Designing for Digital Assembly with a Construction Team of Mobile Robots
Advances in construction automation have primarily focused on creating heavy machines to accomplish repetitive tasks. While this approach is valuable in an assembly-line context, it does not always translate well for the diverse terrain and dynamic nature of construction sites. As a result, the use of automation in the architectural assembly has lagged far behind other industries. To address the challenges of construction-site assembly, this project suggests an alternative technique that uses a fl eet of smaller robots working in parallel. The proposed method, which is inspired by the construction techniques of insect colonies, has several advantages over the use of larger machines. It allows for much greater on-site fl exibility and portability.
It is also easy to scale the operation, by adding or removing additional units as needed. The use of multiple small robots provides operational redundancy that can adapt to the loss of any particular machine. These advantages make the technology particularly suitable for construction in hazardous or inaccessible areas. The use of assembly robots also opens new horizons for design creativity, allowing architects to explore new ideas that would be unwieldy and expensive to construct using traditional techniques. In our tests, we used a team of small mobile robots to fold 2D laser-cut stock into 3D curved structures, and then assemble these units into larger interlocked forms.
For preliminary testing, we used a team of five Cozmo robots. The modular construction units, or “bricks,” were generated from planar stock (heavy-weight cardstock) that was laser-cut to allow pop-up construction into curved three-dimensional forms. The laser cutter was used to inscribe the cardboard along desired folding creases so that applying planar forces at specified points along the perimeter caused the cardboard to spring into a 3D shape. Our system design was motivated by the goal of using relatively simple, independent robots with limited capabilities, so we endeavored to keep the process as streamlined as possible.
The sensory capacities of the robots were limited to perceiving only objects labeled with fiducials. They were also able to query nearby robots to obtain the robot’s ID and to verify that they were working together on the same brick (these processes were also carried out using fiducial numbers). A large fiducial tag at the supply depot was used to give the robots a common coordinate frame and help them navigate from the supply depot to the building site. However, we noted during our tests that the accuracy of the robot’s localization decreased with distance from the depot, particularly when the robots were moving and/or temporarily lost sight of the depot. Future work could potentially use the robots themselves to maintain a more reliable coordinate frame. Even without global position information, it is theoretically possible for the robot team to maintain reliable paths from the supply depot to the construction area by using some robots as stationary landmarks.
One of the unique aspects of this approach is that the individual units did, not store information about the current state of the overall structure or the actions of more distant robots. Furthermore, the robots obtained information about the available construction options only through direct inspection. After leaving the build area, the memory of the structure’s state was not retained since this information was likely to become outdated as other robots continued to make modifications. Our trial demonstrations showed that the Cozmo robots could be readily programmed to build basic architectural structures in this fashion. One of the structures that we constructed using this approach is shown in Figures. As noted above, only the first two steps of the process are currently fully automated. Future work will be carried out to automate the process of stitching the chains into a mesh and lifting them into place. These movements will require coordinated action from many robots at once.
Credit to: Design & Augmented Intelligence Lab
Research Team: Dr. Kalantari (PI), Dr. Becker, Rhema Ike, Han Dang
Year: Since 2017