

The construction industry suffers from such problems as high cost, poor quality, prolonged duration, and substandard safety. Robots have the potential to help alleviate such problems by becoming construction co-workers, yet they are seldom found operating on today’s construction sites. This is primarily due to the industry’s unstructured nature, substantial scale, and loose tolerances, which present additional challenges for robot operation. To help construction robots overcome such challenges and begin functioning as useful partners in human-robot construction teams, this research focuses on advancing two fundamental capabilities: enabling a robot to determine where it is located as it moves about a construction site, and enabling it to determine the actual pose and geometry of its workpieces so it can adapt its work plan and perform work. Specifically, this research first explores the use of a camera-marker sensor system for construction robot localization. To provide a mobile construction robot with the ability to estimate its own pose, a camera-marker sensor system was developed that is affordable, reconfigurable, and functional in GNSS-denied locations, such as urban areas and indoors. Excavation was used as a case study construction activity, where bucket tooth pose served as the key point of interest. The sensor system underwent several iterations of design and testing, and was found capable of estimating bucket tooth position with centimeter-level accuracy. This research also explores a framework to enable a construction robot to leverage its sensors and Building Information Model (BIM) to perceive and autonomously model the actual pose and geometry of its workpieces. Autonomous motion planning and execution methods were also developed and incorporated into the adaptive framework to enable a robot to adapt its work plan to the circumstances it encounters and perform work. The adaptive framework was implemented on a real robot and evaluated using joint filling as a case study construction task. The robot was found capable of identifying the true pose and geometry of a construction joint with an accuracy of 0.11 millimeters and 1.1 degrees. The robot also demonstrated the ability to autonomously adapt its work plan and successfully fill the joint. In all, this research is expected to serve as a basis for enabling robots to function more effectively in challenging construction environments. In particular, this work focuses on enabling robots to operate with greater functionality and versatility using methods that are generalizable to a range of construction activities. This research establishes the foundational blocks needed for humans and robots to leverage their respective strengths and function together as effective construction partners, which will lead to ubiquitous collaborative human-robot teams operating on actual construction sites, and ultimately bring the industry closer to realizing the extensive benefits of robotics.PHDCivil EngineeringUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttps://deepblue.lib.umich.edu/bitstream/2027.42/149785/1/klundeen_1.pd