Physical Parameter Dependencies in Mechanical Reservoir Computing: Structural Analysis, Actuation, and Improved Processing

While physical reservoir computers (PRCs) have tremendous potential for applications in electromechanical and biomechanical systems, their adoption remains slow due to a limited understanding of the impact of mechanical parameters on computational properties. Here, we specifically investigate these relationships concerning the body structure, input parameters, and observed states using classical swinging body dynamical systems: soft tentacles and a simulated multilink pendulum. Principally, we show that the mechanical structure (including stiffness and damping) couples with input parameters (frequency and magnitude) to regulate the manifested computational capabilities, and the type of PRC (if any) that can be achieved. Further, we show that input rate drives a transition between linear memory-based and nonlinear calculations, while the mechanical structure alters the conversion between longer and shorter memories. By investigating sensory limitations, we show that a spatiotemporal structure of the computational properties exists that relates to the mechanical structure, and that can be exploited to improve selected computing tasks in soft tentacle PRCs. Using the parameter dependencies we additionally provide inferences into the computational archetypes of the bodies of established soft robots.