Teaching multifunctionality to nonlinear fluidic networks

Soft robots powered by pressurized fluid are enabling a variety of innovative applications in diverse areas, from biomimetics to rehabilitation. Such soft machines need suitable controllers. Currently, there is no general design strategy for building non-electronic control modules that require few inputs yet enable multiple functionalities.

Here, we demonstrate a fluidic network that acts as just such a multifunctional control device. Crucially, the network structure is designed via a simple computational algorithm, based on recently explored contrastive learning in complex networks. This work presents a practical way to design soft robotic controllers with minimal prior knowledge.