The initial research for this production method was done in collaboration with RoboFold to develop a proof of concept model for the company’s production method. The proof of concept was undertaken at a scale of 1:5 using off-the-shelf, 6-axis robotic arms that were retrofitted with custom vacuum grippers. The production model looks to eliminate the need for tooling by forming sheet metal with a system of curved-crease scoring and folding by the robots. The overall goal for the prototype was to create a system which allowed multiple robots to cooperate to fold an individual piece. The other challenge is that robotic arms are generally used for pick and place type operations rather than following a highly controlled path needed for folding operations. The goal was to create digital versions of the physical robots which accounted for the variations in sizes and range of motion per axis and create a visual solver to determine the kinematic values necessary to fold a particular shape. Robofold employed this system to prototype a production system for Studio Joris Laarman’s Asimov chair which was to be made from a single sheet of folded metal. To do this, the form of the chair was developed in paper and then digitally scanned. The digital mesh of the finished chair was then simulated to fold to a flat sheet by EVOLUTE in Vienna. The folding process is then inverted digitally to create a singular mesh which animates from flat to the final form. The robots are attached to the digital surface at a specified point and as the surface animates it drives the kinematic chain of the robot. The distinct angle of each axis was then run through a calibration filter to determine the value sent to the servo. A background script would then take the calibrated pulse values, convert them to a specified format, and write them into a text file each frame. At this stage the folding simulation did not take into account specific material properties or attachment points so the surface served as a general guide to the physical forming and deviations due to material properties were implemented via an offset layer in the software. According to Greg Epps of Robofold, the placement of the grippers can be determined by understanding the folding mesh as a quad mesh linkage which is broken into interrelated regions based on the major fold lines. Regions which require a higher degree of accuracy, such as connection points, are given priority as they require more precise guiding, but as the linkage becomes more complex, fewer robots are required in relation to the number of folds. In this case, the four robots were used in pairs to create the 2 crease details on the back of the chair which lock the form into place and the fixed arm is used to fold the bottom fin and hold the material so that the 4 bots have the leverage to push the material.
At each simulated frame the angle of each axis per robot, the fixed arm, and 4 solenoids had to be calculated resulting in 29 values being sent to the hardware over a custom protocol. The solenoids where used to allow the pickup and placement of the finished chair by controlling the suction of the vacuum grippers. To accommodate this, a custom circuit and controller were developed based on an Arduino. By embedding the digital model with the physical characteristics this process could be crafted digitally frame-by-frame as an iterative process between the physical and digital model. This direct control of the overall system made it possible to deal with the vast amount of data that the physical robots required to operate.