Then, a so-called Singmaster String is sent to a C / C++ program running on a high-performance computer, which is then able to optimally solve the cube—i.e. in 21 or fewer 90° rotations. The first time the program starts up, approximately 1.5 gigabytes of solutions are generated; they’re arrayed in a tree (brachiating) structure with, of course, the completed cube at the end. These are, needless to say, nowhere near all possible combinations. If none of these combinations matches the cube’s current combination, then it must at least include symmetries since all “non-symmetries” are depicted in the solutions tree. In this case, the cube is rotated until its composition corresponds to one of the anti-symmetrical groups in the solutions tree. Thereafter, it’s simply a matter of following the path to the solved cube. Solution duration: approximately 1 second to 5 minutes.

The result of the program—or, to put it more precisely, the rotation commands—are sent via a Python Script to the robot’s three Arduino Unos, which interpret these in the form of commands for the stepper motors and finally carry them out.

By means of various deep-learning frameworks, thousands of simulations and/or training sets for reinforcement learning are run on the in-house servers at Spengergasse Technical School and the Technical University of Vienna and subsequently tested on the Hexapod. However, to get a clearer picture of its physical surroundings, the Hexapod has recourse to a stereo camera, which puts the information about these surroundings to use. Thanks to object recognition capabilities, the robot can navigate independently in it field of deployment and execute tasks.