An early adopter of robotics in the 20th Century was the manufacturing industry, which realized that highly precise and repetitive motions could be better performed by robots executing scripts. But in the field, work is still performed in much the same way as it has for millennia. Construction projects are still performed by teams of dozens to thousands of specialized workers, who work in parallel, despite countless obstacles forcing adjustments to be made to correct for e.g. manufacturing errors and joining errors. There is no single human who knows the full state of the system, and all decisions are made amidst countless sources of uncertainty. Robots in these settings perform highly specialized tasks, including site inspection, routine repetitive placement (e.g. tile laying robots), and landscape grading. In hazardous environments such as space, construction can only proceed by exposing humans to unnecessary risks or granting robots a greater degree of autonomy than is currently possible.
What research must be done to enable robots to construct or manufacture with complete autonomy?
In 2018, I founded the Field and Space Experimental Robotics Laboratory (FASER Lab), dedicated to pursuing answers to aspects of the construction question. My time at NASA informed my research approach, which has transferred well to my role at Virginia Tech. There and here, large scale prototypes and test articles designed in-house are favored for their closer resemblance to eventual flight hardware. In FASER Lab, wherever possible, we design, build, and verify all robots and test-articles for their various projects. This infrastructure includes a 4-m tall, 8-m reach robotic arm based on NASA’s Lightweight Surface Manipulation System, a Lunar lander robotic crane prototype.
Cross-disciplinary research is required to study the numerous aspects of the construction question.
Each novel research result from FASER Lab’s research addresses at least one of these aspects.
Operations Sequencing
Dr. Joshua Moser investigated operations planning for construction scenarios, and invented a novel mixed integer programming operations planning method in collaboration with a team from VT’s Industrial and Systems Engineering which is the first and only known method that optimally plans tasks for a team of robots that 1) must be able to work in small teams on specific tasks and 2) must re-plan in case of task failure.
A completed arch made collaboratively, in which a forced error led to correct re-planning.
Accurate Truss Assembly and Deployment
Dr. Samantha Glassner Chapin worked with Dr. Everson and Dr. Quartaro to study, design, build, and verify a mixed deployable and assembled radio telescope truss structure, which we called BORG after a suggestion from our NASA colleagues. While doing this, Dr. Chapin investigated incorporating structural mechanism states and prediction methods into Bayesian factor graph methods. We named the method SF-GraphSLAM (SF: Semantic Fiducial). We believe it to be the only published method that incorporates constructed mechanical element semantic knowledge into a workspace estimation factor graph method.
Scaffolding for Structure Deflection and Repair
Dr. Holly Everson also committed years of effort into creating the BORG prototypes, and while doing so, developed a methodology to use an intelligent scaffolding robot to deflect a damaged structure safely to enable repairs, elaborating on work she started with me as an undergraduate, which is crucial for safely handling and correcting errors on structures under load that must be scaffolded to be repaired.
Structure Outfitting with Deformable Objects
Dr. Amy Quartaro developed a mixed flexible-rigid body modeling approach for construction robots to rapidly model and predict the state of deformable linear objects in a workspace, imaged only by a monocular camera. Her work is the first to explicitly enable robots to model cable behavior wherein the cables have kinks. This work is necessary for robots to outfit a completed structure with utilities such as electricity and ventilation.
Visual detection and measurement of a cable to determine its rest properties.
Rock Classification and Measurement
Yiyan Ruan is investigating scene recognition via limited quantities of RGB-D sensor data to enable robots to recognize and measure objects of indeterminate shape (e.g. rocks in a landscape that is to be graded) and has utilized current deep learning techniques to out-perform state-of-the-art methods. His research is currently under review at two journals (see the Publications link).
Reconfigurable robotic truss structures
William Chapin collaborated with a team from Virginia Tech’s Industrial and Systems Engineering department to model and plan optimal reconfiguration motions for serialized Stewart platforms, which can be used both as precise manipulators and as reconfigurable trusses.