The sense-assess-respond feedback loop is a critical aspect of intelligence in both living and material systems. Physical reservoir computing is a strategy toward embodying intelligence into a system by leveraging nonlinear dynamics to perform nonlinear computations. While mechanical reservoirs typically produce nonlinear dynamics due to some material nonlinearity, geometric nonlinearity can also produce nonlinear dynamics due to the finite displacements and rotations in the network. This study examines the effect of both network node degree and geometry on the nonlinear computing performance of simulated mass-spring-damper systems comprised of linear springs. Smaller node degrees and sharper equilibrium angles in the network tended to produce greater nonlinear frequency content. However, the equilibrium geometry is not sufficient to characterize the reservoir performance as the dynamic variation in the spring angles also significantly correlates with nonlinear frequency content. This study highlights the significance of both network topology and dynamic geometry on the performance of mechanical reservoir computers.
Kiyabu et al. (Mon,) studied this question.