High-accuracy mobile manipulation for on-site robotic building construction

Despite the wide-spread usage of robots in manufacturing, robots and automation are yet to make a significant impact on the building construction industry. This is because building construction requires a combination of mobility, accuracy, and autonomy that mobile robotic systems cannot yet provide. This work develops localization, sensor fusion and control strategies to allow mobile manipulation systems to achieve high-accuracy end-effector positioning throughout large workspaces and over long time-scales. We demonstrate that these methods allow such machines to perform the final machining and assembly of materials directly on the construction site and enable to execution of novel digital building processes. We propose the concept of a mobile manipulator, called an In situ Fabricator (IF), purpose-built for performing a wide range of building tasks on the construction site. Our initial prototype, IF1, considers the careful integration of existing robotic hardware and methods into a generic construction machine. We demonstrate the construction of a load-bearing steel-reinforced concrete wall with IF1 on a real construction site. High-accuracy mobile manipulation requires accurate, reliable and globally consistent localization. We investigate how known geometric information about the environment and ongoing building process can be used to improve the localization accuracy of state-of-the-art methods. An object-based simultaneous localization and mapping system is presented which tracks robot motion relative to the structure it is building with sub-centimeter accuracy over the course of long building tasks. This system is used to guide the manual construction of a brick structure through an augmented reality interface, showing how sensed information about the built structure can used to continuously adapt the ongoing building process to compensate for inaccuracies as they appear. In order to ensure the accuracy and robustness of a mobile manipulator’s state estimator throughout large workspaces and over long time-scales, it is desirable to fuse the measurements from multiple heterogeneous sensors online. We present a general and modular framework for online probabilistic sensor fusion, making our C++ implementation available open-source. It is shown that our moving horizon estimator-based framework offers more flexibility in state estimator design and can generate higher quality estimates than conventional filter-based methods. We additionally demonstrate wholebody state estimation for a smaller but more dynamically capable mobile manipulator, IFmini, using visual and inertial sensors on both the base and end-effector. Modern mobile manipulation systems achieve high accuracy by adding process controls to prevent base motion while executing manipulation tasks. This strategy tends to make such systems prohibitively heavy and slow. We address this problem by creating a sensory-motor system which can effectively sense and reject dynamic disturbances coming from to the base of the robot such that high-accuracy end-effector positioning can be achieved without assuming that the base is stationary. It is shown that by fusing localization information with inertial measurements inside of our moving horizon estimator, high-consistency, high-rate and real-time predictive estimates can be generated and used to directly close a task space control loop at the toque level. We demonstrate the ability of the system to reject dynamic disturbances through experiments where the base of a hydraulic manipulator is pushed by hand while the system works to maintain a stationary end-effector.