MEAM Seminar: “Physical and Neural Reservoirs for Mechanically Intelligent Sensing and Control”

Reservoir computing is an information processing paradigm within which the nonlinear dynamics of a system are exploited to perform information processing. While originally proposed in the context of artificial and biological neural network dynamics, computing via dynamics has been shown to be applicable to a wide range of systems. Of note are physical reservoir computers, where a structure’s physical or mechanical nonlinear dynamics are used to perform ‘mechanically intelligent’ information processing. This talk will describe recent work on identifying how reservoir computing can be used to enable sensing and control in compliant systems such as soft robots, where traditional approaches struggle to accommodate the high-dimensionality and nonlinearity of soft, compliant structures. Drawing inspiration from biological systems found in nature such as spider webs and octopus arms, where a reservoir computing paradigm (both physical and neural) may help explain their unique ability to coordinate their soft arms, I will show how both physical and neural dynamics can be leveraged to simplify the need for power-hungry onboard information processing in soft robots and other compliant structures.