Optomechanical Reservoir Computing for Embodied Signal Processing

The sense-assess-respond feedback loop is a key building block for intelligent behavior in living and synthetic materials systems. Soft robotics is an ideal testbed for the development and embodiment of sense-assess-respond feedback networks, due to the compatible integration of novel sensing motifs, diverse multi-functional responses, and the prevalence of many interesting material and geometric nonlinearities to exploit for assessing (or computing) their local environment. In addition, soft robotic systems are typically power and size constrained, further motivating the development of alternative strategies for their environmental processing and control. In this study, we investigate the information processing capacity of a coupled optomechanical spring system with tunable nonlinearities using the computing framework of physical reservoir computing. Reservoir computing is a class of recurrent neural networks that trains only a readout layer of the network dynamics in contrast to tuning all the internal parameters of the network. This simplified training regime opens the door to physical implementations (vs on chip) of this neural network mapping behavior, and harnesses the unique, intrinsic nonlinearities and variabilities of a physical system. We introduce a mechanical analog for the rectified linear unit (ReLU) activation function, and tune the stiffness ratio of the bilinear force-displacement curve through geometric design. Elastomeric light guides are mechanically attached to a 1D experimental ReLU spring network and demonstrate a nonlinear optical transmission vs displacement response when mechanically driven. The nonlinearity in optical intensity is a result of leakage when the lightguide deforms in bending, which follows Snell’s law of refraction. We further develop a spectral projection method to characterize the relationship between system nonlinearity and reservoir computing performance. The analysis partitions the dimensionality increase of the signal among its frequencies, distinguishing between the strengths of the linear vs various nonlinear classes of frequency content. Collectively, the simulation and experimental results of this work demonstrate the benefit of combining different physical nonlinearities for signal processing, as highlighted in the optomechanical reservoir outperforming the mechanical only system. The study also motivates the identification of coupled physical nonlinearities in more complex materials systems, such as soft robotics, to directly perform embodied signal processing and advance their sense-assess-response feedback toward more intelligent behaviors.