Local learning in robotic granular materials

The mechanical properties of granular packings are closely related to particle configurations as well as the interparticle interactions present in the system. Numerical studies on small systems have shown that even a single particle rearrangement can cause a discontinuous change in the shear modulus of a granular material [1]. However, the specifics of how particle-level changes affect the bulk properties of a granular packing remain poorly understood. To study this behavior, we have developed a 2D robotic granular system that allows us to vary the size and pressure of individual soft, deformable particles. This setup enables us to continuously transition between different packing configurations while tracking the system's modulus. Our goal is to uncover local rules based on particle size, deformation, and pressure that govern the bulk properties of the system and use them to predictably modify a packing's behavior. These findings will provide a pathway to create trainable granular metamaterials that can be evolved using local feedback.



[1] Wang, Philip, et al. "Shear response of granular packings compressed above jamming onset." Physical Review E 103.2 (2021): 022902.