Sculpting Folding Pathways through Physical Learning in Mechanical Systems

Creased sheets can possess an exponential number of folding pathways accessible from the flat state. This property poses an engineering challenge in the design of specialized self-folding origami patterns with one or a few prescribed folding pathways. Hence, an alternative means to eliminating undesired folding pathways via physical training is sought. Creased sheets are folded along desired folding pathways and their initially equal crease stiffnesses are allowed to dynamically evolve according to a local rule dependent on folding strain. It is found that undesired folding pathways are eliminated in saddle-node bifurcations, leaving behind only the desired folded pathway. Thus, physical folding, combined with a plasticity rule for crease stiffness, can naturally arrive at design parameters needed for non-linear behaviors that are hard to predict otherwise. This work also explores the capability of mechanical networks to accomplish more complex tasks, like classification of forces. It demonstrates supervised learning in self-folding origami patterns. Self folding origamis are able to learn from examples through a simple dynamic physical learning rule. These self-folding origamis were able to identify the different classes of the iris flowers based on their features, such as the length and width of the petals and sepals of the flower.